From c4821c8e6bf1f81c6ea31e11ebc0fc1666e9337b Mon Sep 17 00:00:00 2001
From: David Robillard <d@drobilla.net>
Date: Thu, 25 Feb 2021 10:27:59 -0500
Subject: Add exess from git@gitlab.com:drobilla/exess.git 4638b1f

---
 meson.build                                        |    6 +-
 subprojects/exess/.clang-format                    |   30 +
 subprojects/exess/.clang-tidy                      |    7 +
 subprojects/exess/.clant.json                      |    4 +
 subprojects/exess/.gitignore                       |    1 +
 subprojects/exess/.gitlab-ci.yml                   |  168 ++
 subprojects/exess/.includes.imp                    |   10 +
 subprojects/exess/NEWS                             |    5 +
 subprojects/exess/README.md                        |   83 +
 .../exess/bindings/cpp/include/exess/exess.hpp     |  707 +++++++++
 subprojects/exess/bindings/cpp/meson.build         |   54 +
 subprojects/exess/bindings/cpp/test/.clang-tidy    |   17 +
 subprojects/exess/bindings/cpp/test/meson.build    |    8 +
 .../exess/bindings/cpp/test/test_exess_hpp.cpp     |  221 +++
 subprojects/exess/doc/c/Doxyfile.in                |   39 +
 subprojects/exess/doc/c/api/meson.build            |    5 +
 subprojects/exess/doc/c/index.rst                  |   10 +
 subprojects/exess/doc/c/man/manlink.in             |    1 +
 subprojects/exess/doc/c/man/meson.build            |  122 ++
 subprojects/exess/doc/c/meson.build                |   53 +
 subprojects/exess/doc/c/overview.rst               |  151 ++
 subprojects/exess/doc/c/xml/meson.build            |   12 +
 subprojects/exess/doc/conf.py.in                   |  112 ++
 subprojects/exess/doc/cpp/Doxyfile.in              |   44 +
 subprojects/exess/doc/cpp/api/meson.build          |    5 +
 subprojects/exess/doc/cpp/index.rst                |   11 +
 subprojects/exess/doc/cpp/meson.build              |   42 +
 subprojects/exess/doc/cpp/overview.rst             |  151 ++
 subprojects/exess/doc/cpp/xml/meson.build          |   14 +
 subprojects/exess/doc/meson.build                  |   17 +
 subprojects/exess/doc/summary.rst                  |    3 +
 subprojects/exess/include/exess/exess.h            | 1633 ++++++++++++++++++++
 subprojects/exess/meson.build                      |  236 +++
 subprojects/exess/meson/meson.build                |  204 +++
 subprojects/exess/meson_options.txt                |   17 +
 subprojects/exess/scripts/dox_to_sphinx.py         |  675 ++++++++
 subprojects/exess/src/.clang-tidy                  |   12 +
 subprojects/exess/src/attributes.h                 |   30 +
 subprojects/exess/src/base64.c                     |  163 ++
 subprojects/exess/src/bigint.c                     |  605 ++++++++
 subprojects/exess/src/bigint.h                     |  118 ++
 subprojects/exess/src/boolean.c                    |   69 +
 subprojects/exess/src/byte.c                       |   45 +
 subprojects/exess/src/canonical.c                  |  309 ++++
 subprojects/exess/src/coerce.c                     |  422 +++++
 subprojects/exess/src/datatype.c                   |   79 +
 subprojects/exess/src/date.c                       |  112 ++
 subprojects/exess/src/date_utils.h                 |   65 +
 subprojects/exess/src/datetime.c                   |  265 ++++
 subprojects/exess/src/decimal.c                    |  267 ++++
 subprojects/exess/src/decimal.h                    |   63 +
 subprojects/exess/src/digits.c                     |  243 +++
 subprojects/exess/src/digits.h                     |   38 +
 subprojects/exess/src/double.c                     |   53 +
 subprojects/exess/src/duration.c                   |  322 ++++
 subprojects/exess/src/exess_config.h               |   79 +
 subprojects/exess/src/float.c                      |   44 +
 subprojects/exess/src/hex.c                        |  131 ++
 subprojects/exess/src/ieee_float.h                 |   63 +
 subprojects/exess/src/int.c                        |   45 +
 subprojects/exess/src/int_math.c                   |   78 +
 subprojects/exess/src/int_math.h                   |   72 +
 subprojects/exess/src/long.c                       |  110 ++
 subprojects/exess/src/macros.h                     |   31 +
 subprojects/exess/src/read_utils.c                 |   51 +
 subprojects/exess/src/read_utils.h                 |   77 +
 subprojects/exess/src/scientific.c                 |  125 ++
 subprojects/exess/src/scientific.h                 |   35 +
 subprojects/exess/src/short.c                      |   45 +
 subprojects/exess/src/soft_float.c                 |  161 ++
 subprojects/exess/src/soft_float.h                 |   71 +
 subprojects/exess/src/strerror.c                   |   70 +
 subprojects/exess/src/string_utils.h               |   74 +
 subprojects/exess/src/strtod.c                     |  405 +++++
 subprojects/exess/src/strtod.h                     |   33 +
 subprojects/exess/src/time.c                       |  172 +++
 subprojects/exess/src/time_utils.h                 |   35 +
 subprojects/exess/src/timezone.c                   |  150 ++
 subprojects/exess/src/timezone.h                   |   59 +
 subprojects/exess/src/ubyte.c                      |   45 +
 subprojects/exess/src/uint.c                       |   45 +
 subprojects/exess/src/ulong.c                      |   94 ++
 subprojects/exess/src/ushort.c                     |   45 +
 subprojects/exess/src/variant.c                    |  431 ++++++
 subprojects/exess/src/warnings.h                   |   46 +
 subprojects/exess/src/write_utils.c                |   50 +
 subprojects/exess/src/write_utils.h                |   85 +
 subprojects/exess/src/year.c                       |   98 ++
 subprojects/exess/test/.clang-tidy                 |   11 +
 subprojects/exess/test/float_test_data.h           |  130 ++
 subprojects/exess/test/int_test_data.h             |   42 +
 subprojects/exess/test/meson.build                 |   76 +
 subprojects/exess/test/num_test_utils.h            |   85 +
 subprojects/exess/test/test_base64.c               |  196 +++
 subprojects/exess/test/test_bigint.c               |  841 ++++++++++
 subprojects/exess/test/test_boolean.c              |  114 ++
 subprojects/exess/test/test_byte.c                 |   98 ++
 subprojects/exess/test/test_canonical.c            |  412 +++++
 subprojects/exess/test/test_coerce.c               |  519 +++++++
 subprojects/exess/test/test_datatype.c             |   81 +
 subprojects/exess/test/test_date.c                 |  257 +++
 subprojects/exess/test/test_datetime.c             |  459 ++++++
 subprojects/exess/test/test_decimal.c              |  260 ++++
 subprojects/exess/test/test_double.c               |  246 +++
 subprojects/exess/test/test_duration.c             |  310 ++++
 subprojects/exess/test/test_float.c                |  240 +++
 subprojects/exess/test/test_hex.c                  |  180 +++
 subprojects/exess/test/test_int.c                  |  129 ++
 subprojects/exess/test/test_int_math.c             |   97 ++
 subprojects/exess/test/test_long.c                 |  144 ++
 subprojects/exess/test/test_short.c                |  100 ++
 subprojects/exess/test/test_strerror.c             |   35 +
 subprojects/exess/test/test_time.c                 |  222 +++
 subprojects/exess/test/test_timezone.c             |  178 +++
 subprojects/exess/test/test_ubyte.c                |  100 ++
 subprojects/exess/test/test_uint.c                 |  128 ++
 subprojects/exess/test/test_ulong.c                |  126 ++
 subprojects/exess/test/test_ushort.c               |  100 ++
 subprojects/exess/test/test_variant.c              |  301 ++++
 subprojects/exess/test/time_test_utils.h           |   52 +
 120 files changed, 17581 insertions(+), 1 deletion(-)
 create mode 100644 subprojects/exess/.clang-format
 create mode 100644 subprojects/exess/.clang-tidy
 create mode 100644 subprojects/exess/.clant.json
 create mode 100644 subprojects/exess/.gitignore
 create mode 100644 subprojects/exess/.gitlab-ci.yml
 create mode 100644 subprojects/exess/.includes.imp
 create mode 100644 subprojects/exess/NEWS
 create mode 100644 subprojects/exess/README.md
 create mode 100644 subprojects/exess/bindings/cpp/include/exess/exess.hpp
 create mode 100644 subprojects/exess/bindings/cpp/meson.build
 create mode 100644 subprojects/exess/bindings/cpp/test/.clang-tidy
 create mode 100644 subprojects/exess/bindings/cpp/test/meson.build
 create mode 100644 subprojects/exess/bindings/cpp/test/test_exess_hpp.cpp
 create mode 100644 subprojects/exess/doc/c/Doxyfile.in
 create mode 100644 subprojects/exess/doc/c/api/meson.build
 create mode 100644 subprojects/exess/doc/c/index.rst
 create mode 100644 subprojects/exess/doc/c/man/manlink.in
 create mode 100644 subprojects/exess/doc/c/man/meson.build
 create mode 100644 subprojects/exess/doc/c/meson.build
 create mode 100644 subprojects/exess/doc/c/overview.rst
 create mode 100644 subprojects/exess/doc/c/xml/meson.build
 create mode 100644 subprojects/exess/doc/conf.py.in
 create mode 100644 subprojects/exess/doc/cpp/Doxyfile.in
 create mode 100644 subprojects/exess/doc/cpp/api/meson.build
 create mode 100644 subprojects/exess/doc/cpp/index.rst
 create mode 100644 subprojects/exess/doc/cpp/meson.build
 create mode 100644 subprojects/exess/doc/cpp/overview.rst
 create mode 100644 subprojects/exess/doc/cpp/xml/meson.build
 create mode 100644 subprojects/exess/doc/meson.build
 create mode 100644 subprojects/exess/doc/summary.rst
 create mode 100644 subprojects/exess/include/exess/exess.h
 create mode 100644 subprojects/exess/meson.build
 create mode 100644 subprojects/exess/meson/meson.build
 create mode 100644 subprojects/exess/meson_options.txt
 create mode 100755 subprojects/exess/scripts/dox_to_sphinx.py
 create mode 100644 subprojects/exess/src/.clang-tidy
 create mode 100644 subprojects/exess/src/attributes.h
 create mode 100644 subprojects/exess/src/base64.c
 create mode 100644 subprojects/exess/src/bigint.c
 create mode 100644 subprojects/exess/src/bigint.h
 create mode 100644 subprojects/exess/src/boolean.c
 create mode 100644 subprojects/exess/src/byte.c
 create mode 100644 subprojects/exess/src/canonical.c
 create mode 100644 subprojects/exess/src/coerce.c
 create mode 100644 subprojects/exess/src/datatype.c
 create mode 100644 subprojects/exess/src/date.c
 create mode 100644 subprojects/exess/src/date_utils.h
 create mode 100644 subprojects/exess/src/datetime.c
 create mode 100644 subprojects/exess/src/decimal.c
 create mode 100644 subprojects/exess/src/decimal.h
 create mode 100644 subprojects/exess/src/digits.c
 create mode 100644 subprojects/exess/src/digits.h
 create mode 100644 subprojects/exess/src/double.c
 create mode 100644 subprojects/exess/src/duration.c
 create mode 100644 subprojects/exess/src/exess_config.h
 create mode 100644 subprojects/exess/src/float.c
 create mode 100644 subprojects/exess/src/hex.c
 create mode 100644 subprojects/exess/src/ieee_float.h
 create mode 100644 subprojects/exess/src/int.c
 create mode 100644 subprojects/exess/src/int_math.c
 create mode 100644 subprojects/exess/src/int_math.h
 create mode 100644 subprojects/exess/src/long.c
 create mode 100644 subprojects/exess/src/macros.h
 create mode 100644 subprojects/exess/src/read_utils.c
 create mode 100644 subprojects/exess/src/read_utils.h
 create mode 100644 subprojects/exess/src/scientific.c
 create mode 100644 subprojects/exess/src/scientific.h
 create mode 100644 subprojects/exess/src/short.c
 create mode 100644 subprojects/exess/src/soft_float.c
 create mode 100644 subprojects/exess/src/soft_float.h
 create mode 100644 subprojects/exess/src/strerror.c
 create mode 100644 subprojects/exess/src/string_utils.h
 create mode 100644 subprojects/exess/src/strtod.c
 create mode 100644 subprojects/exess/src/strtod.h
 create mode 100644 subprojects/exess/src/time.c
 create mode 100644 subprojects/exess/src/time_utils.h
 create mode 100644 subprojects/exess/src/timezone.c
 create mode 100644 subprojects/exess/src/timezone.h
 create mode 100644 subprojects/exess/src/ubyte.c
 create mode 100644 subprojects/exess/src/uint.c
 create mode 100644 subprojects/exess/src/ulong.c
 create mode 100644 subprojects/exess/src/ushort.c
 create mode 100644 subprojects/exess/src/variant.c
 create mode 100644 subprojects/exess/src/warnings.h
 create mode 100644 subprojects/exess/src/write_utils.c
 create mode 100644 subprojects/exess/src/write_utils.h
 create mode 100644 subprojects/exess/src/year.c
 create mode 100644 subprojects/exess/test/.clang-tidy
 create mode 100644 subprojects/exess/test/float_test_data.h
 create mode 100644 subprojects/exess/test/int_test_data.h
 create mode 100644 subprojects/exess/test/meson.build
 create mode 100644 subprojects/exess/test/num_test_utils.h
 create mode 100644 subprojects/exess/test/test_base64.c
 create mode 100644 subprojects/exess/test/test_bigint.c
 create mode 100644 subprojects/exess/test/test_boolean.c
 create mode 100644 subprojects/exess/test/test_byte.c
 create mode 100644 subprojects/exess/test/test_canonical.c
 create mode 100644 subprojects/exess/test/test_coerce.c
 create mode 100644 subprojects/exess/test/test_datatype.c
 create mode 100644 subprojects/exess/test/test_date.c
 create mode 100644 subprojects/exess/test/test_datetime.c
 create mode 100644 subprojects/exess/test/test_decimal.c
 create mode 100644 subprojects/exess/test/test_double.c
 create mode 100644 subprojects/exess/test/test_duration.c
 create mode 100644 subprojects/exess/test/test_float.c
 create mode 100644 subprojects/exess/test/test_hex.c
 create mode 100644 subprojects/exess/test/test_int.c
 create mode 100644 subprojects/exess/test/test_int_math.c
 create mode 100644 subprojects/exess/test/test_long.c
 create mode 100644 subprojects/exess/test/test_short.c
 create mode 100644 subprojects/exess/test/test_strerror.c
 create mode 100644 subprojects/exess/test/test_time.c
 create mode 100644 subprojects/exess/test/test_timezone.c
 create mode 100644 subprojects/exess/test/test_ubyte.c
 create mode 100644 subprojects/exess/test/test_uint.c
 create mode 100644 subprojects/exess/test/test_ulong.c
 create mode 100644 subprojects/exess/test/test_ushort.c
 create mode 100644 subprojects/exess/test/test_variant.c
 create mode 100644 subprojects/exess/test/time_test_utils.h

diff --git a/meson.build b/meson.build
index 13e7ec3b..98ae0d5e 100644
--- a/meson.build
+++ b/meson.build
@@ -126,6 +126,10 @@ else
   prog_args = ['-DSERD_STATIC']
 endif
 
+exess_dep = dependency('exess',
+                       version: '>= 0.0.1',
+                       fallback: ['exess', 'exess_dep'])
+
 # Build shared and/or static library/libraries
 libserd = build_target(
   versioned_name,
@@ -133,7 +137,7 @@ libserd = build_target(
   version: meson.project_version(),
   include_directories: include_directories(['include']),
   c_args: c_warnings + library_args,
-  dependencies: m_dep,
+  dependencies: [exess_dep, m_dep],
   gnu_symbol_visibility: 'hidden',
   install: true,
   target_type: library_type)
diff --git a/subprojects/exess/.clang-format b/subprojects/exess/.clang-format
new file mode 100644
index 00000000..466ffccf
--- /dev/null
+++ b/subprojects/exess/.clang-format
@@ -0,0 +1,30 @@
+---
+AlignConsecutiveAssignments: true
+AlignConsecutiveDeclarations: true
+AlignEscapedNewlinesLeft: true
+BasedOnStyle: Mozilla
+BraceWrapping:
+  AfterNamespace: false
+  AfterClass: true
+  AfterEnum: false
+  AfterExternBlock: false
+  AfterFunction: true
+  AfterStruct: false
+  SplitEmptyFunction: false
+  SplitEmptyRecord: false
+BinPackArguments: false
+BinPackParameters: false
+BreakBeforeBraces: Custom
+Cpp11BracedListStyle: true
+FixNamespaceComments: true
+IndentCaseLabels: false
+IndentPPDirectives: AfterHash
+KeepEmptyLinesAtTheStartOfBlocks: false
+SpacesInContainerLiterals: false
+StatementMacros:
+  - EXESS_API
+  - EXESS_CONST_FUNC
+  - EXESS_PURE_FUNC
+  - _Generic
+  - _Pragma
+...
diff --git a/subprojects/exess/.clang-tidy b/subprojects/exess/.clang-tidy
new file mode 100644
index 00000000..d3e82197
--- /dev/null
+++ b/subprojects/exess/.clang-tidy
@@ -0,0 +1,7 @@
+Checks: >
+  *,
+  -clang-analyzer-security.insecureAPI.DeprecatedOrUnsafeBufferHandling,
+  -llvmlibc-*,
+WarningsAsErrors: '*'
+HeaderFilterRegex: '.*'
+FormatStyle: file
diff --git a/subprojects/exess/.clant.json b/subprojects/exess/.clant.json
new file mode 100644
index 00000000..76cae6d3
--- /dev/null
+++ b/subprojects/exess/.clant.json
@@ -0,0 +1,4 @@
+{
+  "version": "1.0.0",
+  "mapping_files": [".includes.imp"]
+}
diff --git a/subprojects/exess/.gitignore b/subprojects/exess/.gitignore
new file mode 100644
index 00000000..84c048a7
--- /dev/null
+++ b/subprojects/exess/.gitignore
@@ -0,0 +1 @@
+/build/
diff --git a/subprojects/exess/.gitlab-ci.yml b/subprojects/exess/.gitlab-ci.yml
new file mode 100644
index 00000000..52426131
--- /dev/null
+++ b/subprojects/exess/.gitlab-ci.yml
@@ -0,0 +1,168 @@
+stages:
+  - build
+  - deploy
+
+.build_template: &build_definition
+  stage: build
+
+
+arm32_dbg:
+  <<: *build_definition
+  image: lv2plugin/debian-arm32
+  script:
+    - meson . build --cross-file=/usr/share/meson/cross/arm-linux-gnueabihf.ini -Dbuildtype=debug -Dstrict=true -Dwerror=true
+    - ninja -C build test
+
+arm32_rel:
+  <<: *build_definition
+  image: lv2plugin/debian-arm32
+  script:
+    - meson . build --cross-file=/usr/share/meson/cross/arm-linux-gnueabihf.ini -Dbuildtype=release -Dstrict=true -Dwerror=true
+    - ninja -C build test
+
+
+arm64_dbg:
+  <<: *build_definition
+  image: lv2plugin/debian-arm64
+  script:
+    - meson . build --cross-file=/usr/share/meson/cross/aarch64-linux-gnu.ini -Dbuildtype=debug -Dstrict=true -Dwerror=true
+    - ninja -C build test
+
+arm64_rel:
+  <<: *build_definition
+  image: lv2plugin/debian-arm64
+  script:
+    - meson . build --cross-file=/usr/share/meson/cross/aarch64-linux-gnu.ini -Dbuildtype=release -Dstrict=true -Dwerror=true
+    - ninja -C build test
+
+
+x32_dbg:
+  <<: *build_definition
+  image: lv2plugin/debian-x32
+  script:
+    - meson . build --cross-file=/usr/share/meson/cross/i686-linux-gnu.ini -Dbuildtype=debug -Dstrict=true -Dwerror=true
+    - ninja -C build test
+
+x32_rel:
+  <<: *build_definition
+  image: lv2plugin/debian-x32
+  script:
+    - meson . build --cross-file=/usr/share/meson/cross/i686-linux-gnu.ini -Dbuildtype=release -Dstrict=true -Dwerror=true
+    - ninja -C build test
+
+
+x64_dbg:
+  <<: *build_definition
+  image: lv2plugin/debian-x64
+  script:
+    - meson . build -Dbuildtype=debug -Dstrict=true -Dwerror=true -Db_coverage=true
+    - ninja -C build test
+    - ninja -C build coverage-html
+  artifacts:
+    paths:
+      - build/meson-logs/coveragereport
+
+x64_rel:
+  <<: *build_definition
+  image: lv2plugin/debian-x64
+  script:
+    - meson . build -Dbuildtype=release -Dstrict=true -Dwerror=true
+    - ninja -C build test
+
+
+x64_static:
+  <<: *build_definition
+  image: lv2plugin/debian-x64
+  script:
+    - meson . build -Ddefault_library=static -Dstrict=true -Dwerror=true
+    - ninja -C build test
+
+
+x64_sanitize:
+  <<: *build_definition
+  image: lv2plugin/debian-x64-clang
+  script:
+    - meson . build -Db_lundef=false -Dbuildtype=plain -Dstrict=true -Dwerror=true
+    - ninja -C build test
+  variables:
+    CC: "clang"
+    CFLAGS: "-fno-sanitize-recover=all -fsanitize=address -fsanitize=undefined -fsanitize=float-divide-by-zero -fsanitize=implicit-conversion -fsanitize=local-bounds -fsanitize=nullability"
+    LDFLAGS: "-fno-sanitize-recover=all -fsanitize=address -fsanitize=undefined -fsanitize=float-divide-by-zero -fsanitize=implicit-conversion -fsanitize=local-bounds -fsanitize=nullability"
+
+
+mingw32_dbg:
+  <<: *build_definition
+  image: lv2plugin/debian-mingw32
+  script:
+    - meson . build --cross-file=/usr/share/meson/cross/i686-w64-mingw32.ini -Dbuildtype=debug -Dstrict=true -Dwerror=true
+    - ninja -C build test
+
+mingw32_rel:
+  <<: *build_definition
+  image: lv2plugin/debian-mingw32
+  script:
+    - meson . build --cross-file=/usr/share/meson/cross/i686-w64-mingw32.ini -Dbuildtype=release -Dstrict=true -Dwerror=true
+    - ninja -C build test
+
+
+mingw64_dbg:
+  <<: *build_definition
+  image: lv2plugin/debian-mingw64
+  script:
+    - meson . build --cross-file=/usr/share/meson/cross/x86_64-w64-mingw32.ini -Dbuildtype=debug -Dstrict=true -Dwerror=true
+    - ninja -C build test
+
+mingw64_rel:
+  <<: *build_definition
+  image: lv2plugin/debian-mingw64
+  script:
+    - meson . build --cross-file=/usr/share/meson/cross/x86_64-w64-mingw32.ini -Dbuildtype=release -Dstrict=true -Dwerror=true
+    - ninja -C build test
+
+
+mac_dbg:
+  <<: *build_definition
+  tags: [macos]
+  script:
+    - meson . build -Dbuildtype=debug -Dstrict=true -Dwerror=true
+    - ninja -C build test
+
+mac_rel:
+  <<: *build_definition
+  tags: [macos]
+  script:
+    - meson . build -Dbuildtype=release -Dstrict=true -Dwerror=true
+    - ninja -C build test
+
+
+win_dbg:
+  <<: *build_definition
+  tags: [windows,meson]
+  script:
+    - meson . build -Dbuildtype=debug -Dstrict=true -Dwerror=true
+    - ninja -C build test
+
+win_rel:
+  <<: *build_definition
+  tags: [windows,meson]
+  script:
+    - meson . build -Dbuildtype=release -Dstrict=true -Dwerror=true
+    - ninja -C build test
+
+
+pages:
+  stage: deploy
+  script:
+    - mkdir -p .public/doc
+    - mkdir -p .public/c
+    - mv build/meson-logs/coveragereport/ .public/coverage
+    - mv build/doc/c/singlehtml .public/c/singlehtml
+    - mv build/doc/c/html .public/c/html
+    - mv .public public
+  dependencies:
+    - x64_dbg
+  artifacts:
+    paths:
+      - public
+  only:
+    - master
diff --git a/subprojects/exess/.includes.imp b/subprojects/exess/.includes.imp
new file mode 100644
index 00000000..bf54f24d
--- /dev/null
+++ b/subprojects/exess/.includes.imp
@@ -0,0 +1,10 @@
+[
+  { "symbol": [ "int16_t",  "private", "<stdint.h>", "public" ] },
+  { "symbol": [ "int32_t",  "private", "<stdint.h>", "public" ] },
+  { "symbol": [ "int64_t",  "private", "<stdint.h>", "public" ] },
+  { "symbol": [ "int8_t",   "private", "<stdint.h>", "public" ] },
+  { "symbol": [ "uint16_t", "private", "<stdint.h>", "public" ] },
+  { "symbol": [ "uint32_t", "private", "<stdint.h>", "public" ] },
+  { "symbol": [ "uint64_t", "private", "<stdint.h>", "public" ] },
+  { "symbol": [ "uint8_t",  "private", "<stdint.h>", "public" ] }
+]
diff --git a/subprojects/exess/NEWS b/subprojects/exess/NEWS
new file mode 100644
index 00000000..26c425ff
--- /dev/null
+++ b/subprojects/exess/NEWS
@@ -0,0 +1,5 @@
+exess (0.0.1) unstable;
+
+  * Initial release
+
+ -- David Robillard <d@drobilla.net>  Thu, 14 Jan 2021 09:14:34 +0000
diff --git a/subprojects/exess/README.md b/subprojects/exess/README.md
new file mode 100644
index 00000000..c655ace7
--- /dev/null
+++ b/subprojects/exess/README.md
@@ -0,0 +1,83 @@
+Exess
+=====
+
+Exess is a simple library for reading and writing [XSD][] datatypes.
+
+Exess is useful for applications that need to read/write common datatypes
+from/to strings, in a standard and locale-independent format.  It supports
+reading any valid syntax, and writing in canonical form.  The implementation is
+not complete, but includes support for all of the common datatypes that are
+generally useful (the XML-specific and partial Gregorian calendar datatypes are
+omitted).
+
+Conversion to a string and back is lossless for all supported values.  For
+example, writing a `float` number to a string then reading it back will yield
+the exact same `float` as the original value.
+
+The API consists mainly of simple read and write functions for each datatype.
+A variant type is also included which allows generic code to work with values
+of any type.  For flexibility, allocation is handled by the caller, making it
+possible to work on the stack, or read and write values to fields in some
+structure.  Syntax errors are reported with a descriptive error code and
+character offset, allowing friendly error messages to be produced.
+
+Supported Datatypes
+-------------------
+
+Exess supports reading and writing:
+
+  * `boolean`, like "false", "true", "0", or "1".
+
+  * `decimal`, like "1.234" (stored as `double`).
+
+  * `float` and `double`, like "4.2E1" or "4.2e1".
+
+  * The unbounded integer types `integer`, `nonPositiveInteger`,
+    `negativeInteger`, `nonNegativeInteger`, and `nonPositiveInteger` (stored
+    as `int64_t` or `uint64_t`).
+
+  * The fixed size integer types `long`, `int`, `short`, `byte`,
+    `unsignedLong`, `unsignedInt`, `unsignedShort`, and `unsignedByte`.
+
+  * `duration`, like "P1Y6M".
+
+  * `time`, like "12:30:00.00".
+
+  * `date`, like "2001-12-31".
+
+  * `hex`, like "EC5355".
+
+  * `base64`,like "Zm9vYmFy".
+
+Dependencies
+------------
+
+None, except the C standard library.
+
+Building
+--------
+
+A [Meson][] build definition is included which can be used to do a proper
+system installation with a `pkg-config` file, generate IDE projects, run the
+tests, and so on.  For example, the library and tests can be built and run like
+so:
+
+    meson setup build
+    cd build
+    ninja test
+
+Documentation
+-------------
+
+ * [API reference (single page)](https://drobilla.gitlab.io/exess/c/singlehtml)
+ * [API reference (paginated)](https://drobilla.gitlab.io/exess/c/html)
+
+See the [Meson documentation][] for more details on using Meson.
+
+ -- David Robillard <d@drobilla.net>
+
+[XSD]: https://www.w3.org/TR/xmlschema-2/
+
+[Meson]: https://mesonbuild.com/
+
+[Meson documentation]: https://mesonbuild.com/Quick-guide.html
diff --git a/subprojects/exess/bindings/cpp/include/exess/exess.hpp b/subprojects/exess/bindings/cpp/include/exess/exess.hpp
new file mode 100644
index 00000000..7aa6fc5b
--- /dev/null
+++ b/subprojects/exess/bindings/cpp/include/exess/exess.hpp
@@ -0,0 +1,707 @@
+/*
+  Copyright 2021 David Robillard <d@drobilla.net>
+
+  Permission to use, copy, modify, and/or distribute this software for any
+  purpose with or without fee is hereby granted, provided that the above
+  copyright notice and this permission notice appear in all copies.
+
+  THIS SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+  WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+  MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+  ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+  WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+  ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+  OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+*/
+
+#ifndef EXESS_EXESS_HPP
+#define EXESS_EXESS_HPP
+
+#include "exess/exess.h"
+
+#include <cstdint>
+#include <exception>
+#include <iostream>
+#include <string>
+#include <type_traits>
+#include <typeinfo>
+#include <vector>
+
+namespace exess {
+
+/**
+   @defgroup exesspp Exess C++ API
+   This is the C++ wrapper for the exess API.
+   @{
+*/
+
+constexpr const char* EXESS_NONNULL const xsd_uri =
+  "http://www.w3.org/2001/XMLSchema#";
+
+// FIXME: Reorganize
+using Datatype = ExessDatatype;
+using Result   = ExessResult;
+
+using Duration = ExessDuration;
+using DateTime = ExessDateTime;
+using Time     = ExessTime;
+using Date     = ExessDate;
+
+/**
+   @defgroup exesspp_status Status
+   @copydoc exess_status
+   @{
+*/
+
+/// @copydoc ExessStatus
+enum class Status {
+  success,                ///< @copydoc EXESS_SUCCESS
+  expected_end,           ///< @copydoc EXESS_EXPECTED_END
+  expected_boolean,       ///< @copydoc EXESS_EXPECTED_BOOLEAN
+  expected_integer,       ///< @copydoc EXESS_EXPECTED_INTEGER
+  expected_duration,      ///< @copydoc EXESS_EXPECTED_DURATION
+  expected_sign,          ///< @copydoc EXESS_EXPECTED_SIGN
+  expected_digit,         ///< @copydoc EXESS_EXPECTED_DIGIT
+  expected_colon,         ///< @copydoc EXESS_EXPECTED_COLON
+  expected_dash,          ///< @copydoc EXESS_EXPECTED_DASH
+  expected_time_sep,      ///< @copydoc EXESS_EXPECTED_TIME_SEP
+  expected_time_tag,      ///< @copydoc EXESS_EXPECTED_TIME_TAG
+  expected_date_tag,      ///< @copydoc EXESS_EXPECTED_DATE_TAG
+  expected_hex,           ///< @copydoc EXESS_EXPECTED_HEX
+  expected_base64,        ///< @copydoc EXESS_EXPECTED_BASE64
+  bad_order,              ///< @copydoc EXESS_BAD_ORDER
+  bad_value,              ///< @copydoc EXESS_BAD_VALUE
+  out_of_range,           ///< @copydoc EXESS_OUT_OF_RANGE
+  no_space,               ///< @copydoc EXESS_NO_SPACE
+  would_reduce_precision, ///< @copydoc EXESS_WOULD_REDUCE_PRECISION
+  would_round,            ///< @copydoc EXESS_WOULD_ROUND
+  would_truncate,         ///< @copydoc EXESS_WOULD_TRUNCATE
+  unsupported,            ///< @copydoc EXESS_UNSUPPORTED
+};
+
+/// @copydoc exess_strerror
+inline const char* EXESS_NONNULL
+strerror(const Status status)
+{
+  return exess_strerror(static_cast<ExessStatus>(status));
+}
+
+inline std::ostream&
+operator<<(std::ostream& stream, const Status status)
+{
+  return stream << strerror(status);
+}
+
+/**
+   @}
+
+   Datatype Traits
+*/
+
+namespace detail {
+
+template<class T>
+struct DatatypeTraits {};
+
+template<>
+struct DatatypeTraits<bool> {
+  static constexpr auto datatype       = EXESS_BOOLEAN;
+  static constexpr auto read_function  = exess_read_boolean;
+  static constexpr auto write_function = exess_write_boolean;
+};
+
+template<>
+struct DatatypeTraits<double> {
+  static constexpr auto datatype       = EXESS_DOUBLE;
+  static constexpr auto read_function  = exess_read_double;
+  static constexpr auto write_function = exess_write_double;
+};
+
+template<>
+struct DatatypeTraits<float> {
+  static constexpr auto datatype       = EXESS_FLOAT;
+  static constexpr auto read_function  = exess_read_float;
+  static constexpr auto write_function = exess_write_float;
+};
+
+template<>
+struct DatatypeTraits<int64_t> {
+  static constexpr auto datatype       = EXESS_LONG;
+  static constexpr auto read_function  = exess_read_long;
+  static constexpr auto write_function = exess_write_long;
+};
+
+template<>
+struct DatatypeTraits<int32_t> {
+  static constexpr auto datatype       = EXESS_INT;
+  static constexpr auto read_function  = exess_read_int;
+  static constexpr auto write_function = exess_write_int;
+};
+
+template<>
+struct DatatypeTraits<int16_t> {
+  static constexpr auto datatype       = EXESS_SHORT;
+  static constexpr auto read_function  = exess_read_short;
+  static constexpr auto write_function = exess_write_short;
+};
+
+template<>
+struct DatatypeTraits<int8_t> {
+  static constexpr auto datatype       = EXESS_BYTE;
+  static constexpr auto read_function  = exess_read_byte;
+  static constexpr auto write_function = exess_write_byte;
+};
+
+template<>
+struct DatatypeTraits<uint64_t> {
+  static constexpr auto datatype       = EXESS_ULONG;
+  static constexpr auto read_function  = exess_read_ulong;
+  static constexpr auto write_function = exess_write_ulong;
+};
+
+template<>
+struct DatatypeTraits<uint32_t> {
+  static constexpr auto datatype       = EXESS_UINT;
+  static constexpr auto read_function  = exess_read_uint;
+  static constexpr auto write_function = exess_write_uint;
+};
+
+template<>
+struct DatatypeTraits<uint16_t> {
+  static constexpr auto datatype       = EXESS_USHORT;
+  static constexpr auto read_function  = exess_read_ushort;
+  static constexpr auto write_function = exess_write_ushort;
+};
+
+template<>
+struct DatatypeTraits<uint8_t> {
+  static constexpr auto datatype       = EXESS_UBYTE;
+  static constexpr auto read_function  = exess_read_ubyte;
+  static constexpr auto write_function = exess_write_ubyte;
+};
+
+template<>
+struct DatatypeTraits<Duration> {
+  static constexpr auto datatype       = EXESS_DURATION;
+  static constexpr auto read_function  = exess_read_duration;
+  static constexpr auto write_function = exess_write_duration;
+};
+
+template<>
+struct DatatypeTraits<DateTime> {
+  static constexpr auto datatype       = EXESS_DATETIME;
+  static constexpr auto read_function  = exess_read_datetime;
+  static constexpr auto write_function = exess_write_datetime;
+};
+
+template<>
+struct DatatypeTraits<Time> {
+  static constexpr auto datatype       = EXESS_TIME;
+  static constexpr auto read_function  = exess_read_time;
+  static constexpr auto write_function = exess_write_time;
+};
+
+template<>
+struct DatatypeTraits<Date> {
+  static constexpr auto datatype       = EXESS_DATE;
+  static constexpr auto read_function  = exess_read_date;
+  static constexpr auto write_function = exess_write_date;
+};
+
+} // namespace detail
+
+/**
+   @defgroup exesspp_read_write Reading and Writing
+   @{
+*/
+
+/**
+   Read a value from a string.
+
+   @param out Set to the parsed value.
+   @param str String input.
+   @return The `count` of characters read, and a `status` code.
+*/
+template<class T>
+inline Result
+read(T* EXESS_NONNULL out, const char* EXESS_NONNULL str)
+{
+  return detail::DatatypeTraits<T>::read_function(out, str);
+}
+
+/**
+   Write a value to a canonical string.
+
+   @param value Value to write.
+   @param buf_size The size of `buf` in bytes.
+   @param buf Output buffer, or null to only measure.
+
+   @return The `count` of characters in the output, and `status`
+   #Status::success, or #Status::no_space if the buffer is too small.
+*/
+template<class T>
+inline Result
+write(const T& value, const size_t buf_size, char* EXESS_NONNULL buf)
+{
+  return detail::DatatypeTraits<T>::write_function(value, buf_size, buf);
+}
+
+/**
+   Return a value as a string.
+
+   This is a wrapper for write() that allocates a new string of the appropriate
+   length, writes the value to it, and returns it.
+
+   @param value The value to convert to a string.
+   @return The value as a string, or the empty string on error.
+*/
+template<class T>
+inline std::string
+to_string(const T& value)
+{
+  auto r = detail::DatatypeTraits<T>::write_function(value, 0, nullptr);
+  if (r.status) {
+    return {};
+  }
+
+  // In C++17, std::string::data() allows mutable access
+#if __cplusplus >= 201703L
+  std::string string(r.count, ' ');
+  r = detail::DatatypeTraits<T>::write_function(
+    value, r.count + 1, string.data());
+
+  // Before, we had to allocate somewhere else
+#else
+  std::vector<char> buf(r.count + 1, '\0');
+  r = detail::DatatypeTraits<T>::write_function(value, r.count + 1, buf.data());
+
+  std::string string(buf.data());
+#endif
+
+  return r.status ? "" : string;
+}
+
+/**
+   @}
+*/
+
+// TODO: hex, base64
+
+struct Variant : ExessVariant {
+  explicit Variant(const Status status) noexcept
+    : ExessVariant{}
+  {
+    datatype        = EXESS_NOTHING;
+    value.as_status = static_cast<ExessStatus>(status);
+  }
+
+  explicit Variant(const bool v) noexcept
+    : ExessVariant{}
+  {
+    datatype      = EXESS_BOOLEAN;
+    value.as_bool = v;
+  }
+
+  explicit Variant(const double v) noexcept
+    : ExessVariant{}
+  {
+    datatype        = EXESS_DOUBLE;
+    value.as_double = v;
+  }
+
+  explicit Variant(const float v) noexcept
+    : ExessVariant{}
+  {
+    datatype       = EXESS_FLOAT;
+    value.as_float = v;
+  }
+
+  explicit Variant(const int64_t v) noexcept
+    : ExessVariant{}
+  {
+    datatype      = EXESS_LONG;
+    value.as_long = v;
+  }
+
+  explicit Variant(const int32_t v) noexcept
+    : ExessVariant{}
+  {
+    datatype     = EXESS_INT;
+    value.as_int = v;
+  }
+
+  explicit Variant(const int16_t v) noexcept
+    : ExessVariant{}
+  {
+    datatype       = EXESS_SHORT;
+    value.as_short = v;
+  }
+
+  explicit Variant(const int8_t v) noexcept
+    : ExessVariant{}
+  {
+    datatype      = EXESS_BYTE;
+    value.as_byte = v;
+  }
+
+  explicit Variant(const uint64_t v) noexcept
+    : ExessVariant{}
+  {
+    datatype       = EXESS_ULONG;
+    value.as_ulong = v;
+  }
+
+  explicit Variant(const uint32_t v) noexcept
+    : ExessVariant{}
+  {
+    datatype      = EXESS_UINT;
+    value.as_uint = v;
+  }
+
+  explicit Variant(const uint16_t v) noexcept
+    : ExessVariant{}
+  {
+    datatype        = EXESS_USHORT;
+    value.as_ushort = v;
+  }
+
+  explicit Variant(const uint8_t v) noexcept
+    : ExessVariant{}
+  {
+    datatype       = EXESS_UBYTE;
+    value.as_ubyte = v;
+  }
+
+  explicit Variant(const Duration v) noexcept
+    : ExessVariant{}
+  {
+    datatype          = EXESS_DURATION;
+    value.as_duration = v;
+  }
+
+  explicit Variant(const DateTime v) noexcept
+    : ExessVariant{}
+  {
+    datatype          = EXESS_DATETIME;
+    value.as_datetime = v;
+  }
+
+  explicit Variant(const Time v) noexcept
+    : ExessVariant{}
+  {
+    datatype      = EXESS_TIME;
+    value.as_time = v;
+  }
+
+  explicit Variant(const Date v) noexcept
+    : ExessVariant{}
+  {
+    datatype      = EXESS_DATE;
+    value.as_date = v;
+  }
+};
+
+inline Variant
+make_decimal(const double value) noexcept
+{
+  Variant result{value};
+  result.datatype = EXESS_DECIMAL;
+  return result;
+}
+
+inline Variant
+make_integer(const int64_t value) noexcept
+{
+  Variant result{value};
+  result.datatype = EXESS_INTEGER;
+  return result;
+}
+
+inline Variant
+make_non_positive_integer(const int64_t value) noexcept
+{
+  if (value > 0) {
+    return Variant{Status::out_of_range};
+  }
+
+  Variant result{value};
+  result.datatype = EXESS_NON_POSITIVE_INTEGER;
+  return result;
+}
+
+inline Variant
+make_negative_integer(const int64_t value) noexcept
+{
+  if (value >= 0) {
+    return Variant{Status::out_of_range};
+  }
+
+  Variant result{value};
+  result.datatype = EXESS_NEGATIVE_INTEGER;
+  return result;
+}
+
+inline Variant
+make_non_negative_integer(const uint64_t value) noexcept
+{
+  Variant result{value};
+  result.datatype = EXESS_NON_NEGATIVE_INTEGER;
+  return result;
+}
+
+inline Variant
+make_positive_integer(const uint64_t value) noexcept
+{
+  if (value == 0) {
+    return Variant{Status::out_of_range};
+  }
+
+  Variant result{value};
+  result.datatype = EXESS_POSITIVE_INTEGER;
+  return result;
+}
+
+// enum class Datatype {
+//   NOTHING,              ///< Sentinel for unknown datatypes or errors
+//   BOOLEAN,              ///< xsd:boolean (see @ref exess_boolean)
+//   DECIMAL,              ///< xsd:decimal (see @ref exess_decimal)
+//   DOUBLE,               ///< xsd:double (see @ref exess_double)
+//   FLOAT,                ///< xsd:float (see @ref exess_float)
+//   INTEGER,              ///< xsd:integer (see @ref exess_long)
+//   NON_POSITIVE_INTEGER, ///< xsd:nonPositiveInteger (see @ref exess_long)
+//   NEGATIVE_INTEGER,     ///< xsd:negativeInteger (see @ref exess_long)
+//   LONG,                 ///< xsd:long (see @ref exess_long)
+//   INT,                  ///< xsd:integer (see @ref exess_int)
+//   SHORT,                ///< xsd:short (see @ref exess_short)
+//   BYTE,                 ///< xsd:byte (see @ref exess_byte)
+//   NON_NEGATIVE_INTEGER, ///< xsd:nonNegativeInteger (see @ref exess_ulong)
+//   ULONG,                ///< xsd:unsignedLong (see @ref exess_ulong)
+//   UINT,                 ///< xsd:unsignedInt (see @ref exess_uint)
+//   USHORT,               ///< xsd:unsignedShort (see @ref exess_ushort)
+//   UBYTE,                ///< xsd:unsignedByte (see @ref exess_ubyte)
+//   POSITIVE_INTEGER,     ///< xsd:positiveInteger (see @ref exess_ulong)
+//   DURATION,             ///< xsd:duration (see @ref exess_duration)
+//   DATETIME,             ///< xsd:dateTime (see @ref exess_datetime)
+//   TIME,                 ///< xsd:time (see @ref exess_time)
+//   DATE,                 ///< xsd:date (see @ref exess_date)
+//   HEX,                  ///< xsd:hexBinary (see @ref exess_hex)
+//   BASE64,               ///< xsd:base64Binary (see @ref exess_base64)
+// }
+
+//
+
+/**
+   Return a pointer to the value of `variant` if it has type `T`.
+
+   This is safe to call on any variant, and will only return a pointer to the
+   value if the variant has the matching type and the value is valid to read.
+
+   @return A pointer to the value in `variant`, or null.
+*/
+template<class T>
+constexpr const T* EXESS_NULLABLE
+get_if(const Variant& variant) noexcept
+{
+  return (variant.datatype == detail::DatatypeTraits<T>::datatype)
+           ? reinterpret_cast<const T*>(&variant.value)
+           : nullptr;
+}
+
+/**
+   Return a pointer to the Status value in `variant` if it exists.
+*/
+template<>
+constexpr const Status* EXESS_NULLABLE
+get_if<Status>(const Variant& variant) noexcept
+{
+  return variant.datatype == EXESS_NOTHING
+           ? reinterpret_cast<const Status*>(&variant.value.as_status)
+           : nullptr;
+}
+
+/**
+   Return a pointer to the value of `variant` if it is a `double`.
+
+   This specialization works for both #EXESS_DECIMAL and #EXESS_DOUBLE values.
+*/
+template<>
+constexpr const double* EXESS_NULLABLE
+get_if<double>(const Variant& variant) noexcept
+{
+  return (variant.datatype == EXESS_DECIMAL || variant.datatype == EXESS_DOUBLE)
+           ? &variant.value.as_double
+           : nullptr;
+}
+
+/**
+   Return a pointer to the value of `variant` if it is an `int64_t` (long).
+
+   This specialization works for #EXESS_INTEGER, #EXESS_NON_POSITIVE_INTEGER,
+   #EXESS_NEGATIVE_INTEGER, and #EXESS_LONG values.
+*/
+template<>
+constexpr const int64_t* EXESS_NULLABLE
+get_if<int64_t>(const Variant& variant) noexcept
+{
+  return (variant.datatype >= EXESS_INTEGER && variant.datatype <= EXESS_LONG)
+           ? &variant.value.as_long
+           : nullptr;
+}
+
+/**
+   Return a pointer to the value of `variant` if it is a `uint64_t` (ulong).
+
+   This specialization works for #EXESS_NON_NEGATIVE_INTEGER, #EXESS_ULONG, and
+   #EXESS_POSITIVE_INTEGER values.
+*/
+template<>
+constexpr const uint64_t* EXESS_NULLABLE
+get_if<uint64_t>(const Variant& variant) noexcept
+{
+  return (variant.datatype == EXESS_NON_NEGATIVE_INTEGER ||
+          variant.datatype == EXESS_ULONG ||
+          variant.datatype == EXESS_POSITIVE_INTEGER)
+           ? &variant.value.as_ulong
+           : nullptr;
+}
+
+/**
+   Return the value of a variant with the given type.
+*/
+template<class T>
+const T&
+get(const Variant& variant)
+{
+  if (const T* const pointer = get_if<T>(variant)) {
+    return *pointer;
+  }
+
+  if (variant.datatype == EXESS_NOTHING) {
+    throw std::runtime_error{"Empty exess::Variant access"};
+  }
+
+  throw std::runtime_error{
+    std::string("Bad exess::Variant access: ") +
+    std::string(exess_datatype_uri(variant.datatype) + strlen(xsd_uri)) +
+    " from " + typeid(T).name()};
+}
+
+template<class T>
+constexpr size_t
+max_length()
+{
+  return 0u;
+}
+
+template<>
+constexpr size_t
+max_length<bool>()
+{
+  return EXESS_MAX_BOOLEAN_LENGTH;
+}
+
+template<>
+constexpr size_t
+max_length<double>()
+{
+  return EXESS_MAX_DOUBLE_LENGTH;
+}
+
+template<>
+constexpr size_t
+max_length<float>()
+{
+  return EXESS_MAX_FLOAT_LENGTH;
+}
+
+template<>
+constexpr size_t
+max_length<int64_t>()
+{
+  return EXESS_MAX_LONG_LENGTH;
+}
+
+template<>
+constexpr size_t
+max_length<int32_t>()
+{
+  return EXESS_MAX_INT_LENGTH;
+}
+
+template<>
+constexpr size_t
+max_length<int16_t>()
+{
+  return EXESS_MAX_SHORT_LENGTH;
+}
+
+template<>
+constexpr size_t
+max_length<int8_t>()
+{
+  return EXESS_MAX_BYTE_LENGTH;
+}
+
+template<>
+constexpr size_t
+max_length<uint64_t>()
+{
+  return EXESS_MAX_ULONG_LENGTH;
+}
+
+template<>
+constexpr size_t
+max_length<uint32_t>()
+{
+  return EXESS_MAX_UINT_LENGTH;
+}
+
+template<>
+constexpr size_t
+max_length<uint16_t>()
+{
+  return EXESS_MAX_USHORT_LENGTH;
+}
+
+template<>
+constexpr size_t
+max_length<uint8_t>()
+{
+  return EXESS_MAX_UBYTE_LENGTH;
+}
+
+template<>
+constexpr size_t
+max_length<Duration>()
+{
+  return EXESS_MAX_DURATION_LENGTH;
+}
+
+template<>
+constexpr size_t
+max_length<DateTime>()
+{
+  return EXESS_MAX_DATETIME_LENGTH;
+}
+
+template<>
+constexpr size_t
+max_length<Time>()
+{
+  return EXESS_MAX_TIME_LENGTH;
+}
+
+template<>
+constexpr size_t
+max_length<Date>()
+{
+  return EXESS_MAX_DATE_LENGTH;
+}
+
+/**
+   @}
+*/
+
+} // namespace exess
+
+#endif // EXESS_EXESS_HPP
diff --git a/subprojects/exess/bindings/cpp/meson.build b/subprojects/exess/bindings/cpp/meson.build
new file mode 100644
index 00000000..61e3616e
--- /dev/null
+++ b/subprojects/exess/bindings/cpp/meson.build
@@ -0,0 +1,54 @@
+add_languages(['cpp'])
+cpp = meson.get_compiler('cpp')
+
+# Set ultra strict warnings for developers, if requested
+if get_option('strict')
+  if cpp.get_id() == 'clang'
+    cpp_suppressions = [
+      '-Wno-c++14-extensions',
+      '-Wno-c++98-compat',
+      '-Wno-c++98-compat-pedantic',
+      '-Wno-documentation-unknown-command',
+      '-Wno-nullability-extension',
+      '-Wno-padded',
+    ]
+
+  elif cpp.get_id() == 'gcc'
+    cpp_suppressions = [
+      '-Wno-inline',
+      '-Wno-padded',
+      '-Wno-switch-default',
+      '-Wno-unused-const-variable',
+      '-Wno-useless-cast',
+    ]
+
+  elif cpp.get_id() == 'msvc'
+    cpp_suppressions = [
+      '/wd4028',  # formal parameter different from declaration
+      '/wd4204',  # nonstandard extension: non-constant aggregate initializer
+      '/wd4706',  # assignment within conditional expression
+      '/wd4710',  # function not inlined
+      '/wd4711',  # function selected for automatic inline expansion
+      '/wd4820',  # padding added after data member
+      '/wd5045',  # will insert Spectre mitigation
+    ]
+  endif
+
+else
+  if cpp.get_id() == 'clang'
+    cpp_suppressions = [
+      '-Wno-nullability-extension',
+    ]
+  endif
+endif
+
+exess_cpp_args = cc.get_supported_arguments(cpp_suppressions)
+
+cpp_headers = ['include/exess/exess.hpp']
+cpp_header_files = files(cpp_headers)
+
+exess_hpp_dep = declare_dependency(
+  dependencies: exess_dep,
+  include_directories: include_directories('include'))
+
+subdir('test')
diff --git a/subprojects/exess/bindings/cpp/test/.clang-tidy b/subprojects/exess/bindings/cpp/test/.clang-tidy
new file mode 100644
index 00000000..a0307839
--- /dev/null
+++ b/subprojects/exess/bindings/cpp/test/.clang-tidy
@@ -0,0 +1,17 @@
+Checks: >
+  *,
+  -*-magic-numbers,
+  -*-uppercase-literal-suffix,
+  -clang-analyzer-nullability.NullableDereferenced,
+  -cppcoreguidelines-pro-bounds-array-to-pointer-decay,
+  -cppcoreguidelines-pro-type-reinterpret-cast,
+  -fuchsia-default-arguments-calls,
+  -fuchsia-overloaded-operator,
+  -hicpp-no-array-decay,
+  -llvmlibc-*,
+  -misc-non-private-member-variables-in-classes,
+  -modernize-unary-static-assert,
+  -modernize-use-trailing-return-type,
+WarningsAsErrors: '*'
+HeaderFilterRegex: '.*'
+FormatStyle: file
diff --git a/subprojects/exess/bindings/cpp/test/meson.build b/subprojects/exess/bindings/cpp/test/meson.build
new file mode 100644
index 00000000..e04c21c9
--- /dev/null
+++ b/subprojects/exess/bindings/cpp/test/meson.build
@@ -0,0 +1,8 @@
+test('exess_hpp',
+     executable('test_exess_hpp',
+                'test_exess_hpp.cpp',
+                cpp_args: exess_cpp_args + prog_args,
+                dependencies: [exess_hpp_dep],
+                include_directories: include_directories('../../../test',
+                                                         '../../../src')),
+     suite: 'cpp')
diff --git a/subprojects/exess/bindings/cpp/test/test_exess_hpp.cpp b/subprojects/exess/bindings/cpp/test/test_exess_hpp.cpp
new file mode 100644
index 00000000..e940f4b1
--- /dev/null
+++ b/subprojects/exess/bindings/cpp/test/test_exess_hpp.cpp
@@ -0,0 +1,221 @@
+/*
+  Copyright 2021 David Robillard <d@drobilla.net>
+
+  Permission to use, copy, modify, and/or distribute this software for any
+  purpose with or without fee is hereby granted, provided that the above
+  copyright notice and this permission notice appear in all copies.
+
+  THIS SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+  WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+  MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+  ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+  WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+  ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+  OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+*/
+
+#undef NDEBUG
+
+#include "float_test_data.h"
+
+#include "exess/exess.h"
+#include "exess/exess.hpp"
+
+#include <cassert>
+#include <cstdint>
+#include <cstdlib>
+#include <iostream>
+#include <sstream>
+#include <stdexcept>
+#include <string>
+
+namespace exess {
+namespace {
+
+void
+test_status()
+{
+  std::ostringstream ss;
+  ss << Status::no_space;
+
+  std::cerr << ss.str();
+  assert(ss.str() == "Insufficient space");
+}
+
+void
+test_read()
+{
+  bool a_bool = false;
+  assert(!read(&a_bool, "true").status);
+  assert(a_bool == true);
+
+  double a_double = 0.0;
+  assert(!read(&a_double, "4.2E16").status);
+  assert(double_matches(a_double, 42000000000000000.0));
+
+  float a_float = 0.0f;
+  assert(!read(&a_float, "4.2E7").status);
+  assert(float_matches(a_float, 42000000.0f));
+
+  int64_t a_long = 0;
+  assert(!read(&a_long, "-4200000000").status);
+  assert(a_long == -4200000000);
+
+  int32_t a_int = 0;
+  assert(!read(&a_int, "-42000").status);
+  assert(a_int == -42000);
+
+  int16_t a_short = 0;
+  assert(!read(&a_short, "-420").status);
+  assert(a_short == -420);
+
+  int8_t a_byte = 0;
+  assert(!read(&a_byte, "-42").status);
+  assert(a_byte == -42);
+
+  uint64_t a_ulong = 0u;
+  assert(!read(&a_ulong, "4200000000").status);
+  assert(a_ulong == 4200000000);
+
+  uint32_t a_uint = 0u;
+  assert(!read(&a_uint, "42000").status);
+  assert(a_uint == 42000);
+
+  uint16_t a_ushort = 0u;
+  assert(!read(&a_ushort, "420").status);
+  assert(a_ushort == 420);
+
+  uint8_t a_ubyte = 0u;
+  assert(!read(&a_ubyte, "42").status);
+  assert(a_ubyte == 42);
+}
+
+void
+test_to_string()
+{
+  assert(to_string(true) == "true");
+  assert(to_string(42000000000000000.0) == "4.2E16");
+  assert(to_string(42000000.0f) == "4.2E7");
+  assert(to_string(int64_t(-4200000000)) == "-4200000000");
+  assert(to_string(int32_t(-42000)) == "-42000");
+  assert(to_string(int16_t(-420)) == "-420");
+  assert(to_string(int8_t(-42)) == "-42");
+  assert(to_string(uint64_t(4200000000u)) == "4200000000");
+  assert(to_string(uint32_t(42000u)) == "42000");
+  assert(to_string(uint16_t(420u)) == "420");
+  assert(to_string(uint8_t(42u)) == "42");
+}
+
+void
+test_max_length()
+{
+  static_assert(max_length<bool>() == EXESS_MAX_BOOLEAN_LENGTH, "");
+  static_assert(max_length<double>() == EXESS_MAX_DOUBLE_LENGTH, "");
+  static_assert(max_length<float>() == EXESS_MAX_FLOAT_LENGTH, "");
+  static_assert(max_length<int64_t>() == EXESS_MAX_LONG_LENGTH, "");
+  static_assert(max_length<int32_t>() == EXESS_MAX_INT_LENGTH, "");
+  static_assert(max_length<int16_t>() == EXESS_MAX_SHORT_LENGTH, "");
+  static_assert(max_length<int8_t>() == EXESS_MAX_BYTE_LENGTH, "");
+  static_assert(max_length<uint64_t>() == EXESS_MAX_ULONG_LENGTH, "");
+  static_assert(max_length<uint32_t>() == EXESS_MAX_UINT_LENGTH, "");
+  static_assert(max_length<uint16_t>() == EXESS_MAX_USHORT_LENGTH, "");
+  static_assert(max_length<uint8_t>() == EXESS_MAX_UBYTE_LENGTH, "");
+  static_assert(max_length<Duration>() == EXESS_MAX_DURATION_LENGTH, "");
+  static_assert(max_length<DateTime>() == EXESS_MAX_DATETIME_LENGTH, "");
+  static_assert(max_length<Date>() == EXESS_MAX_DATE_LENGTH, "");
+  static_assert(max_length<Time>() == EXESS_MAX_TIME_LENGTH, "");
+}
+
+template<class T>
+void
+check_get_throws(const Variant& variant)
+{
+  bool caught = false;
+
+  try {
+    get<T>(variant);
+  } catch (const std::runtime_error&) {
+    caught = true;
+  }
+
+  assert(caught);
+}
+
+void
+test_variant()
+{
+  const auto a_nothing              = Variant{Status::success};
+  const auto a_bool                 = Variant{true};
+  const auto a_decimal              = make_decimal(1.2);
+  const auto a_double               = Variant{3.4};
+  const auto a_float                = Variant{5.6f};
+  const auto a_integer              = make_integer(7);
+  const auto a_non_positive_integer = make_non_positive_integer(-8);
+  const auto a_negative_integer     = make_negative_integer(-9);
+  const auto a_long                 = Variant{int64_t(10)};
+  const auto a_int                  = Variant{int32_t(11)};
+  const auto a_short                = Variant{int16_t(12)};
+  const auto a_byte                 = Variant{int8_t(13)};
+  const auto a_non_negative_integer = make_non_negative_integer(14u);
+  const auto a_ulong                = Variant{uint64_t(15u)};
+  const auto a_uint                 = Variant{uint32_t(16u)};
+  const auto a_ushort               = Variant{uint16_t(17u)};
+  const auto a_ubyte                = Variant{uint8_t(18u)};
+  const auto a_positive_integer     = make_positive_integer(19u);
+
+  try {
+    assert(get<Status>(a_nothing) == Status::success);
+    assert(get<bool>(a_bool) == true);
+    assert(double_matches(get<double>(a_decimal), 1.2));
+    assert(double_matches(get<double>(a_double), 3.4));
+    assert(float_matches(get<float>(a_float), 5.6f));
+    assert(get<int64_t>(a_integer) == 7);
+    assert(get<int64_t>(a_non_positive_integer) == -8);
+    assert(get<int64_t>(a_negative_integer) == -9);
+    assert(get<int64_t>(a_long) == 10);
+    assert(get<int32_t>(a_int) == 11);
+    assert(get<int16_t>(a_short) == 12);
+    assert(get<int8_t>(a_byte) == 13);
+    assert(get<uint64_t>(a_non_negative_integer) == 14u);
+    assert(get<uint64_t>(a_ulong) == 15u);
+    assert(get<uint32_t>(a_uint) == 16u);
+    assert(get<uint16_t>(a_ushort) == 17u);
+    assert(get<uint8_t>(a_ubyte) == 18u);
+    assert(get<uint64_t>(a_positive_integer) == 19u);
+  } catch (const std::runtime_error&) {
+    abort();
+  }
+
+  check_get_throws<int>(a_nothing);
+  check_get_throws<int>(a_bool);
+  check_get_throws<int>(a_double);
+  check_get_throws<int>(a_float);
+  check_get_throws<int>(a_integer);
+  check_get_throws<int>(a_non_positive_integer);
+  check_get_throws<int>(a_negative_integer);
+  check_get_throws<int>(a_long);
+  check_get_throws<bool>(a_int);
+  check_get_throws<int>(a_short);
+  check_get_throws<int>(a_byte);
+  check_get_throws<int>(a_non_negative_integer);
+  check_get_throws<int>(a_ulong);
+  check_get_throws<int>(a_uint);
+  check_get_throws<int>(a_ushort);
+  check_get_throws<int>(a_ubyte);
+  check_get_throws<int>(a_positive_integer);
+}
+
+} // namespace
+} // namespace exess
+
+int
+main()
+{
+  exess::test_status();
+  exess::test_read();
+  exess::test_to_string();
+  exess::test_max_length();
+  exess::test_variant();
+
+  return 0;
+}
diff --git a/subprojects/exess/doc/c/Doxyfile.in b/subprojects/exess/doc/c/Doxyfile.in
new file mode 100644
index 00000000..9a768f89
--- /dev/null
+++ b/subprojects/exess/doc/c/Doxyfile.in
@@ -0,0 +1,39 @@
+PROJECT_NAME           = Exess
+PROJECT_BRIEF          = "A library for reading and writing XSD datatypes"
+
+QUIET                  = YES
+WARN_AS_ERROR          = YES
+WARN_IF_UNDOCUMENTED   = NO
+WARN_NO_PARAMDOC       = NO
+
+JAVADOC_AUTOBRIEF      = YES
+
+FULL_PATH_NAMES        = NO
+CASE_SENSE_NAMES       = YES
+HIDE_IN_BODY_DOCS      = YES
+REFERENCES_LINK_SOURCE = NO
+
+GENERATE_HTML          = NO
+GENERATE_LATEX         = NO
+GENERATE_XML           = YES
+XML_PROGRAMLISTING     = NO
+SHOW_FILES             = NO
+
+ENABLE_PREPROCESSING = YES
+SKIP_FUNCTION_MACROS = NO
+
+MACRO_EXPANSION        = YES
+EXPAND_ONLY_PREDEF     = YES
+PREDEFINED             = EXESS_API \
+                         EXESS_CONST_API \
+                         EXESS_CONST_FUNC= \
+                         EXESS_NONNULL= \
+                         EXESS_NULLABLE= \
+                         EXESS_PURE_API \
+                         EXESS_PURE_FUNC= \
+
+RECURSIVE              = YES
+STRIP_FROM_PATH        = @EXESS_SRCDIR@
+INPUT                  = @EXESS_SRCDIR@/include
+
+OUTPUT_DIRECTORY       = @DOX_OUTPUT@
diff --git a/subprojects/exess/doc/c/api/meson.build b/subprojects/exess/doc/c/api/meson.build
new file mode 100644
index 00000000..1a3d6f94
--- /dev/null
+++ b/subprojects/exess/doc/c/api/meson.build
@@ -0,0 +1,5 @@
+c_exess_rst = custom_target(
+  'Exess C API Sphinx Input',
+  command: [dox_to_sphinx, '-f', '@INPUT0@', meson.current_build_dir()],
+  input: [c_index_xml] + c_rst_files,
+  output: 'exess.rst')
diff --git a/subprojects/exess/doc/c/index.rst b/subprojects/exess/doc/c/index.rst
new file mode 100644
index 00000000..396a2333
--- /dev/null
+++ b/subprojects/exess/doc/c/index.rst
@@ -0,0 +1,10 @@
+#####
+Exess
+#####
+
+.. include:: summary.rst
+
+.. toctree::
+
+   overview
+   api/exess
diff --git a/subprojects/exess/doc/c/man/manlink.in b/subprojects/exess/doc/c/man/manlink.in
new file mode 100644
index 00000000..95accb8a
--- /dev/null
+++ b/subprojects/exess/doc/c/man/manlink.in
@@ -0,0 +1 @@
+.so @TARGET@
diff --git a/subprojects/exess/doc/c/man/meson.build b/subprojects/exess/doc/c/man/meson.build
new file mode 100644
index 00000000..d2a31557
--- /dev/null
+++ b/subprojects/exess/doc/c/man/meson.build
@@ -0,0 +1,122 @@
+# Run Sphinx to generate group man pages
+
+# Group man pages generated by Sphinx
+man_pages = [
+  'exess_base64.3',
+  'exess_binary.3',
+  'exess_boolean.3',
+  'exess_coercion.3',
+  'exess_byte.3',
+  'exess_datatypes.3',
+  'exess_date.3',
+  'exess_datetime.3',
+  'exess_decimal.3',
+  'exess_double.3',
+  'exess_duration.3',
+  'exess_float.3',
+  'exess_hex.3',
+  'exess_int.3',
+  'exess_long.3',
+  # 'exess_numbers.3',
+  'exess_short.3',
+  'exess_status.3',
+  'exess_time.3',
+  'exess_timezone.3',
+  'exess_ubyte.3',
+  'exess_uint.3',
+  'exess_ulong.3',
+  'exess_ushort.3',
+  'exess_variant.3',
+]
+
+# Run Sphinx to generate man pages
+man_docs = custom_target(
+  'man pages for exess',
+  command: [sphinx_build, '-M', 'man',
+            meson.current_build_dir() / '..', meson.current_build_dir() / '..',
+            '-E', '-q', '-t', 'man'],
+  input: [c_rst_files, c_exess_rst, c_index_xml],
+  output: man_pages,
+  build_by_default: true,
+  install: true,
+  install_dir: get_option('mandir') / 'man3')
+
+
+
+# message(man_docs.extract_all_objects())
+
+# foreach man_page : man_pages
+#   install_man(man_docs[0])#meson.current_build_dir() / man_page)
+# endforeach
+
+# "Links" from symbol names to corresponding page (one-line includes)
+man_links = [
+  ['ExessStatus.3', 'exess_status.3'],
+  ['ExessResult.3', 'exess_status.3'],
+  ['exess_strerror.3', 'exess_status.3'],
+  ['exess_read_decimal.3', 'exess_decimal.3'],
+  ['exess_write_decimal.3', 'exess_decimal.3'],
+  ['exess_read_double.3', 'exess_double.3'],
+  ['exess_write_double.3', 'exess_double.3'],
+  ['exess_read_float.3', 'exess_float.3'],
+  ['exess_write_float.3', 'exess_float.3'],
+  ['exess_read_boolean.3', 'exess_boolean.3'],
+  ['exess_write_boolean.3', 'exess_boolean.3'],
+  ['exess_read_long.3', 'exess_long.3'],
+  ['exess_write_long.3', 'exess_long.3'],
+  ['exess_read_int.3', 'exess_int.3'],
+  ['exess_write_int.3', 'exess_int.3'],
+  ['exess_read_short.3', 'exess_short.3'],
+  ['exess_write_short.3', 'exess_short.3'],
+  ['exess_read_byte.3', 'exess_byte.3'],
+  ['exess_write_byte.3', 'exess_byte.3'],
+  ['exess_read_ulong.3', 'exess_ulong.3'],
+  ['exess_write_ulong.3', 'exess_ulong.3'],
+  ['exess_read_uint.3', 'exess_uint.3'],
+  ['exess_write_uint.3', 'exess_uint.3'],
+  ['exess_read_ushort.3', 'exess_ushort.3'],
+  ['exess_write_ushort.3', 'exess_ushort.3'],
+  ['exess_read_ubyte.3', 'exess_ubyte.3'],
+  ['exess_write_ubyte.3', 'exess_ubyte.3'],
+  ['ExessDuration.3', 'exess_duration.3'],
+  ['exess_read_duration.3', 'exess_duration.3'],
+  ['exess_write_duration.3', 'exess_duration.3'],
+  ['ExessDateTime.3', 'exess_datetime.3'],
+  ['exess_read_datetime.3', 'exess_datetime.3'],
+  ['exess_write_datetime.3', 'exess_datetime.3'],
+  ['ExessTimezone.3', 'exess_timezone.3'],
+  ['ExessDate.3', 'exess_date.3'],
+  ['exess_read_date.3', 'exess_date.3'],
+  ['exess_write_date.3', 'exess_date.3'],
+  ['ExessTime.3', 'exess_time.3'],
+  ['exess_read_time.3', 'exess_time.3'],
+  ['exess_write_time.3', 'exess_time.3'],
+  ['ExessBlob.3', 'exess_binary.3'],
+  ['ExessDatatype.3', 'exess_datatypes.3'],
+  ['ExessValue.3', 'exess_variant.3'],
+  ['ExessVariant.3', 'exess_variant.3'],
+  ['exess_read_variant.3', 'exess_variant.3'],
+  ['exess_write_variant.3', 'exess_variant.3'],
+  ['exess_write_canonical.3', 'exess_variant.3'],
+  ['ExessCoercionFlag.3', 'exess_variant.3'],
+  ['ExessCoercionFlags.3', 'exess_variant.3'],
+  ['exess_coerce.3', 'exess_variant.3'],
+]
+
+foreach man_link : man_links
+  link = man_link[0]
+  target = man_link[1]
+
+  config = configuration_data()
+  config.set('TARGET', target)
+
+  manlink = configure_file(configuration: config,
+                           input: 'manlink.in',
+                           output: link,
+                           install: true,
+                           install_dir: get_option('mandir') / 'man3')
+
+  # message(target)
+
+  # install_man(manlink)
+endforeach
diff --git a/subprojects/exess/doc/c/meson.build b/subprojects/exess/doc/c/meson.build
new file mode 100644
index 00000000..64c02180
--- /dev/null
+++ b/subprojects/exess/doc/c/meson.build
@@ -0,0 +1,53 @@
+config = configuration_data()
+config.set('EXESS_VERSION', meson.project_version())
+
+conf_py = configure_file(configuration: config,
+                         input: '../conf.py.in',
+                         output: 'conf.py')
+
+configure_file(copy: true, input: '../summary.rst', output: 'summary.rst')
+
+c_rst_files = files(
+  'index.rst',
+  'overview.rst',
+)
+
+foreach f : c_rst_files
+  configure_file(copy: true, input: f, output: '@PLAINNAME@')
+endforeach
+
+subdir('xml')
+subdir('api')
+
+custom_target(
+  'Exess C API Documentation (singlehtml)',
+  command: [sphinx_build, '-M', 'singlehtml',
+            meson.current_build_dir(), meson.current_build_dir(),
+            '-E', '-q', '-t', 'singlehtml'],
+  input: [c_rst_files, c_exess_rst, c_index_xml],
+  output: 'singlehtml',
+  build_by_default: true,
+  install: true,
+  install_dir: docdir / 'exess-0')
+
+custom_target(
+  'Exess C API Documentation (html)',
+  command: [sphinx_build, '-M', 'html',
+            meson.current_build_dir(), meson.current_build_dir(),
+            '-E', '-q', '-t', 'html'],
+  input: [c_rst_files, c_exess_rst, c_index_xml],
+  output: 'html',
+  build_by_default: true,
+  install: true,
+  install_dir: docdir / 'exess-0')
+
+# custom_target(
+#   'man pages for exess',
+#   command: [sphinx_build, '-M', 'man',
+#             meson.current_build_dir(), meson.current_build_dir(),
+#             '-E', '-q', '-t', 'man'],
+#   input: [c_rst_files, c_exess_rst, c_index_xml],
+#   output: ['man/exess_base64.3'],
+#   build_by_default: true)
+
+subdir('man')
diff --git a/subprojects/exess/doc/c/overview.rst b/subprojects/exess/doc/c/overview.rst
new file mode 100644
index 00000000..1a2d3f6c
--- /dev/null
+++ b/subprojects/exess/doc/c/overview.rst
@@ -0,0 +1,151 @@
+########
+Overview
+########
+
+.. default-domain:: c
+.. highlight:: c
+
+The complete API is declared in ``exess.h``:
+
+.. code-block:: c
+
+   #include <exess/exess.h>
+
+**************
+Reading Values
+**************
+
+Each supported type has a read function that takes a pointer to an output value,
+and a string to read.
+It reads the value after skipping any leading whitespace,
+then returns an :struct:`ExessResult` with a ``status`` code and the ``count`` of characters read.
+For example:
+
+.. code-block:: c
+
+   int32_t     value = 0;
+   ExessResult r     = exess_read_int(&value, "1234");
+   if (!r.status) {
+     fprintf(stderr, "Read %zu bytes as int %d\n", r.count, value);
+   }
+
+If there was a syntax error,
+the status code indicates the specific problem.
+If a value was read but didn't end at whitespace or the end of the string,
+the status :enumerator:`EXESS_EXPECTED_END` is returned.
+This indicates that there is trailing garbage in the string,
+so the parse may not be complete or correct depending on the context.
+
+
+**************
+Writing Values
+**************
+
+The corresponding write function takes a value to write,
+a buffer size in bytes, and a buffer to write to.
+It returns an :struct:`ExessResult`,
+with a ``status`` code and the ``count`` of characters written,
+not including the trailing null byte.
+
+For datatypes with a bounded length,
+a constant like :var:`EXESS_MAX_INT_LENGTH` is the maximum length of the canonical representation of any value.
+This can be used to allocate buffers statically or on the stack,
+for example:
+
+.. code-block:: c
+
+   char buf[EXESS_MAX_INT_LENGTH + 1] = {0};
+
+   ExessResult r = exess_write_int(1234, sizeof(buf), buf);
+   if (!r.status) {
+     printf("Write error: %s\n", exess_strerror(r.status));
+   }
+
+******************
+Allocating Strings
+******************
+
+Exess doesn't do any allocation itself,
+so the calling code is responsible for providing a large enough buffer for output.
+The `count` returned by write functions can be used to determine the space required for a specific value.
+If the write function is called with a null output buffer,
+then this count is still returned as if a value were written.
+This can be used to precisely allocate memory for the string,
+taking care to allocate an extra byte for the null terminator.
+For example:
+
+.. code-block:: c
+
+   ExessResult r   = exess_write_int(1234, 0, NULL);
+   char*       str = (char*)calloc(r.count + 1, 1);
+
+   r = exess_write_int(1234, r.count + 1, buf);
+
+Note that for some types,
+this operation can be about as expensive as actually writing the value.
+For example, it requires binary to decimal conversion for floating point numbers.
+For ``float`` and ``double``,
+since the length is bounded and relatively small,
+it may be better to write immediately to a static buffer,
+then copy the result to the final destination.
+
+********
+Variants
+********
+
+The fundamental read and write functions all have similar semantics,
+but different type signatures since they use different value types.
+:struct:`ExessVariant` is a tagged union that can hold any supported value,
+allowing generic code to work with values of any type.
+
+Any value can be read with :func:`exess_read_variant` and written with :func:`exess_write_variant`,
+which work similarly to the fundamental read and write functions,
+except the read function takes an additional ``datatype`` parameter.
+The expected datatype must be provided,
+attempting to infer a datatype from the string content is not supported.
+
+Datatypes
+=========
+
+:enum:`ExessDatatype` enumerates all of the supported variant datatypes.
+The special value :enumerator:`EXESS_NOTHING` is used as a sentinel for unknown datatypes or other errors.
+
+If you have a datatype URI, then :func:`exess_datatype_from_uri()` can be used
+to map it to a datatype.  If the URI is not for a supported datatype, then it will return :enumerator:`EXESS_NOTHING`.
+
+Unbounded Numeric Types
+=======================
+
+There are 6 unbounded numeric types:
+decimal, integer, nonPositiveInteger, negativeInteger, nonNegativeInteger, and positiveInteger.
+:struct:`ExessVariant` supports reading and writing these types,
+but stores them in the largest corresponding native type:
+``double``, ``int64_t``, or ``uint64_t``.
+If the value doesn't fit in this type,
+then :func:`exess_read_variant` will return an :enumerator:`EXESS_OUT_OF_RANGE` error.
+
+Writing Canonical Form
+======================
+
+Since values are always written in canonical form,
+:struct:`ExessVariant` can be used as a generic mechanism to convert any string to canonical form:
+simply read a value,
+then write it.
+If the value itself isn't required,
+then :func:`exess_write_canonical` can be used to do this in a single step.
+For example, this will print ``123``:
+
+.. code-block:: c
+
+   char buf[4] = {0};
+
+   ExessResult r = exess_write_canonical(" +123", EXESS_INT, sizeof(buf), buf);
+   if (!r) {
+     printf("%s\n", buf);
+   }
+
+Note that it is better to use :func:`exess_write_canonical` if the value isn't required,
+since it supports transforming some values outside the range of :struct:`ExessVariant`.
+Specifically,
+decimal and integer strings will be transformed directly,
+avoiding conversion into values and the limits of the machine's numeric types.
diff --git a/subprojects/exess/doc/c/xml/meson.build b/subprojects/exess/doc/c/xml/meson.build
new file mode 100644
index 00000000..cfa726ed
--- /dev/null
+++ b/subprojects/exess/doc/c/xml/meson.build
@@ -0,0 +1,12 @@
+config = configuration_data()
+config.set('EXESS_SRCDIR', exess_src_root)
+config.set('DOX_OUTPUT', meson.current_build_dir() / '..')
+
+c_doxyfile = configure_file(configuration: config,
+                            input: '../Doxyfile.in',
+                            output: 'Doxyfile')
+
+c_index_xml = custom_target('exess-c-index.xml',
+                            command: [doxygen, '@INPUT0@'],
+                            input: [c_doxyfile] + c_header_files,
+                            output: 'index.xml')
diff --git a/subprojects/exess/doc/conf.py.in b/subprojects/exess/doc/conf.py.in
new file mode 100644
index 00000000..3eba5bc1
--- /dev/null
+++ b/subprojects/exess/doc/conf.py.in
@@ -0,0 +1,112 @@
+# Project information
+
+project = "Exess"
+copyright = "2021, David Robillard"
+author = "David Robillard"
+release = "@EXESS_VERSION@"
+
+# General configuration
+
+exclude_patterns = ["xml"]
+language = "en"
+nitpicky = True
+pygments_style = "friendly"
+
+# Ignore everything opaque or external for nitpicky mode
+_opaque = [
+    "int16_t",
+    "int32_t",
+    "int64_t",
+    "int8_t",
+    "size_t",
+    "uint16_t",
+    "uint32_t",
+    "uint64_t",
+    "uint8_t",
+]
+
+_c_nitpick_ignore = map(lambda x: ("c:identifier", x), _opaque)
+_cpp_nitpick_ignore = map(lambda x: ("cpp:identifier", x), _opaque)
+nitpick_ignore = list(_c_nitpick_ignore) + list(_cpp_nitpick_ignore)
+
+# HTML output
+
+html_copy_source = False
+html_short_title = "Exess"
+html_theme = "sphinx_lv2_theme"
+
+if tags.has("singlehtml"):
+    html_sidebars = {
+        "**": [
+            "globaltoc.html",
+        ]
+    }
+
+    html_theme_options = {
+        "body_max_width": "48em",
+        "body_min_width": "48em",
+        "description": "A library for reading and writing XSD datatypes",
+        "show_footer_version": True,
+        "show_logo_version": False,
+        "logo_name": True,
+        "logo_width": "8em",
+        "nosidebar": False,
+        "page_width": "80em",
+        "sidebar_width": "18em",
+        "globaltoc_maxdepth": 3,
+        "globaltoc_collapse": False,
+    }
+
+else:
+    html_theme_options = {
+        "body_max_width": "60em",
+        "body_min_width": "40em",
+        "description": "A library for reading and writing XSD datatypes",
+        "show_footer_version": True,
+        "show_logo_version": False,
+        "logo_name": True,
+        "logo_width": "8em",
+        "nosidebar": True,
+        "page_width": "60em",
+        "sidebar_width": "14em",
+        "globaltoc_maxdepth": 1,
+        "globaltoc_collapse": True,
+    }
+
+# Man page output
+
+groups = {
+    "status": "Status",
+    # "numbers": "",
+    "decimal": "Decimal Strings",
+    "double": "Double Strings",
+    "float": "Float Strings",
+    "boolean": "Boolean Strings",
+    "long": "Long Strings",
+    "int": "Int Strings",
+    "short": "Short Strings",
+    "byte": "Byte Strings",
+    "ulong": "Unsigned Long Strings",
+    "uint": "Unsigned Int Strings",
+    "ushort": "Unsigned Short Strings",
+    "ubyte": "Unsigned Byte Strings",
+    # 'time_and_date': "",
+    "duration": "Duration Strings",
+    "datetime": "Datetime Strings",
+    "timezone": "Time zone strings",
+    "date": "Date strings",
+    "time": "Time strings",
+    "binary": "",
+    "hex": "Hex Binary Strings",
+    "base64": "Base64 Binary Strings",
+    "datatypes": "Datatypes",
+    "variant": "Value Variant",
+    # 'generics': "",
+    "coercion": "Value Type Coercion",
+}
+
+author = "David Robillard <d@drobilla.net>"
+man_pages = []
+for group, title in groups.items():
+    name = "exess_" + group
+    man_pages += [("api/" + name, name, title, author, 3)]
diff --git a/subprojects/exess/doc/cpp/Doxyfile.in b/subprojects/exess/doc/cpp/Doxyfile.in
new file mode 100644
index 00000000..d254aa77
--- /dev/null
+++ b/subprojects/exess/doc/cpp/Doxyfile.in
@@ -0,0 +1,44 @@
+PROJECT_NAME           = Exess
+PROJECT_BRIEF          = "A library for reading and writing XSD datatypes"
+
+QUIET                  = YES
+WARN_AS_ERROR          = YES
+WARN_IF_UNDOCUMENTED   = NO
+WARN_NO_PARAMDOC       = NO
+
+JAVADOC_AUTOBRIEF      = YES
+
+CASE_SENSE_NAMES       = YES
+EXCLUDE_SYMBOLS        = exess::detail
+EXTRACT_LOCAL_CLASSES  = NO
+EXTRACT_PRIVATE        = NO
+HIDE_IN_BODY_DOCS      = YES
+HIDE_UNDOC_CLASSES     = YES
+HIDE_UNDOC_MEMBERS     = YES
+REFERENCES_LINK_SOURCE = NO
+
+GENERATE_HTML          = NO
+GENERATE_LATEX         = NO
+GENERATE_XML           = YES
+XML_PROGRAMLISTING     = NO
+SHOW_FILES             = NO
+
+ENABLE_PREPROCESSING   = YES
+SKIP_FUNCTION_MACROS   = NO
+
+EXPAND_ONLY_PREDEF     = YES
+MACRO_EXPANSION        = YES
+PREDEFINED             = EXESS_API \
+                         EXESS_CONST_API \
+                         EXESS_CONST_FUNC= \
+                         EXESS_NONNULL= \
+                         EXESS_NULLABLE= \
+                         EXESS_PURE_API \
+                         EXESS_PURE_FUNC= \
+
+RECURSIVE              = YES
+STRIP_FROM_PATH        = @EXESS_SRCDIR@
+INPUT                  = @EXESS_SRCDIR@/include \
+                         @EXESS_SRCDIR@/bindings/cpp/include
+
+OUTPUT_DIRECTORY       = @DOX_OUTPUT@
diff --git a/subprojects/exess/doc/cpp/api/meson.build b/subprojects/exess/doc/cpp/api/meson.build
new file mode 100644
index 00000000..6cdf4e04
--- /dev/null
+++ b/subprojects/exess/doc/cpp/api/meson.build
@@ -0,0 +1,5 @@
+cpp_exess_rst = custom_target(
+  'Exess C++ API Sphinx Input',
+  command: [dox_to_sphinx, '-l', 'cpp', '-f', '@INPUT@', 'doc/cpp/api'],
+  input: cpp_index_xml,
+  output: 'exess.rst')
diff --git a/subprojects/exess/doc/cpp/index.rst b/subprojects/exess/doc/cpp/index.rst
new file mode 100644
index 00000000..a88d3007
--- /dev/null
+++ b/subprojects/exess/doc/cpp/index.rst
@@ -0,0 +1,11 @@
+#####
+Exess
+#####
+
+.. include:: summary.rst
+
+.. toctree::
+
+   overview
+   api/exesspp
+   api/exess
diff --git a/subprojects/exess/doc/cpp/meson.build b/subprojects/exess/doc/cpp/meson.build
new file mode 100644
index 00000000..27dc34c5
--- /dev/null
+++ b/subprojects/exess/doc/cpp/meson.build
@@ -0,0 +1,42 @@
+config = configuration_data()
+config.set('EXESS_VERSION', meson.project_version())
+
+conf_py = configure_file(configuration: config,
+                         input: '../conf.py.in',
+                         output: 'conf.py')
+
+configure_file(copy: true, input: '../summary.rst', output: 'summary.rst')
+
+cpp_rst_files = files(
+  'index.rst',
+  'overview.rst',
+)
+
+foreach f : cpp_rst_files
+  configure_file(copy: true, input: f, output: '@PLAINNAME@')
+endforeach
+
+subdir('xml')
+subdir('api')
+
+docs = custom_target(
+  'Exess C++ API Documentation (singlehtml)',
+  command: [sphinx_build, '-M', 'singlehtml',
+            meson.current_build_dir(), meson.current_build_dir(),
+            '-E', '-q', '-t', 'singlehtml'],
+  input: [cpp_rst_files, cpp_exess_rst, cpp_index_xml],
+  output: 'singlehtml',
+  build_by_default: true,
+  install: true,
+  install_dir: docdir / 'exessxx-0')
+
+docs = custom_target(
+  'Exess C++ API Documentation (html)',
+  command: [sphinx_build, '-M', 'html',
+            meson.current_build_dir(), meson.current_build_dir(),
+            '-E', '-q', '-t', 'html'],
+  input: [cpp_rst_files, cpp_exess_rst, cpp_index_xml],
+  output: 'html',
+  build_by_default: true,
+  install: true,
+  install_dir: docdir / 'exessxx-0')
diff --git a/subprojects/exess/doc/cpp/overview.rst b/subprojects/exess/doc/cpp/overview.rst
new file mode 100644
index 00000000..3e59a953
--- /dev/null
+++ b/subprojects/exess/doc/cpp/overview.rst
@@ -0,0 +1,151 @@
+########
+Overview
+########
+
+.. default-domain:: cpp
+.. highlight:: cpp
+.. namespace:: exess
+
+The complete API is declared in ``exess.hpp``:
+
+.. code-block:: cpp
+
+   #include <exess/exess.hpp>
+
+**************
+Reading Values
+**************
+
+Each supported type has a read function that takes a pointer to an output value,
+and a string to read.
+It reads the value after skipping any leading whitespace,
+then returns an :struct:`ExessResult` with a ``status`` code and the ``count`` of characters read.
+For example:
+
+.. code-block:: cpp
+
+   int32_t     value = 0;
+   ExessResult r     = exess_read_int(&value, "1234");
+   if (!r.status) {
+     fprintf(stderr, "Read %zu bytes as int %d\n", r.count, value);
+   }
+
+If there was a syntax error,
+the status code indicates the specific problem.
+If a value was read but didn't end at whitespace or the end of the string,
+the status :enumerator:`EXESS_EXPECTED_END` is returned.
+This indicates that there is trailing garbage in the string,
+so the parse may not be complete or correct depending on the context.
+
+**************
+Writing Values
+**************
+
+The corresponding write function takes a value to write,
+a buffer size in bytes, and a buffer to write to.
+It returns an :struct:`ExessResult`,
+with a ``status`` code and the ``count`` of characters written,
+not including the trailing null byte.
+
+For datatypes with a bounded length,
+the `constexpr` function template :func:`max_length` returns the maximum length of the canonical representation of any value.
+This can be used to allocate buffers statically or on the stack,
+for example:
+
+.. code-block:: cpp
+
+   char buf[exess::max_length<int>() + 1] = {0};
+
+   exess::Result r = exess::write(1234, sizeof(buf), buf);
+   if (r.status != exess::Status::success) {
+     std::cerr << "Write error: " << exess::strerror(r.status) << "\n";
+   }
+
+******************
+Allocating Strings
+******************
+
+Exess doesn't do any allocation itself,
+so the calling code is responsible for providing a large enough buffer for output.
+The `count` returned by write functions can be used to determine the space required for a specific value.
+If the write function is called with a null output buffer,
+then this count is still returned as if a value were written.
+This can be used to precisely allocate memory for the string,
+taking care to allocate an extra byte for the null terminator.
+For example:
+
+.. code-block:: cpp
+
+   exess::Result r   = exess::write(1234, 0, NULL);
+   char*         str = (char*)calloc(r.count + 1, 1);
+
+   r = exess_write_int(1234, r.count + 1, buf);
+
+Note that for some types,
+this operation can be about as expensive as actually writing the value.
+For example, it requires binary to decimal conversion for floating point numbers.
+For ``float`` and ``double``,
+since the length is bounded and relatively small,
+it may be better to write immediately to a static buffer,
+then copy the result to the final destination.
+
+********
+Variants
+********
+
+The fundamental read and write functions all have similar semantics,
+but different type signatures since they use different value types.
+:struct:`ExessVariant` is a tagged union that can hold any supported value,
+allowing generic code to work with values of any type.
+
+Any value can be read with :func:`exess_read_variant` and written with :func:`exess_write_variant`,
+which work similarly to the fundamental read and write functions,
+except the read function takes an additional ``datatype`` parameter.
+The expected datatype must be provided,
+attempting to infer a datatype from the string content is not supported.
+
+Datatypes
+=========
+
+:enum:`ExessDatatype` enumerates all of the supported variant datatypes.
+The special value :enumerator:`EXESS_NOTHING` is used as a sentinel for unknown datatypes or other errors.
+
+If you have a datatype URI, then :func:`exess_datatype_from_uri()` can be used
+to map it to a datatype.  If the URI is not for a supported datatype, then it will return :enumerator:`EXESS_NOTHING`.
+
+Unbounded Numeric Types
+=======================
+
+There are 6 unbounded numeric types:
+decimal, integer, nonPositiveInteger, negativeInteger, nonNegativeInteger, and positiveInteger.
+:struct:`ExessVariant` supports reading and writing these types,
+but stores them in the largest corresponding native type:
+``double``, ``int64_t``, or ``uint64_t``.
+If the value doesn't fit in this type,
+then :func:`exess_read_variant` will return an :enumerator:`EXESS_OUT_OF_RANGE` error.
+
+Writing Canonical Form
+======================
+
+Since values are always written in canonical form,
+:struct:`ExessVariant` can be used as a generic mechanism to convert any string to canonical form:
+simply read a value,
+then write it.
+If the value itself isn't required,
+then :func:`exess_write_canonical` can be used to do this in a single step.
+For example, this will print ``123``:
+
+.. code-block:: cpp
+
+   char buf[4] = {0};
+
+   ExessResult r = exess_write_canonical(" +123", EXESS_INT, sizeof(buf), buf);
+   if (!r) {
+     printf("%s\n", buf);
+   }
+
+Note that it is better to use :func:`exess_write_canonical` if the value isn't required,
+since it supports transforming some values outside the range of :struct:`ExessVariant`.
+Specifically,
+decimal and integer strings will be transformed directly,
+avoiding conversion into values and the limits of the machine's numeric types.
diff --git a/subprojects/exess/doc/cpp/xml/meson.build b/subprojects/exess/doc/cpp/xml/meson.build
new file mode 100644
index 00000000..ee529bca
--- /dev/null
+++ b/subprojects/exess/doc/cpp/xml/meson.build
@@ -0,0 +1,14 @@
+config = configuration_data()
+config.set('EXESS_SRCDIR', exess_src_root)
+config.set('DOX_OUTPUT', meson.current_build_dir() / '..')
+
+cpp_doxyfile = configure_file(configuration: config,
+                              input: '../Doxyfile.in',
+                              output: 'Doxyfile')
+
+cpp_index_xml = custom_target(
+  'exess-cpp-index.xml',
+  command: [doxygen, '@INPUT0@'],
+  input: [cpp_doxyfile] + c_header_files + cpp_header_files,
+  output: 'index.xml')
+
diff --git a/subprojects/exess/doc/meson.build b/subprojects/exess/doc/meson.build
new file mode 100644
index 00000000..e69c4f05
--- /dev/null
+++ b/subprojects/exess/doc/meson.build
@@ -0,0 +1,17 @@
+docdir = get_option('datadir') / 'doc'
+
+doxygen = find_program('doxygen', required: get_option('docs'))
+dox_to_sphinx = find_program('../scripts/dox_to_sphinx.py')
+sphinx_build = find_program('sphinx-build', required: get_option('docs'))
+
+build_docs = doxygen.found() and sphinx_build.found()
+
+if build_docs
+  subdir('c')
+  subdir('cpp')
+endif
+
+if meson.version().version_compare('>=0.53.0')
+  summary('Documentation', build_docs, bool_yn: true)
+endif
+
diff --git a/subprojects/exess/doc/summary.rst b/subprojects/exess/doc/summary.rst
new file mode 100644
index 00000000..7e4dd84f
--- /dev/null
+++ b/subprojects/exess/doc/summary.rst
@@ -0,0 +1,3 @@
+Exess is a simple C library for reading and writing XSD_ datatypes.
+
+.. _XSD: https://www.w3.org/TR/xmlschema-2/
diff --git a/subprojects/exess/include/exess/exess.h b/subprojects/exess/include/exess/exess.h
new file mode 100644
index 00000000..f327eb2e
--- /dev/null
+++ b/subprojects/exess/include/exess/exess.h
@@ -0,0 +1,1633 @@
+/*
+  Copyright 2021 David Robillard <d@drobilla.net>
+
+  Permission to use, copy, modify, and/or distribute this software for any
+  purpose with or without fee is hereby granted, provided that the above
+  copyright notice and this permission notice appear in all copies.
+
+  THIS SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+  WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+  MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+  ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+  WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+  ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+  OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+*/
+
+#ifndef EXESS_EXESS_H
+#define EXESS_EXESS_H
+
+#include <stdbool.h>
+#include <stddef.h>
+#include <stdint.h>
+
+#if defined(_WIN32) && !defined(EXESS_STATIC) && defined(EXESS_INTERNAL)
+#  define EXESS_API __declspec(dllexport)
+#elif defined(_WIN32) && !defined(EXESS_STATIC)
+#  define EXESS_API __declspec(dllimport)
+#elif defined(__GNUC__)
+#  define EXESS_API __attribute__((visibility("default")))
+#else
+#  define EXESS_API
+#endif
+
+#ifdef __GNUC__
+#  define EXESS_PURE_FUNC __attribute__((pure))
+#  define EXESS_CONST_FUNC __attribute__((const))
+#else
+#  define EXESS_PURE_FUNC
+#  define EXESS_CONST_FUNC
+#endif
+
+#if defined(__clang__) && __clang_major__ >= 7
+#  define EXESS_NONNULL _Nonnull
+#  define EXESS_NULLABLE _Nullable
+#else
+#  define EXESS_NONNULL
+#  define EXESS_NULLABLE
+#endif
+
+// Pure API functions have no observable side-effects
+#define EXESS_PURE_API \
+  EXESS_API            \
+  EXESS_PURE_FUNC
+
+// Const API functions are pure, and read no memory other than their parameters
+#define EXESS_CONST_API \
+  EXESS_API             \
+  EXESS_CONST_FUNC
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+/**
+   @defgroup exess Exess C API
+   This is the complete public C API of exess.
+   @{
+*/
+
+/// The base URI of XML Schema, `http://www.w3.org/2001/XMLSchema#`
+#define EXESS_XSD_URI "http://www.w3.org/2001/XMLSchema#"
+
+/**
+   @defgroup exess_status Status
+
+   Status codes and return values used for error handling.
+
+   Success and various specific errors are reported by an integer status code,
+   which can be converted to a string to produce friendly error messages.
+   Reading and writing functions return a "result", which has a status code
+   along with a count of bytes read or written.
+
+   @{
+*/
+
+/// Status code to describe errors or other relevant situations
+typedef enum {
+  EXESS_SUCCESS,                ///< Success
+  EXESS_EXPECTED_END,           ///< Expected end of value
+  EXESS_EXPECTED_BOOLEAN,       ///< Expected "false", "true", "0" or "1"
+  EXESS_EXPECTED_INTEGER,       ///< Expected an integer value
+  EXESS_EXPECTED_DURATION,      ///< Expected a duration starting with 'P'
+  EXESS_EXPECTED_SIGN,          ///< Expected '-' or '+'
+  EXESS_EXPECTED_DIGIT,         ///< Expected a digit
+  EXESS_EXPECTED_COLON,         ///< Expected ':'
+  EXESS_EXPECTED_DASH,          ///< Expected '-'
+  EXESS_EXPECTED_TIME_SEP,      ///< Expected 'T'
+  EXESS_EXPECTED_TIME_TAG,      ///< Expected 'H', 'M', or 'S'
+  EXESS_EXPECTED_DATE_TAG,      ///< Expected 'Y', 'M', or 'D'
+  EXESS_EXPECTED_HEX,           ///< Expected a hexadecimal character
+  EXESS_EXPECTED_BASE64,        ///< Expected a base64 character
+  EXESS_BAD_ORDER,              ///< Invalid field order
+  EXESS_BAD_VALUE,              ///< Invalid value
+  EXESS_OUT_OF_RANGE,           ///< Value out of range for datatype
+  EXESS_NO_SPACE,               ///< Insufficient space
+  EXESS_WOULD_REDUCE_PRECISION, ///< Precision reducing coercion required
+  EXESS_WOULD_ROUND,            ///< Rounding coercion required
+  EXESS_WOULD_TRUNCATE,         ///< Truncating coercion required
+  EXESS_UNSUPPORTED,            ///< Unsupported value
+} ExessStatus;
+
+/**
+   Result returned from a read or write function.
+
+   This combines a status code with a byte offset, so it can be used to
+   determine how many characters were read or written, or what error occurred
+   at what character offset.
+*/
+typedef struct {
+  ExessStatus status; ///< Status code
+  size_t      count;  ///< Number of bytes read or written, excluding null
+} ExessResult;
+
+/**
+   Return a string describing a status code in plain English.
+
+   The returned string is always one sentence, with an uppercase first
+   character, and no trailing period.
+*/
+EXESS_CONST_API
+const char* EXESS_NONNULL
+exess_strerror(ExessStatus status);
+
+/**
+   @}
+   @defgroup exess_numbers Numbers
+   Datatypes for numbers
+   @{
+*/
+
+/**
+   @defgroup exess_decimal Decimal
+
+   An xsd:decimal is a decimal number of arbitrary precision, but this
+   implementation only supports values that fit in a `double`.
+
+   Unlike xsd:double, xsd:decimal is written in numeric form, never in
+   scientific notation.  Special infinity and NaN values are not supported.
+   Note that the xsd:decimal representation for some numbers is very long, so
+   xsd:double may be a better choice for values in a wide range.
+
+   Canonical form has no leading "+" sign, and at most 1 leading or trailing
+   zero such that there is at least 1 digit on either side of the decimal
+   point, like "12.34", "-1.0", and "0.0".
+
+   Non-canonical form allows a leading "+", any number of leading and trailing
+   zeros, any number of digits (including zero) on either side of the point,
+   and does not require a decimal point, like "+1", "01", "-.5", "4.", and
+   "42".
+
+   @{
+*/
+
+/// The maximum length of an xsd:decimal string from exess_write_decimal(), 327
+#define EXESS_MAX_DECIMAL_LENGTH 327
+
+/**
+   Read an xsd:decimal string after any leading whitespace.
+
+   Values beyond the range of `decimal` will produce `-INF` or `INF`, and
+   return an error because these are not valid decimal values.
+
+   @param out Set to the parsed value, or NaN on error.
+   @param str String input.
+   @return The `count` of characters read, and a `status` code.
+*/
+EXESS_API
+ExessResult
+exess_read_decimal(double* EXESS_NONNULL out, const char* EXESS_NONNULL str);
+
+/**
+   Write a canonical xsd:decimal string.
+
+   @param value Value to write.
+   @param buf_size The size of `buf` in bytes.
+   @param buf Output buffer, or null to only measure.
+
+   @return The `count` of characters in the output, and `status`
+   #EXESS_SUCCESS, or #EXESS_NO_SPACE if the buffer is too small.
+*/
+EXESS_API
+ExessResult
+exess_write_decimal(double value, size_t buf_size, char* EXESS_NULLABLE buf);
+
+/**
+   @}
+   @defgroup exess_double Double
+
+   An xsd:double is an IEEE-754 64-bit floating point number, written in
+   scientific notation.
+
+   Canonical form has no leading "+" sign, at most 1 leading or trailing zero
+   such that there is at least 1 digit on either side of the decimal point, and
+   always has an exponent, like "12.34E56", "-1.0E-2", and "-0.0E0".  The
+   special values negative infinity, positive infinity, and not-a-number are
+   written "-INF", "INF", and "NaN", respectively.
+
+   Non-canonical form allows a leading "+", any number of leading and trailing
+   zeros, any number of digits (including zero) on either side of the point,
+   and does not require an exponent or decimal point, like "+1E3", "1E+3",
+   ".5E3", "4.2", and "42".
+
+   @{
+*/
+
+/// The maximum length of a canonical xsd:double string, 24
+#define EXESS_MAX_DOUBLE_LENGTH 24
+
+/**
+   Read an xsd:double string after any leading whitespace.
+
+   Values beyond the range of `double` will produce `-INF` or `INF`.
+
+   @param out Set to the parsed value, or `NAN` on error.
+   @param str String input.
+   @return The `count` of characters read, and a `status` code.
+*/
+EXESS_API
+ExessResult
+exess_read_double(double* EXESS_NONNULL out, const char* EXESS_NONNULL str);
+
+/**
+   Write a canonical xsd:double string.
+
+   Any `double` value is supported.  Reading the resulting string with
+   exess_read_double() will produce exactly `value`, except the extra bits in
+   NaNs are not preserved.
+
+   @param value Value to write.
+   @param buf_size The size of `buf` in bytes.
+   @param buf Output buffer, or null to only measure.
+
+   @return The `count` of characters in the output and `status` #EXESS_SUCCESS,
+   or #EXESS_NO_SPACE if the buffer is too small.
+*/
+EXESS_API
+ExessResult
+exess_write_double(double value, size_t buf_size, char* EXESS_NULLABLE buf);
+
+/**
+   @}
+   @defgroup exess_float Float
+
+   An xsd:float is an IEEE-754 32-bit floating point number, written in
+   scientific notation.
+
+   The lexical form is the same as xsd:double, the only difference is that the
+   value space of xsd:float is smaller.  See @ref exess_double for details.
+
+   @{
+*/
+
+/// The maximum length of a canonical xsd:float string, 15
+#define EXESS_MAX_FLOAT_LENGTH 15
+
+/**
+   Read an xsd:float string after any leading whitespace.
+
+   Values beyond the range of `float` will produce `-INF` or `INF`.
+
+   @param out Set to the parsed value, or `NAN` on error.
+   @param str String input.
+   @return The `count` of characters read, and a `status` code.
+*/
+EXESS_API
+ExessResult
+exess_read_float(float* EXESS_NONNULL out, const char* EXESS_NONNULL str);
+
+/**
+   Write a canonical xsd:float string.
+
+   Any `float` value is supported.  Reading the resulting string with
+   exess_read_float() will produce exactly `value`, except the extra bits in
+   NaNs are not preserved.
+
+   @param value Value to write.
+   @param buf_size The size of `buf` in bytes.
+   @param buf Output buffer, or null to only measure.
+
+   @return The `count` of characters in the output, and `status`
+   #EXESS_SUCCESS, or #EXESS_NO_SPACE if the buffer is too small.
+*/
+EXESS_API
+ExessResult
+exess_write_float(float value, size_t buf_size, char* EXESS_NULLABLE buf);
+
+/**
+   @}
+   @defgroup exess_boolean Boolean
+
+   An xsd:boolean has only two possible values, canonically written as "false"
+   and "true".  The non-canonical forms "0" and "1" are also supported.
+
+   @{
+*/
+
+/// The maximum length of a canonical xsd:boolean string, 5
+#define EXESS_MAX_BOOLEAN_LENGTH 5
+
+/**
+   Read an xsd:boolean string after any leading whitespace.
+
+   @param out Set to the parsed value, or false on error.
+   @param str String input.
+   @return The `count` of characters read, and a `status` code.
+*/
+EXESS_API
+ExessResult
+exess_read_boolean(bool* EXESS_NONNULL out, const char* EXESS_NONNULL str);
+
+/**
+   Write a canonical xsd:boolean string.
+
+   @param value Value to write.
+   @param buf_size The size of `buf` in bytes.
+   @param buf Output buffer, or null to only measure.
+
+   @return The `count` of characters in the output, and `status`
+   #EXESS_SUCCESS, or #EXESS_NO_SPACE if the buffer is too small.
+*/
+EXESS_API
+ExessResult
+exess_write_boolean(bool value, size_t buf_size, char* EXESS_NULLABLE buf);
+
+/**
+   @}
+   @defgroup exess_long Long
+
+   An xsd:long is a signed 64-bit integer, written in decimal.
+
+   Values range from -9223372036854775808 to 9223372036854775807 inclusive.
+
+   Canonical form has no leading "+" sign and no leading zeros (except for the
+   number "0"), like "-1", "0", and "1234".
+
+   Non-canonical form allows a leading "+" and any number of leading zeros,
+   like "01" and "+0001234".
+
+   @{
+*/
+
+/// The maximum length of a canonical xsd:long string, 20
+#define EXESS_MAX_LONG_LENGTH 20
+
+/**
+   Read an xsd:long string after any leading whitespace.
+
+   @param out Set to the parsed value, or zero on error.
+   @param str String input.
+   @return The `count` of characters read, and a `status` code.
+*/
+EXESS_API
+ExessResult
+exess_read_long(int64_t* EXESS_NONNULL out, const char* EXESS_NONNULL str);
+
+/**
+   Write a canonical xsd:long string.
+
+   @param value Value to write.
+   @param buf_size The size of `buf` in bytes.
+   @param buf Output buffer, or null to only measure.
+
+   @return The `count` of characters in the output, and `status`
+   #EXESS_SUCCESS, or #EXESS_NO_SPACE if the buffer is too small.
+*/
+EXESS_API
+ExessResult
+exess_write_long(int64_t value, size_t buf_size, char* EXESS_NULLABLE buf);
+
+/**
+   @}
+   @defgroup exess_int Int
+
+   An xsd:int is a signed 32-bit integer.
+
+   Values range from -2147483648 to 2147483647 inclusive.
+
+   The lexical form is the same as xsd:long, the only difference is that the
+   value space of xsd:int is smaller.  See @ref exess_long for details.
+
+   @{
+*/
+
+/// The maximum length of a canonical xsd:int string, 11
+#define EXESS_MAX_INT_LENGTH 11
+
+/**
+   Read an xsd:int string after any leading whitespace.
+
+   @param out Set to the parsed value, or zero on error.
+   @param str String input.
+   @return The `count` of characters read, and a `status` code.
+*/
+EXESS_API
+ExessResult
+exess_read_int(int32_t* EXESS_NONNULL out, const char* EXESS_NONNULL str);
+
+/**
+   Write a canonical xsd:int string.
+
+   @param value Value to write.
+   @param buf_size The size of `buf` in bytes.
+   @param buf Output buffer, or null to only measure.
+
+   @return The `count` of characters in the output, and `status`
+   #EXESS_SUCCESS, or #EXESS_NO_SPACE if the buffer is too small.
+*/
+EXESS_API
+ExessResult
+exess_write_int(int32_t value, size_t buf_size, char* EXESS_NULLABLE buf);
+
+/**
+   @}
+   @defgroup exess_short Short
+
+   An xsd:short is a signed 16-bit integer.
+
+   Values range from -32768 to 32767 inclusive.
+
+   The lexical form is the same as xsd:long, the only difference is that the
+   value space of xsd:short is smaller.  See @ref exess_long for details.
+
+   @{
+*/
+
+/// The maximum length of a canonical xsd:short string, 6
+#define EXESS_MAX_SHORT_LENGTH 6
+
+/**
+   Read an xsd:short string after any leading whitespace.
+
+   @param out Set to the parsed value, or zero on error.
+   @param str String input.
+   @return The `count` of characters read, and a `status` code.
+*/
+EXESS_API
+ExessResult
+exess_read_short(int16_t* EXESS_NONNULL out, const char* EXESS_NONNULL str);
+
+/**
+   Write a canonical xsd:short string.
+
+   @param value Value to write.
+   @param buf_size The size of `buf` in bytes.
+   @param buf Output buffer, or null to only measure.
+
+   @return The `count` of characters in the output, and `status`
+   #EXESS_SUCCESS, or #EXESS_NO_SPACE if the buffer is too small.
+*/
+EXESS_API
+ExessResult
+exess_write_short(int16_t value, size_t buf_size, char* EXESS_NULLABLE buf);
+
+/**
+   @}
+   @defgroup exess_byte Byte
+
+   An xsd:byte is a signed 8-bit integer.
+
+   Values range from -128 to 127 inclusive.
+
+   The lexical form is the same as xsd:long, the only difference is that the
+   value space of xsd:byte is smaller.  See @ref exess_long for details.
+
+   @{
+*/
+
+/// The maximum length of a canonical xsd:byte string, 4
+#define EXESS_MAX_BYTE_LENGTH 4
+
+/**
+   Read an xsd:byte string after any leading whitespace.
+
+   @param out Set to the parsed value, or zero on error.
+   @param str String input.
+   @return The `count` of characters read, and a `status` code.
+*/
+EXESS_API
+ExessResult
+exess_read_byte(int8_t* EXESS_NONNULL out, const char* EXESS_NONNULL str);
+
+/**
+   Write a canonical xsd:byte string.
+
+   @param value Value to write.
+   @param buf_size The size of `buf` in bytes.
+   @param buf Output buffer, or null to only measure.
+
+   @return The `count` of characters in the output, and `status`
+   #EXESS_SUCCESS, or #EXESS_NO_SPACE if the buffer is too small.
+*/
+EXESS_API
+ExessResult
+exess_write_byte(int8_t value, size_t buf_size, char* EXESS_NULLABLE buf);
+
+/**
+   @}
+   @defgroup exess_ulong Unsigned Long
+
+   An xsd:unsignedLong is an unsigned 64-bit integer, written in decimal.
+
+   Values range from 0 to 18446744073709551615 inclusive.
+
+   Canonical form has no leading "+" sign and no leading zeros (except for the
+   number "0"), like "0", and "1234".
+
+   Non-canonical form allows any number of leading zeros, like "01" and
+   "0001234".
+
+   @{
+*/
+
+/// The maximum length of a canonical xsd:unsignedLong string, 20
+#define EXESS_MAX_ULONG_LENGTH 20
+
+/**
+   Read an xsd:unsignedLong string after any leading whitespace.
+
+   @param out Set to the parsed value, or zero on error.
+   @param str String input.
+   @return The `count` of characters read, and a `status` code.
+*/
+EXESS_API
+ExessResult
+exess_read_ulong(uint64_t* EXESS_NONNULL out, const char* EXESS_NONNULL str);
+
+/**
+   Write a canonical xsd:unsignedLong string.
+
+   @param value Value to write.
+   @param buf_size The size of `buf` in bytes.
+   @param buf Output buffer, or null to only measure.
+
+   @return The `count` of characters in the output, and `status`
+   #EXESS_SUCCESS, or #EXESS_NO_SPACE if the buffer is too small.
+*/
+EXESS_API
+ExessResult
+exess_write_ulong(uint64_t value, size_t buf_size, char* EXESS_NULLABLE buf);
+
+/**
+   @}
+   @defgroup exess_uint Unsigned Int
+
+   An xsd:unsignedInt is an unsigned 32-bit integer.
+
+   Values range from 0 to 4294967295 inclusive.
+
+   The lexical form is the same as xsd:unsignedLong, the only difference is
+   that the value space of xsd:unsignedInt is smaller.  See @ref exess_ulong
+   for details.
+
+   @{
+*/
+
+/// The maximum length of a canonical xsd:unsignedInt string, 10
+#define EXESS_MAX_UINT_LENGTH 10
+
+/**
+   Read an xsd:unsignedInt string after any leading whitespace.
+
+   @param out Set to the parsed value, or zero on error.
+   @param str String input.
+   @return The `count` of characters read, and a `status` code.
+*/
+EXESS_API
+ExessResult
+exess_read_uint(uint32_t* EXESS_NONNULL out, const char* EXESS_NONNULL str);
+
+/**
+   Write a canonical xsd:unsignedInt string.
+
+   @param value Value to write.
+   @param buf_size The size of `buf` in bytes.
+   @param buf Output buffer, or null to only measure.
+
+   @return The `count` of characters in the output, and `status`
+   #EXESS_SUCCESS, or #EXESS_NO_SPACE if the buffer is too small.
+*/
+EXESS_API
+ExessResult
+exess_write_uint(uint32_t value, size_t buf_size, char* EXESS_NULLABLE buf);
+
+/**
+   @}
+   @defgroup exess_ushort Unsigned Short
+
+   An xsd:unsignedShort is an unsigned 16-bit integer.
+
+   Values range from 0 to 65535 inclusive.
+
+   The lexical form is the same as xsd:unsignedLong, the only difference is
+   that the value space of xsd:unsignedShort is smaller.  See @ref exess_ulong
+   for details.
+
+   @{
+*/
+
+/// The maximum length of a canonical xsd:unsignedShort string, 5
+#define EXESS_MAX_USHORT_LENGTH 5
+
+/**
+   Read an xsd:unsignedShort string after any leading whitespace.
+
+   @param out Set to the parsed value, or zero on error.
+   @param str String input.
+   @return The `count` of characters read, and a `status` code.
+*/
+EXESS_API
+ExessResult
+exess_read_ushort(uint16_t* EXESS_NONNULL out, const char* EXESS_NONNULL str);
+
+/**
+   Write a canonical xsd:unsignedShort string.
+
+   @param value Value to write.
+   @param buf_size The size of `buf` in bytes.
+   @param buf Output buffer, or null to only measure.
+
+   @return The `count` of characters in the output, and `status`
+   #EXESS_SUCCESS, or #EXESS_NO_SPACE if the buffer is too small.
+*/
+EXESS_API
+ExessResult
+exess_write_ushort(uint16_t value, size_t buf_size, char* EXESS_NULLABLE buf);
+
+/**
+   @}
+   @defgroup exess_ubyte Unsigned Byte
+
+   An xsd:unsignedByte is an unsigned 8-bit integer.  Values range from 0 to
+   255 inclusive.
+
+   The lexical form is the same as xsd:unsignedLong, the only difference is
+   that the value space of xsd:unsignedByte is smaller.  See @ref exess_ulong
+   for details.
+
+   @{
+*/
+
+/// The maximum length of a canonical xsd:unsignedByte string, 3
+#define EXESS_MAX_UBYTE_LENGTH 3
+
+/**
+   Read an xsd:unsignedByte string after any leading whitespace.
+
+   @param out Set to the parsed value, or zero on error.
+   @param str String input.
+   @return The `count` of characters read, and a `status` code.
+*/
+EXESS_API
+ExessResult
+exess_read_ubyte(uint8_t* EXESS_NONNULL out, const char* EXESS_NONNULL str);
+
+/**
+   Write a canonical xsd:unsignedByte string.
+
+   @param value Value to write.
+   @param buf_size The size of `buf` in bytes.
+   @param buf Output buffer, or null to only measure.
+
+   @return The `count` of characters in the output, and `status`
+   #EXESS_SUCCESS, or #EXESS_NO_SPACE if the buffer is too small.
+*/
+EXESS_API
+ExessResult
+exess_write_ubyte(uint8_t value, size_t buf_size, char* EXESS_NULLABLE buf);
+
+/**
+   @}
+   @}
+   @defgroup exess_time_and_date Time and Date
+   Datatypes for dates, times, and durations of time.
+   @{
+*/
+
+/**
+   @defgroup exess_duration Duration
+
+   An xsd:duration is a positive or negative duration of time, written in ISO
+   8601 format like "PnYnMnDTnHnMnS" where each "n" is a number and fields may
+   be omitted if they are zero.
+
+   All numbers must be integers, except for seconds which may be a decimal.  If
+   seconds is a decimal, then at least one digit must follow the decimal point.
+   A negative duration is written with "-" as the first character, for example
+   "-P60D".
+
+   Canonical form omits all zero fields and writes no leading or trailing
+   zeros, except for the zero duration which is written "P0Y", for example
+   "P1DT2H", "PT30M", or "PT4.5S".
+
+   Non-canonical form allows zero fields, leading zeros, and for seconds to be
+   written as a decimal even if it is integer, for example "P06D", "PT7.0S", or
+   "P0Y0M01DT06H00M00S".
+
+   @{
+*/
+
+/// The maximum length of an xsd:duration string from exess_write_duration(), 41
+#define EXESS_MAX_DURATION_LENGTH 41
+
+/**
+   A duration of time (xsd:duration value).
+
+   To save space and to simplify arithmetic, this representation only stores
+   two values: integer months, and decimal seconds (to nanosecond precision).
+   These values are converted to and from the other fields during writing and
+   reading.  Years and months are stored as months, and days, hours, minutes,
+   and seconds are stored as seconds.
+
+   The sign of all members must match, so a negative duration has all
+   non-positive members, and a positive duration has all non-negative members.
+*/
+typedef struct {
+  int32_t months;
+  int32_t seconds;
+  int32_t nanoseconds;
+} ExessDuration;
+
+/**
+   Read an xsd:duration string after any leading whitespace.
+
+   @param out Set to the parsed value, or zero on error.
+   @param str String input.
+   @return The `count` of characters read, and a `status` code.
+*/
+EXESS_API
+ExessResult
+exess_read_duration(ExessDuration* EXESS_NONNULL out,
+                    const char* EXESS_NONNULL    str);
+
+/**
+   Write a canonical xsd:duration string.
+
+   @param value Value to write.
+   @param buf_size The size of `buf` in bytes.
+   @param buf Output buffer, or null to only measure.
+
+   @return #EXESS_SUCCESS on success, #EXESS_NO_SPACE if the buffer is too
+   small, or #EXESS_BAD_VALUE if the value is invalid.
+*/
+EXESS_API
+ExessResult
+exess_write_duration(ExessDuration        value,
+                     size_t               buf_size,
+                     char* EXESS_NULLABLE buf);
+/**
+   @}
+   @defgroup exess_datetime Datetime
+
+   An xsd:datetime is a date and time in either UTC or local time.
+
+   Strings have the form YYYY-MM-DDTHH:MM:SS with at least 4 year digits
+   (negative or positive), and all other fields positive two-digit integers
+   except seconds which may be a decimal, for example "2001-02-03T12:13:14.56".
+   Nanosecond precision is supported.
+
+   A local datetime has no suffix, a datetime with a time zone is always in
+   UTC, and is written with a "Z" suffix, for example 2001-02-03T12:13:14Z.
+
+   Canonical form only includes a decimal point if the number of seconds is not
+   an integer.
+
+   This implementation supports up to nanosecond resolution.
+
+   @{
+*/
+
+/// The maximum length of an xsd:dateTime string from exess_write_datetime(), 32
+#define EXESS_MAX_DATETIME_LENGTH 32
+
+/**
+   A date and time (xsd:dateTime value).
+
+   This representation follows the syntax, except the UTC flag is stored
+   between the date and time for more efficient packing.
+*/
+typedef struct {
+  int16_t  year;       ///< Year: any positive or negative value
+  uint8_t  month;      ///< Month: [1, 12]
+  uint8_t  day;        ///< Day: [1, 31]
+  uint8_t  is_utc;     ///< True if this is UTC (not local) time
+  uint8_t  hour;       ///< Hour: [0, 23]
+  uint8_t  minute;     ///< Minute: [0, 59]
+  uint8_t  second;     ///< Second: [0, 59]
+  uint32_t nanosecond; ///< Nanosecond: [0, 999999999]
+} ExessDateTime;
+
+/**
+   Add a duration to a datetime.
+
+   This advances or rewinds the datetime by the given duration, depending on
+   whether the duration is positive or negative.
+
+   If underflow or overflow occur, then this will return an infinite value.  A
+   positive infinity has all fields at maximum, and a negative infinity has all
+   fields at minimum, except `is_utc` which is preserved from the input (so
+   infinities are comparable with the values they came from).  Since 0 and 255
+   are never valid months, these can be tested for by checking if the year and
+   month are `INT16_MIN` and 0, or `INT16_MAX` and `INT8_MAX`.
+
+   @return `s + d`, or an infinite past or infinite future if underflow or
+   overflow occurs.
+*/
+EXESS_CONST_API
+ExessDateTime
+exess_add_datetime_duration(ExessDateTime s, ExessDuration d);
+
+/**
+   Read an xsd:dateTime string after any leading whitespace.
+
+   @param out Set to the parsed value, or zero on error.
+   @param str String input.
+   @return The `count` of characters read, and a `status` code.
+*/
+EXESS_API
+ExessResult
+exess_read_datetime(ExessDateTime* EXESS_NONNULL out,
+                    const char* EXESS_NONNULL    str);
+
+/**
+   Write a canonical xsd:datetime string.
+
+   @param value Value to write.
+   @param buf_size The size of `buf` in bytes.
+   @param buf Output buffer, or null to only measure.
+
+   @return #EXESS_SUCCESS on success, #EXESS_NO_SPACE if the buffer is too
+   small, or #EXESS_BAD_VALUE if the value is invalid.
+*/
+EXESS_API
+ExessResult
+exess_write_datetime(ExessDateTime        value,
+                     size_t               buf_size,
+                     char* EXESS_NULLABLE buf);
+
+/**
+   @}
+   @defgroup exess_timezone Timezones
+
+   Date and time values can have a timezone qualifier suffix.  Note that
+   timezone is not a datatype, one only exists as a part of another value.
+
+   Canonical form starts with a sign, followed by two-digit hour and minute
+   offsets separated by a colon, like "-06:00" and "+02:30".  The zero offset,
+   UTC, is written "Z".
+
+   Non-canonical form also allows writing UTC as "-00:00" or "+00:00".
+
+   This implementation only supports a resolution of 15 minutes, that is, only
+   offsets at 0, 15, 30, and 45 minutes within an hour.
+
+   @{
+*/
+
+/// A time zone offset for a date or time value
+typedef struct {
+  int8_t quarter_hours; ///< Offset in quarter hours: [-56, 56]
+} ExessTimezone;
+
+/// Sentinel value for local time, `INT8_MAX`
+#define EXESS_LOCAL INT8_MAX
+
+/**
+   @}
+   @defgroup exess_date Date
+   An xsd:date is a year, month, and day, with optional timezone.
+   @{
+*/
+
+/// The maximum length of an xsd:date string from exess_write_date(), 18
+#define EXESS_MAX_DATE_LENGTH 18
+
+/// Date (xsd:date)
+typedef struct {
+  int16_t       year;
+  uint8_t       month;
+  uint8_t       day;
+  ExessTimezone zone;
+} ExessDate;
+
+/**
+   Read an xsd:date string after any leading whitespace.
+
+   @param out Set to the parsed value, or zero on error.
+   @param str String input.
+   @return The `count` of characters read, and a `status` code.
+*/
+EXESS_API
+ExessResult
+exess_read_date(ExessDate* EXESS_NONNULL out, const char* EXESS_NONNULL str);
+
+/**
+   Write a canonical xsd:date string.
+
+   The output is always in canonical form, like `2001-04-12` or
+   `-2001-10-26+02:00`.
+
+   @param value Value to write.
+   @param buf_size The size of `buf` in bytes.
+   @param buf Output buffer, or null to only measure.
+
+   @return #EXESS_SUCCESS on success, #EXESS_NO_SPACE if the buffer is too
+   small, or #EXESS_BAD_VALUE if the value is invalid.
+*/
+EXESS_API
+ExessResult
+exess_write_date(ExessDate value, size_t buf_size, char* EXESS_NULLABLE buf);
+
+/**
+   @}
+   @defgroup exess_time Time
+
+   An xsd:time is a time of day, with optional timezone.
+
+   @{
+*/
+
+/// The maximum length of an xsd:time string from exess_write_time(), 24
+#define EXESS_MAX_TIME_LENGTH 24
+
+/// Time (xsd:time)
+typedef struct {
+  ExessTimezone zone;       ///< Time zone
+  uint8_t       hour;       ///< Hour: [0, 23]
+  uint8_t       minute;     ///< Minute: [0, 59]
+  uint8_t       second;     ///< Second: [0, 59]
+  uint32_t      nanosecond; ///< Nanosecond: [0, 999999999]
+} ExessTime;
+
+/**
+   Read an xsd:time string after any leading whitespace.
+
+   @param out Set to the parsed value, or zero on error.
+   @param str String input.
+   @return The `count` of characters read, and a `status` code.
+*/
+EXESS_API
+ExessResult
+exess_read_time(ExessTime* EXESS_NONNULL out, const char* EXESS_NONNULL str);
+
+/**
+   Write a canonical xsd:time string.
+
+   The output is always in canonical form, like "12:15" or "02:00Z".
+
+   @param value Value to write.
+   @param buf_size The size of `buf` in bytes.
+   @param buf Output buffer, or null to only measure.
+
+   @return #EXESS_SUCCESS on success, #EXESS_NO_SPACE if the buffer is too
+   small, or #EXESS_BAD_VALUE if the value is invalid.
+*/
+EXESS_API
+ExessResult
+exess_write_time(ExessTime value, size_t buf_size, char* EXESS_NULLABLE buf);
+
+/**
+   @}
+   @}
+   @defgroup exess_binary Binary
+   Datatypes for arbitrary binary data.
+   @{
+*/
+
+typedef struct {
+  size_t               size;
+  void* EXESS_NULLABLE data;
+} ExessBlob;
+
+/**
+   @defgroup exess_base64 Base64
+   An xsd:base64Binary is arbitrary binary data in base64 encoding.
+   @{
+*/
+
+/**
+   Return the maximum number of bytes required to decode `length` bytes of
+   base64.
+
+   The returned value is an upper bound which is only exact for canonical
+   strings.
+
+   @param length The number of input (text) bytes to decode.
+   @return The size of a decoded value in bytes.
+*/
+EXESS_CONST_API
+size_t
+exess_base64_decoded_size(size_t length);
+
+/**
+   Read a binary value from a base64 string.
+
+   Canonical syntax is a multiple of 4 base64 characters, with either 1 or 2
+   trailing "=" characters as necessary, like "Zm9vYg==", with no whitespace.
+   All whitespace is skipped when reading.
+
+   The caller must allocate a large enough buffer to read the value, otherwise
+   an #EXESS_NO_SPACE error will be returned.  The required space can be
+   calculated with exess_base64_decoded_size().
+
+   When this is called, the output blob must have the size of the available
+   buffer in bytes, and a pointer to the buffer.  On return, the size will be
+   set to the exact size of the decoded data, which may be smaller than the
+   initial available size.  Only these first bytes are written, the rest of the
+   buffer is not modified.
+
+   @param out The blob to set to the decoded binary data.
+   @param str String to parse.
+   @return The `count` of characters read, and a `status` code.
+*/
+EXESS_API
+ExessResult
+exess_read_base64(ExessBlob* EXESS_NONNULL out, const char* EXESS_NONNULL str);
+
+/**
+   Write a canonical xsd:base64Binary string.
+
+   The data is always written in canonical form, as a multiple of 4 characters
+   with no whitespace and 1 or 2 trailing "=" characters as padding if
+   necessary.
+
+   @param data_size The size of `data` in bytes.
+   @param data Data to write to a string.
+   @param buf_size The size of `buf` in bytes.
+   @param buf Output buffer, or null to only measure.
+
+   @return The `count` of characters in the output, and `status`
+   #EXESS_SUCCESS, or #EXESS_NO_SPACE if the buffer is too small.
+*/
+EXESS_API
+ExessResult
+exess_write_base64(size_t                    data_size,
+                   const void* EXESS_NONNULL data,
+                   size_t                    buf_size,
+                   char* EXESS_NULLABLE      buf);
+
+/**
+   @}
+   @defgroup exess_hex Hex
+   An xsd:hexBinary is arbitrary binary data in hexadecimal encoding.
+   @{
+*/
+
+/**
+   Return the maximum number of bytes required to decode `length` bytes of hex.
+
+   The returned value is an upper bound which is only exact for canonical
+   strings.
+
+   @param length The number of input (text) bytes to decode.
+   @return The size of a decoded value in bytes.
+*/
+EXESS_CONST_API
+size_t
+exess_hex_decoded_size(size_t length);
+
+/**
+   Read a binary value from a hex string.
+
+   Canonical syntax is an even number of uppercase hexadecimal digits with no
+   whitespace, like "666F6F".  Lowercase hexadecimal is also supported, and all
+   whitespace is skipped when reading.
+
+   The caller must allocate a large enough buffer to read the value, otherwise
+   an #EXESS_NO_SPACE error will be returned.  The required space can be
+   calculated with exess_hex_decoded_size().
+
+   When this is called, the output blob must have the size of the available
+   buffer in bytes, and a pointer to the buffer.  On return, the size will be
+   set to the exact size of the decoded data, which may be smaller than the
+   initial available size.  Only these first bytes are written, the rest of the
+   buffer is not modified.
+
+   @param out The blob to set to the decoded binary data.
+   @param str String to parse.
+   @return The `count` of characters read, and a `status` code.
+*/
+EXESS_API
+ExessResult
+exess_read_hex(ExessBlob* EXESS_NONNULL out, const char* EXESS_NONNULL str);
+
+/**
+   Write a canonical xsd:hexBinary string.
+
+   The data is always written in canonical form, as an even number of uppercase
+   hexadecimal digits with no whitespace.
+
+   @param data_size The size of `data` in bytes.
+   @param data Data to write to a string.
+   @param buf_size The size of `buf` in bytes.
+   @param buf Output buffer, or null to only measure.
+
+   @return The `count` of characters in the output, and `status`
+   #EXESS_SUCCESS, or #EXESS_NO_SPACE if the buffer is too small.
+*/
+EXESS_API
+ExessResult
+exess_write_hex(size_t                    data_size,
+                const void* EXESS_NONNULL data,
+                size_t                    buf_size,
+                char* EXESS_NULLABLE      buf);
+
+/**
+   @}
+   @}
+   @defgroup exess_datatypes Datatypes
+   Runtime integer tags for supported datatypes with conversion to/from URIs.
+   @{
+*/
+
+typedef enum {
+  EXESS_NOTHING,              ///< Sentinel for unknown datatypes or errors
+  EXESS_BOOLEAN,              ///< xsd:boolean (see @ref exess_boolean)
+  EXESS_DECIMAL,              ///< xsd:decimal (see @ref exess_decimal)
+  EXESS_DOUBLE,               ///< xsd:double (see @ref exess_double)
+  EXESS_FLOAT,                ///< xsd:float (see @ref exess_float)
+  EXESS_INTEGER,              ///< xsd:integer (see @ref exess_long)
+  EXESS_NON_POSITIVE_INTEGER, ///< xsd:nonPositiveInteger (see @ref exess_long)
+  EXESS_NEGATIVE_INTEGER,     ///< xsd:negativeInteger (see @ref exess_long)
+  EXESS_LONG,                 ///< xsd:long (see @ref exess_long)
+  EXESS_INT,                  ///< xsd:integer (see @ref exess_int)
+  EXESS_SHORT,                ///< xsd:short (see @ref exess_short)
+  EXESS_BYTE,                 ///< xsd:byte (see @ref exess_byte)
+  EXESS_NON_NEGATIVE_INTEGER, ///< xsd:nonNegativeInteger (see @ref exess_ulong)
+  EXESS_ULONG,                ///< xsd:unsignedLong (see @ref exess_ulong)
+  EXESS_UINT,                 ///< xsd:unsignedInt (see @ref exess_uint)
+  EXESS_USHORT,               ///< xsd:unsignedShort (see @ref exess_ushort)
+  EXESS_UBYTE,                ///< xsd:unsignedByte (see @ref exess_ubyte)
+  EXESS_POSITIVE_INTEGER,     ///< xsd:positiveInteger (see @ref exess_ulong)
+  EXESS_DURATION,             ///< xsd:duration (see @ref exess_duration)
+  EXESS_DATETIME,             ///< xsd:dateTime (see @ref exess_datetime)
+  EXESS_TIME,                 ///< xsd:time (see @ref exess_time)
+  EXESS_DATE,                 ///< xsd:date (see @ref exess_date)
+  EXESS_HEX,                  ///< xsd:hexBinary (see @ref exess_hex)
+  EXESS_BASE64,               ///< xsd:base64Binary (see @ref exess_base64)
+} ExessDatatype;
+
+/**
+   Return the URI for a supported datatype.
+
+   This only returns URIs that start with
+   `http://www.w3.org/2001/XMLSchema#`.
+
+   @param datatype Datatype tag.
+   @return The URI of the datatype, or null for #EXESS_NOTHING.
+*/
+EXESS_CONST_API
+const char* EXESS_NULLABLE
+exess_datatype_uri(ExessDatatype datatype);
+
+/**
+   Return the datatype tag for a datatype URI.
+
+   @return A datatype tag, or #EXESS_NOTHING if the URI is not a supported
+   datatype.
+*/
+EXESS_PURE_API
+ExessDatatype
+exess_datatype_from_uri(const char* EXESS_NONNULL uri);
+
+/**
+   Return whether a datatype has an upper bound on value sizes.
+
+   This returns true for all datatypes except #EXESS_DECIMAL, #EXESS_INTEGER
+   and its half-bounded subtypes #EXESS_NON_POSITIVE_INTEGER,
+   #EXESS_NEGATIVE_INTEGER, #EXESS_NON_NEGATIVE_INTEGER, and
+   #EXESS_POSITIVE_INTEGER, and the binary types #EXESS_HEX and #EXESS_BASE64.
+
+   For bounded datatypes, the maximum length of the string representation is
+   available in `exess_max_lengths` array, or as static constants like
+   #EXESS_MAX_INT_LENGTH.
+
+   @return True if values of the given datatype have a maximum size.
+*/
+EXESS_CONST_API
+bool
+exess_datatype_is_bounded(ExessDatatype datatype);
+
+/**
+   @}
+   @defgroup exess_variant Variant
+   An ExessVariant is a tagged union that can hold any supported datatype.
+   @{
+*/
+
+typedef union {
+  ExessStatus   as_status;
+  ExessBlob     as_blob;
+  bool          as_bool;
+  double        as_double;
+  float         as_float;
+  int64_t       as_long;
+  int32_t       as_int;
+  int16_t       as_short;
+  int8_t        as_byte;
+  uint64_t      as_ulong;
+  uint32_t      as_uint;
+  uint16_t      as_ushort;
+  uint8_t       as_ubyte;
+  ExessDuration as_duration;
+  ExessDateTime as_datetime;
+  ExessTime     as_time;
+  ExessDate     as_date;
+} ExessValue;
+
+/**
+   Any supported value.
+
+   A variant is either nothing, or a value of a specific supported type.  The
+   nothing variant has datatype #EXESS_NOTHING.
+
+   The value fields (everything other than datatype) are stored in an anonymous
+   union, only the field corresponding to the datatype is active.  This should
+   not be used for type punning, use exess_coerce() for that instead.
+*/
+typedef struct {
+  ExessDatatype datatype;
+  ExessValue    value;
+} ExessVariant;
+
+/**
+   @defgroup exess_variant_constructors Constructors
+   @{
+*/
+
+/// Return a nothing (null) variant, with a status code to signal errors
+EXESS_CONST_API
+ExessVariant
+exess_make_nothing(ExessStatus status);
+
+/// Return a boolean variant with the given value
+EXESS_CONST_API
+ExessVariant
+exess_make_boolean(bool value);
+
+/// Return a decimal variant with the given value
+EXESS_CONST_API
+ExessVariant
+exess_make_decimal(double value);
+
+/// Return a double variant with the given value
+EXESS_CONST_API
+ExessVariant
+exess_make_double(double value);
+
+/// Return a float variant with the given value
+EXESS_CONST_API
+ExessVariant
+exess_make_float(float value);
+
+/// Return a long variant with the given value
+EXESS_CONST_API
+ExessVariant
+exess_make_long(int64_t value);
+
+/// Return an int variant with the given value
+EXESS_CONST_API
+ExessVariant
+exess_make_int(int32_t value);
+
+/// Return a short variant with the given value
+EXESS_CONST_API
+ExessVariant
+exess_make_short(int16_t value);
+
+/// Return a byte variant with the given value
+EXESS_CONST_API
+ExessVariant
+exess_make_byte(int8_t value);
+
+/// Return a ulong variant with the given value
+EXESS_CONST_API
+ExessVariant
+exess_make_ulong(uint64_t value);
+
+/// Return a uint variant with the given value
+EXESS_CONST_API
+ExessVariant
+exess_make_uint(uint32_t value);
+
+/// Return a ushort variant with the given value
+EXESS_CONST_API
+ExessVariant
+exess_make_ushort(uint16_t value);
+
+/// Return a ubyte variant with the given value
+EXESS_CONST_API
+ExessVariant
+exess_make_ubyte(uint8_t value);
+
+/// Return a duration variant with the given value
+EXESS_CONST_API
+ExessVariant
+exess_make_duration(ExessDuration value);
+
+/// Return a datetime variant with the given value
+EXESS_CONST_API
+ExessVariant
+exess_make_datetime(ExessDateTime value);
+
+/// Return a time variant with the given value
+EXESS_CONST_API
+ExessVariant
+exess_make_time(ExessTime value);
+
+/// Return a date variant with the given value
+EXESS_CONST_API
+ExessVariant
+exess_make_date(ExessDate value);
+
+/// Return a hex binary variant with the given value
+EXESS_CONST_API
+ExessVariant
+exess_make_hex(ExessBlob value);
+
+/// Return a base64 binary variant with the given value
+EXESS_CONST_API
+ExessVariant
+exess_make_base64(ExessBlob value);
+
+/**
+   @}
+   @defgroup exess_variant_accessors Accessors
+   @{
+*/
+
+/**
+   Return the status of a variant.
+
+   This returns #EXESS_SUCCESS for any valid value, or the stored status for a
+   #EXESS_NOTHING variant.
+*/
+EXESS_PURE_API
+ExessStatus
+exess_get_status(const ExessVariant* EXESS_NONNULL variant);
+
+/// Return a pointer to the value if `variant` is a boolean, otherwise null
+EXESS_PURE_API
+const bool* EXESS_NULLABLE
+exess_get_boolean(const ExessVariant* EXESS_NONNULL variant);
+
+/**
+   Return a pointer to the value if `variant` is a double, otherwise null.
+
+   This will also access the value for #EXESS_DECIMAL.
+*/
+EXESS_PURE_API
+const double* EXESS_NULLABLE
+exess_get_double(const ExessVariant* EXESS_NONNULL variant);
+
+/// Return a pointer to the value if `variant` is a float, otherwise null
+EXESS_PURE_API
+const float* EXESS_NULLABLE
+exess_get_float(const ExessVariant* EXESS_NONNULL variant);
+
+/**
+   Return a pointer to the value if `variant` is a long, otherwise null.
+
+   This will also access the value for #EXESS_INTEGER,
+   #EXESS_NON_POSITIVE_INTEGER, and #EXESS_NEGATIVE_INTEGER.
+*/
+EXESS_PURE_API
+const int64_t* EXESS_NULLABLE
+exess_get_long(const ExessVariant* EXESS_NONNULL variant);
+
+/// Return a pointer to the value if `variant` is an int, otherwise null
+EXESS_PURE_API
+const int32_t* EXESS_NULLABLE
+exess_get_int(const ExessVariant* EXESS_NONNULL variant);
+
+/// Return a pointer to the value if `variant` is a short, otherwise null
+EXESS_PURE_API
+const int16_t* EXESS_NULLABLE
+exess_get_short(const ExessVariant* EXESS_NONNULL variant);
+
+/// Return a pointer to the value if `variant` is a byte, otherwise null
+EXESS_PURE_API
+const int8_t* EXESS_NULLABLE
+exess_get_byte(const ExessVariant* EXESS_NONNULL variant);
+
+/**
+   Return a pointer to the value if `variant` is a ulong, otherwise null.
+
+   This will also access the value for #EXESS_NON_NEGATIVE_INTEGER and
+   #EXESS_POSITIVE_INTEGER.
+*/
+EXESS_PURE_API
+const uint64_t* EXESS_NULLABLE
+exess_get_ulong(const ExessVariant* EXESS_NONNULL variant);
+
+/// Return a pointer to the value if `variant` is a uint, otherwise null
+EXESS_PURE_API
+const uint32_t* EXESS_NULLABLE
+exess_get_uint(const ExessVariant* EXESS_NONNULL variant);
+
+/// Return a pointer to the value if `variant` is a ushort, otherwise null
+EXESS_PURE_API
+const uint16_t* EXESS_NULLABLE
+exess_get_ushort(const ExessVariant* EXESS_NONNULL variant);
+
+/// Return a pointer to the value if `variant` is a ubyte, otherwise null
+EXESS_PURE_API
+const uint8_t* EXESS_NULLABLE
+exess_get_ubyte(const ExessVariant* EXESS_NONNULL variant);
+
+/// Return a pointer to the value if `variant` is a duration, otherwise null
+EXESS_PURE_API
+const ExessDuration* EXESS_NULLABLE
+exess_get_duration(const ExessVariant* EXESS_NONNULL variant);
+
+/// Return a pointer to the value if `variant` is a datetime, otherwise null
+EXESS_PURE_API
+const ExessDateTime* EXESS_NULLABLE
+exess_get_datetime(const ExessVariant* EXESS_NONNULL variant);
+
+/// Return a pointer to the value if `variant` is a time, otherwise null
+EXESS_PURE_API
+const ExessTime* EXESS_NULLABLE
+exess_get_time(const ExessVariant* EXESS_NONNULL variant);
+
+/// Return a pointer to the value if `variant` is a date, otherwise null
+EXESS_PURE_API
+const ExessDate* EXESS_NULLABLE
+exess_get_date(const ExessVariant* EXESS_NONNULL variant);
+
+/// Return a pointer to the value if `variant` is a date, otherwise null
+EXESS_PURE_API
+const ExessBlob* EXESS_NULLABLE
+exess_get_blob(const ExessVariant* EXESS_NONNULL variant);
+
+/**
+   @}
+*/
+
+//
+
+/**
+   Read any supported datatype from a string.
+
+   For reading binary blobs from base64 or hex, the `as_blob` field of `out`
+   must have the size of the available buffer in bytes, and a pointer to the
+   buffer.  On return, the size will be set to the exact size of the decoded
+   data, which may be smaller than the initial available size.  Only these
+   first bytes are written, the rest of the buffer is not modified.
+
+   @param out Set to the parsed value, or nothing on error.
+   @param datatype The datatype to read the string as.
+   @param str String input.
+   @return The `count` of characters read, and a `status` code.
+*/
+EXESS_API
+ExessResult
+exess_read_variant(ExessVariant* EXESS_NONNULL out,
+                   ExessDatatype               datatype,
+                   const char* EXESS_NONNULL   str);
+
+/**
+   Write any supported xsd datatype to a canonical string.
+
+   @param value Value to write.
+   @param buf_size The size of `buf` in bytes.
+   @param buf Output buffer, or null to only measure.
+
+   @return The `count` of characters in the output, and `status`
+   #EXESS_SUCCESS, or #EXESS_NO_SPACE if the buffer is too small.
+*/
+EXESS_API
+ExessResult
+exess_write_variant(ExessVariant         value,
+                    size_t               buf_size,
+                    char* EXESS_NULLABLE buf);
+
+/**
+   Rewrite a supported xsd datatype in canonical form.
+
+   @param value Input value string.
+   @param datatype Datatype of value.
+   @param buf_size The size of `buf` in bytes.
+   @param buf Output buffer, or null to only measure.
+
+   @return The `count` of characters in the output, and a `status` code.  The
+   status may be an error from reading or writing, but the `count` always
+   refers to the number of characters written.
+*/
+EXESS_API
+ExessResult
+exess_write_canonical(const char* EXESS_NONNULL value,
+                      ExessDatatype             datatype,
+                      size_t                    buf_size,
+                      char* EXESS_NULLABLE      buf);
+
+/**
+  @defgroup exess_coercion Datatype coercion
+
+  Values can be converted between some datatypes using exess_coerce().  This is
+  particularly useful for reducing the number of datatypes that the application
+  needs to explicitly handle.
+
+  @{
+*/
+
+/**
+   Coercion flags.
+
+   These values are ORed together to enable different kinds of lossy conversion
+   in exess_coerce().
+*/
+typedef enum {
+  /**
+     Only do lossless datatype coercions.
+
+     A lossless coercion is when the value has been perfectly preserved in the
+     target datatype, and coercing it back will result in the same value.
+
+     For some datatype combinations this will always be the case, for example
+     from short to long.  For others it will depend on the value, for example
+     only the numbers 0 and 1 coerce to boolean without loss.
+  */
+  EXESS_LOSSLESS = 0u,
+
+  /**
+     Allow datatype coercions that reduce the precision of values.
+
+     This allows coercions that are lossy only in terms of precision, so the
+     resulting value is approximately equal to the original value.
+     Specifically, this allows coercing double to float.
+  */
+  EXESS_REDUCE_PRECISION = 1u << 0u,
+
+  /**
+     Allow datatype coercions that round to the nearest integer.
+
+     This allows coercing floating point numbers to integers by rounding to the
+     nearest integer, with halfway cases rounding towards even (the default
+     IEEE-754 rounding order).
+  */
+  EXESS_ROUND = 1u << 1u,
+
+  /**
+     Allow datatype coercions that truncate significant parts of values.
+
+     This allows coercions that lose data beyond simple precision loss.
+     Specifically, this allows coercing any number to boolean, datetime to
+     date, and datetime to time.
+  */
+  EXESS_TRUNCATE = 1u << 2u,
+} ExessCoercionFlag;
+
+/// Bitwise OR of #ExessCoercionFlag values
+typedef uint32_t ExessCoercionFlags;
+
+/**
+   Coerce a value to another datatype if possible.
+
+   @param value Value to coerce.
+
+   @param datatype Datatype to convert to.
+
+   @param coercions Enabled coercion flags.  If this is #EXESS_LOSSLESS (zero),
+   then #EXESS_SUCCESS is only returned if the resulting value can be coerced
+   back to the original type without any loss of data.  Otherwise, the lossy
+   coercions enabled by the set bits will be attempted.
+
+   @return #EXESS_SUCCESS on successful conversion, #EXESS_OUT_OF_RANGE if the
+   value is outside the range of the target type,
+   #EXESS_WOULD_REDUCE_PRECISION, #EXESS_WOULD_ROUND, or #EXESS_WOULD_TRUNCATE
+   if the required coercion is not enabled, or #EXESS_UNSUPPORTED if conversion
+   between the types is not supported at all.
+*/
+EXESS_API
+ExessVariant
+exess_coerce(ExessVariant       value,
+             ExessDatatype      datatype,
+             ExessCoercionFlags coercions);
+
+/**
+   @}
+   @}
+*/
+
+/**
+   The maximum length of the string representation of datatypes.
+
+   For datatypes that do not have such a bound, the value is zero.
+*/
+static const size_t exess_max_lengths[] = {
+  0, // Unknown
+  EXESS_MAX_BOOLEAN_LENGTH,
+  0, // decimal
+  EXESS_MAX_DOUBLE_LENGTH,
+  EXESS_MAX_FLOAT_LENGTH,
+  0, // integer
+  0, // nonPositiveInteger
+  0, // negativeInteger
+  EXESS_MAX_LONG_LENGTH,
+  EXESS_MAX_INT_LENGTH,
+  EXESS_MAX_SHORT_LENGTH,
+  EXESS_MAX_BYTE_LENGTH,
+  0, // nonNegativeInteger
+  EXESS_MAX_ULONG_LENGTH,
+  EXESS_MAX_UINT_LENGTH,
+  EXESS_MAX_USHORT_LENGTH,
+  EXESS_MAX_UBYTE_LENGTH,
+  0, // positiveInteger
+  EXESS_MAX_DURATION_LENGTH,
+  EXESS_MAX_DATETIME_LENGTH,
+  EXESS_MAX_TIME_LENGTH,
+  EXESS_MAX_DATE_LENGTH,
+  0, // hexBinary
+  0, // base64Binary
+};
+
+/**
+   @}
+*/
+
+#ifdef __cplusplus
+} // extern "C"
+#endif
+
+#endif // EXESS_EXESS_H
diff --git a/subprojects/exess/meson.build b/subprojects/exess/meson.build
new file mode 100644
index 00000000..9e2e96b0
--- /dev/null
+++ b/subprojects/exess/meson.build
@@ -0,0 +1,236 @@
+project('exess', ['c'],
+        version: '0.0.1',
+        license: 'ISC',
+        meson_version: '>= 0.49.0',
+        default_options: [
+          'b_ndebug=if-release',
+          'buildtype=release',
+          'c_std=c99',
+          'cpp_std=c++11',
+          'default_library=shared',
+          'warning_level=3',
+        ])
+
+exess_src_root = meson.current_source_dir()
+major_version = meson.project_version().split('.')[0]
+version_suffix = '-@0@'.format(major_version)
+versioned_name = 'exess' + version_suffix
+
+# Load build tools
+pkg = import('pkgconfig')
+cc = meson.get_compiler('c')
+add_languages(['cpp'])
+cpp = meson.get_compiler('cpp')
+
+# Set ultra strict warnings for developers, if requested
+c_warnings = []
+c_suppressions = []
+if get_option('strict')
+  if meson.is_subproject()
+    c_warnings = []
+  else
+    subdir('meson')
+    add_project_arguments(all_c_warnings, language: ['c'])
+  endif
+
+  if cc.get_id() == 'clang'
+    c_suppressions = [
+      '-Wno-bad-function-cast',
+      '-Wno-nullability-extension',
+      '-Wno-padded',
+    ]
+
+  elif cc.get_id() == 'gcc'
+    c_suppressions = [
+      '-Wno-aggregate-return',
+      '-Wno-inline',
+      '-Wno-padded',
+      '-Wno-strict-overflow',
+      '-Wno-switch-default',
+      '-Wno-unsuffixed-float-constants',
+      '-Wno-unused-const-variable',
+    ]
+
+  elif cc.get_id() == 'msvc'
+    c_suppressions = [
+      '/wd4028',  # formal parameter different from declaration
+      '/wd4204',  # nonstandard extension: non-constant aggregate initializer
+      '/wd4706',  # assignment within conditional expression
+      '/wd4710',  # function not inlined
+      '/wd4711',  # function selected for automatic inline expansion
+      '/wd4820',  # padding added after data member
+      '/wd5045',  # will insert Spectre mitigation
+    ]
+  endif
+else
+  if cc.get_id() == 'clang'
+    c_suppressions = [
+      '-Wno-nullability-extension',
+    ]
+  endif
+endif
+
+exess_c_args = cc.get_supported_arguments(c_suppressions)
+
+message(exess_c_args)
+
+if cc.get_id() == 'msvc'
+  # Suppress warnings in system headers
+  add_project_arguments(['/experimental:external',
+                         '/external:W0',
+                         '/external:anglebrackets'],
+                        language: ['c'])
+endif
+
+# Detect compiler features
+
+feature_checks = {
+  'builtin_clz': 'return __builtin_clz(1);',
+  'builtin_clzll': 'return __builtin_clzll(1);',
+}
+
+checks = get_option('checks')
+foreach name, fragment : feature_checks
+  opt = get_option('use_@0@'.format(name))
+  code = 'int main(void) { @0@ }'.format(fragment)
+  define_name = 'HAVE_@0@'.format(name.to_upper())
+
+  if opt.enabled()
+    add_project_arguments(['-D@0@=1'.format(define_name)], language: ['c'])
+  elif opt.disabled()
+    add_project_arguments(['-D@0@=0'.format(define_name)], language: ['c'])
+  elif checks
+    if cc.links(code, name: name)
+      add_project_arguments(['-D@0@=1'.format(define_name)], language: ['c'])
+    else
+      add_project_arguments(['-D@0@=0'.format(define_name)], language: ['c'])
+    endif
+  endif
+endforeach
+
+# System libraries
+m_dep = cc.find_library('m', required: false)
+
+# Determine library type and the flags needed to build it
+if get_option('default_library') == 'both'
+  if host_machine.system() == 'windows'
+    error('default_library=both is not supported on Windows')
+  endif
+
+  library_type = 'both_libraries'
+  library_args = ['-DEXESS_INTERNAL']
+  prog_args = []
+elif get_option('default_library') == 'shared'
+  library_type = 'shared_library'
+  library_args = ['-DEXESS_INTERNAL']
+  prog_args = []
+else
+  library_type = 'static_library'
+  library_args = ['-DEXESS_INTERNAL', '-DEXESS_STATIC']
+  prog_args = ['-DEXESS_STATIC']
+endif
+
+c_headers = ['include/exess/exess.h']
+c_header_files = files(c_headers)
+
+sources = [
+  'src/base64.c',
+  'src/bigint.c',
+  'src/boolean.c',
+  'src/byte.c',
+  'src/canonical.c',
+  'src/coerce.c',
+  'src/datatype.c',
+  'src/date.c',
+  'src/datetime.c',
+  'src/decimal.c',
+  'src/digits.c',
+  'src/double.c',
+  'src/duration.c',
+  'src/float.c',
+  'src/hex.c',
+  'src/int.c',
+  'src/int_math.c',
+  'src/long.c',
+  'src/read_utils.c',
+  'src/scientific.c',
+  'src/short.c',
+  'src/soft_float.c',
+  'src/strerror.c',
+  'src/strtod.c',
+  'src/time.c',
+  'src/timezone.c',
+  'src/ubyte.c',
+  'src/uint.c',
+  'src/ulong.c',
+  'src/ushort.c',
+  'src/variant.c',
+  'src/write_utils.c',
+  'src/year.c',
+]
+
+# Build shared and/or static library/libraries
+libexess = build_target(
+  versioned_name,
+  sources,
+  c_args: exess_c_args + library_args,
+  dependencies: [m_dep],
+  gnu_symbol_visibility: 'hidden',
+  include_directories: include_directories(['include']),
+  install: true,
+  target_type: library_type,
+  version: meson.project_version())
+
+# Generage pkg-config file
+pkg.generate(
+  libexess,
+  name: 'Exess',
+  filebase: versioned_name,
+  subdirs: [versioned_name],
+  version: meson.project_version(),
+  description: 'A simple and efficient regular expression implementation')
+
+# Install header to a versioned include directory
+install_headers(c_headers, subdir: versioned_name / 'exess')
+
+exess_dep = declare_dependency(
+  link_with: libexess,
+  include_directories: include_directories(['include']))
+
+subdir('bindings/cpp')
+
+if not get_option('docs').disabled()
+  subdir('doc')
+endif
+
+if get_option('tests')
+  if library_type == 'both_libraries'
+    libexess_static = libexess.get_static_lib()
+  elif library_type == 'shared_library'
+    libexess_static = static_library(
+      versioned_name,
+      sources,
+      include_directories: include_directories(['include', 'src']),
+      c_args: exess_c_args + library_args,
+      dependencies: [m_dep],
+      gnu_symbol_visibility: 'default')
+  else
+    libexess_static = libexess
+  endif
+
+  exess_static_dep = declare_dependency(
+    include_directories: include_directories(['include']),
+    dependencies: [m_dep],
+    link_with: libexess_static)
+
+  subdir('test')
+endif
+
+if meson.version().version_compare('>=0.53.0')
+  summary('Tests', get_option('tests'), bool_yn: true)
+
+  summary('Install prefix', get_option('prefix'))
+  summary('Headers', get_option('prefix') / get_option('includedir'))
+  summary('Libraries', get_option('prefix') / get_option('libdir'))
+endif
+
diff --git a/subprojects/exess/meson/meson.build b/subprojects/exess/meson/meson.build
new file mode 100644
index 00000000..5dd0de38
--- /dev/null
+++ b/subprojects/exess/meson/meson.build
@@ -0,0 +1,204 @@
+# General code to enable approximately all warnings.
+#
+# This is trivial for clang and MSVC, but GCC does not have such an option, and
+# has several esoteric warnings, so we need to enable everything we want
+# explicitly.  We enable everything that does not require a value argument,
+# except for warnings that are only relevant for very old languages (earlier
+# than C99 or C++11) or non-standard extensions.
+#
+# Omitted common warnings:
+#
+# Wabi=
+# Waggregate-return
+# Walloc-size-larger-than=BYTES
+# Walloca-larger-than=BYTES
+# Wframe-larger-than=BYTES
+# Wlarger-than=BYTES
+# Wstack-usage=BYTES
+# Wsystem-headers
+# Wtraditional
+# Wtraditional-conversion
+# Wtrampolines
+# Wvla-larger-than=BYTES
+#
+# Omitted C warnings:
+#
+# Wc90-c99-compat
+# Wdeclaration-after-statement
+# Wtraditional
+# Wtraditional-conversion
+#
+# Omitted C++ warnings:
+#
+# Wnamespaces
+# Wtemplates
+
+gcc_common_warnings = [
+  '-Walloc-zero',
+  '-Walloca',
+  '-Wanalyzer-too-complex',
+  '-Warith-conversion',
+  '-Warray-bounds=2',
+  '-Wattribute-alias=2',
+  '-Wcast-align=strict',
+  '-Wcast-qual',
+  '-Wconversion',
+  '-Wdate-time',
+  '-Wdisabled-optimization',
+  '-Wdouble-promotion',
+  '-Wduplicated-branches',
+  '-Wduplicated-cond',
+  '-Wfloat-equal',
+  '-Wformat-overflow=2',
+  '-Wformat-signedness',
+  '-Wformat-truncation=2',
+  '-Wformat=2',
+  '-Wimplicit-fallthrough=2',
+  '-Winit-self',
+  '-Winline',
+  '-Winvalid-pch',
+  '-Wlogical-op',
+  '-Wmissing-declarations',
+  '-Wmissing-include-dirs',
+  '-Wmultichar',
+  '-Wnormalized=nfc',
+  '-Wnull-dereference',
+  '-Wpacked',
+  '-Wpadded',
+  '-Wredundant-decls',
+  '-Wscalar-storage-order',
+  '-Wshadow',
+  '-Wshift-overflow=2',
+  '-Wsizeof-array-argument',
+  '-Wstack-protector',
+  '-Wstrict-aliasing=3',
+  '-Wstrict-overflow=5',
+  '-Wstringop-overflow=3',
+  '-Wsuggest-attribute=cold',
+  '-Wsuggest-attribute=const',
+  '-Wsuggest-attribute=format',
+  '-Wsuggest-attribute=malloc',
+  '-Wsuggest-attribute=noreturn',
+  '-Wsuggest-attribute=pure',
+  '-Wswitch-default',
+  '-Wswitch-enum',
+  '-Wsync-nand',
+  '-Wundef',
+  '-Wunused-const-variable=2',
+  '-Wunused-macros',
+  '-Wvarargs',
+  '-Wvector-operation-performance',
+  '-Wvla',
+  '-Wwrite-strings',
+]
+
+gcc_c_warnings = [
+  '-Wbad-function-cast',
+  '-Wc++-compat',
+  '-Wc99-c11-compat',
+  '-Wdesignated-init',
+  '-Wdiscarded-array-qualifiers',
+  '-Wdiscarded-qualifiers',
+  '-Wincompatible-pointer-types',
+  '-Wjump-misses-init',
+  '-Wmissing-prototypes',
+  '-Wnested-externs',
+  '-Wold-style-definition',
+  '-Wstrict-prototypes',
+  '-Wunsuffixed-float-constants',
+]
+
+# Set all_c_warnings for the current C compiler
+if is_variable('cc') and not is_variable('all_c_warnings')
+  if cc.get_id() == 'clang'
+    all_c_warnings = ['-Weverything']
+  elif cc.get_id() == 'gcc'
+    all_c_warnings = gcc_common_warnings + [
+      '-Wbad-function-cast',
+      '-Wc++-compat',
+      '-Wc99-c11-compat',
+      '-Wdesignated-init',
+      '-Wdiscarded-array-qualifiers',
+      '-Wdiscarded-qualifiers',
+      '-Wincompatible-pointer-types',
+      '-Wjump-misses-init',
+      '-Wmissing-prototypes',
+      '-Wnested-externs',
+      '-Wold-style-definition',
+      '-Wstrict-prototypes',
+      '-Wunsuffixed-float-constants',
+    ]
+  elif cc.get_id() == 'msvc'
+    all_c_warnings = ['/Wall']
+  else
+    all_c_warnings = []
+  endif
+
+  all_c_warnings = cc.get_supported_arguments(all_c_warnings)
+  
+endif
+
+# Set all_cpp_warnings for the current C++ compiler
+if is_variable('cpp') and not is_variable('all_cpp_warnings')
+  if cpp.get_id() == 'clang'
+    all_cpp_warnings = [
+      '-Weverything',
+      '-Wno-c++98-compat',
+      '-Wno-c++98-compat-pedantic'
+    ]
+  elif cpp.get_id() == 'gcc'
+    all_cpp_warnings = gcc_common_warnings + [
+      '-Wabi-tag',
+      '-Waligned-new=all',
+      '-Wcatch-value=3',
+      '-Wcomma-subscript',
+      '-Wconditionally-supported',
+      '-Wctor-dtor-privacy',
+      '-Wdeprecated-copy-dtor',
+      '-Weffc++',
+      '-Wextra-semi',
+      '-Wmismatched-tags',
+      '-Wmultiple-inheritance',
+      '-Wnoexcept',
+      '-Wnoexcept-type',
+      '-Wnon-virtual-dtor',
+      '-Wold-style-cast',
+      '-Woverloaded-virtual',
+      '-Wplacement-new=2',
+      '-Wredundant-tags',
+      '-Wregister',
+      '-Wsign-promo',
+      '-Wstrict-null-sentinel',
+      '-Wsuggest-final-methods',
+      '-Wsuggest-final-types',
+      '-Wsuggest-override',
+      '-Wuseless-cast',
+      '-Wvirtual-inheritance',
+      '-Wvolatile',
+      '-Wzero-as-null-pointer-constant',
+    ]
+  elif cpp.get_id() == 'msvc'
+    all_cpp_warnings = ['/Wall']
+  else
+    all_cpp_warnings = []
+  endif
+
+  all_cpp_warnings = cpp.get_supported_arguments(all_cpp_warnings)
+  
+endif
+
+# Set all_objc_warnings for the current Objective C compiler
+if is_variable('objcc') and not is_variable('all_objc_warnings')
+  all_objc_warnings = []
+  if objcc.get_id() == 'clang'
+    all_objc_warnings = ['-Weverything']
+  elif objc.get_id() == 'gcc'
+    all_objc_warnings = gcc_common_warnings + [
+      '-Wno-direct-ivar-access',
+    ]
+  else
+    all_objc_warnings = []
+  endif
+
+  all_objc_warnings = objcc.get_supported_arguments(all_objc_warnings)
+endif
diff --git a/subprojects/exess/meson_options.txt b/subprojects/exess/meson_options.txt
new file mode 100644
index 00000000..2970a1c4
--- /dev/null
+++ b/subprojects/exess/meson_options.txt
@@ -0,0 +1,17 @@
+option('checks', type: 'boolean', value: true, yield: true,
+       description: 'Check for features with the build system')
+
+option('docs', type: 'feature', value: 'auto', yield: true,
+       description: 'Build documentation')
+
+option('strict', type: 'boolean', value: false, yield: true,
+       description: 'Enable ultra-strict warnings')
+
+option('tests', type: 'boolean', value: true, yield: true,
+       description: 'Build tests')
+
+option('use_builtin_clz', type: 'feature', value: 'auto', yield: true,
+       description: 'Use __builtin_clz')
+       
+option('use_builtin_clzll', type: 'feature', value: 'auto', yield: true,
+       description: 'Use __builtin_clzll')
diff --git a/subprojects/exess/scripts/dox_to_sphinx.py b/subprojects/exess/scripts/dox_to_sphinx.py
new file mode 100755
index 00000000..557d0474
--- /dev/null
+++ b/subprojects/exess/scripts/dox_to_sphinx.py
@@ -0,0 +1,675 @@
+#!/usr/bin/env python3
+
+# Copyright 2020 David Robillard <d@drobilla.net>
+#
+# Permission to use, copy, modify, and/or distribute this software for any
+# purpose with or without fee is hereby granted, provided that the above
+# copyright notice and this permission notice appear in all copies.
+#
+# THIS SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+# WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+# MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+# ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+# WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+# ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+# OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+
+"""
+Write Sphinx markup from Doxygen XML.
+
+Takes a path to a directory of XML generated by Doxygen, and emits a directory
+with a reStructuredText file for every documented symbol.
+"""
+
+import argparse
+import os
+import sys
+import textwrap
+import xml.etree.ElementTree
+
+__author__ = "David Robillard"
+__date__ = "2020-11-18"
+__email__ = "d@drobilla.net"
+__license__ = "ISC"
+__version__ = __date__.replace("-", ".")
+
+
+def load_index(index_path):
+    """
+    Load the index from XML.
+
+    :returns: A dictionary from ID to skeleton records with basic information
+    for every documented entity.  Some records have an ``xml_filename`` key
+    with the filename of a definition file.  These files will be loaded later
+    to flesh out the records in the index.
+    """
+
+    root = xml.etree.ElementTree.parse(index_path).getroot()
+    index = {}
+
+    for compound in root:
+        compound_id = compound.get("refid")
+        compound_kind = compound.get("kind")
+        compound_name = compound.find("name").text
+        if compound_kind in ["dir", "file", "page"]:
+            continue
+
+        # Add record for compound (compounds appear only once in the index)
+        assert compound_id not in index
+        index[compound_id] = {
+            "kind": compound_kind,
+            "name": compound_name,
+            "xml_filename": compound_id + ".xml",
+            "children": [],
+        }
+
+        name_prefix = (
+            ("%s::" % compound_name) if compound_kind == "namespace" else ""
+        )
+
+        for child in compound.findall("member"):
+            if child.get("refid") in index:
+                assert compound_kind == "group"
+                continue
+
+            # Everything has a kind and a name
+            child_record = {
+                "kind": child.get("kind"),
+                "name": name_prefix + child.find("name").text,
+            }
+
+            if child.get("kind") == "enum":
+                # Enums are not compounds, but we want to resolve the parent of
+                # their values so they are not written as top level documents
+                child_record["children"] = []
+
+            if child.get("kind") == "enumvalue":
+                # Remove namespace prefix
+                child_record["name"] = child.find("name").text
+
+            if child.get("kind") == "variable":
+                if child_record["name"][0] == "@":
+                    # Remove placeholder name from anonymous struct or union
+                    child_record["name"] = ""
+                else:
+                    # Remove namespace prefix
+                    child_record["name"] = child.find("name").text
+
+            index[child.get("refid")] = child_record
+
+    return index
+
+
+def resolve_index(index, root):
+    """
+    Walk a definition document and extend the index for linking.
+
+    This does two things: sets the "parent" and "children" fields of all
+    applicable records, and sets the "strong" field of enums so that the
+    correct Sphinx role can be used when referring to them.
+    """
+
+    def add_child(index, parent_id, child_id):
+        parent = index[parent_id]
+        child = index[child_id]
+
+        if child["kind"] == "enumvalue":
+            assert parent["kind"] == "enum"
+            assert "parent" not in child or child["parent"] == parent_id
+            child["parent"] = parent_id
+
+        else:
+            if parent["kind"] in ["class", "struct", "union"]:
+                assert "parent" not in child or child["parent"] == parent_id
+                child["parent"] = parent_id
+
+        if child_id not in parent["children"]:
+            parent["children"] += [child_id]
+
+    compound = root.find("compounddef")
+    compound_kind = compound.get("kind")
+
+    if compound_kind == "group":
+        for subgroup in compound.findall("innergroup"):
+            add_child(index, compound.get("id"), subgroup.get("refid"))
+
+        for klass in compound.findall("innerclass"):
+            add_child(index, compound.get("id"), klass.get("refid"))
+
+    for section in compound.findall("sectiondef"):
+        if section.get("kind").startswith("private"):
+            for member in section.findall("memberdef"):
+                if member.get("id") in index:
+                    del index[member.get("id")]
+        else:
+            for member in section.findall("memberdef"):
+                member_id = member.get("id")
+                add_child(index, compound.get("id"), member_id)
+
+                if member.get("kind") == "enum":
+                    index[member_id]["strong"] = member.get("strong") == "yes"
+                    for value in member.findall("enumvalue"):
+                        add_child(index, member_id, value.get("id"))
+
+
+def sphinx_role(record, lang):
+    """
+    Return the Sphinx role used for a record.
+
+    This is used for the description directive like ".. c:function::", and
+    links like ":c:func:`foo`.
+    """
+
+    kind = record["kind"]
+
+    if kind in ["class", "function", "namespace", "struct", "union"]:
+        return lang + ":" + kind
+
+    if kind == "define":
+        return "c:macro"
+
+    if kind == "group":
+        return "ref"
+
+    if kind == "enum":
+        return lang + (":enum-class" if record["strong"] else ":enum")
+
+    if kind == "typedef":
+        return lang + ":type"
+
+    if kind == "enumvalue":
+        return lang + ":enumerator"
+
+    if kind == "variable":
+        return lang + (":member" if "parent" in record else ":var")
+
+    raise RuntimeError("No known role for kind '%s'" % kind)
+
+
+def child_identifier(lang, parent_name, child_name):
+    """
+    Return the identifier for an enum value or struct member.
+
+    Sphinx, for some reason, uses a different syntax for this in C and C++.
+    """
+
+    separator = "::" if lang == "cpp" else "."
+
+    return "%s%s%s" % (parent_name, separator, child_name)
+
+
+def link_markup(index, lang, refid):
+    """Return a Sphinx link for a Doxygen reference."""
+
+    record = index[refid]
+    kind, name = record["kind"], record["name"]
+    role = sphinx_role(record, lang)
+
+    if kind in ["class", "define", "enum", "struct", "typedef", "union"]:
+        return ":%s:`%s`" % (role, name)
+
+    if kind == "group":
+        return ":ref:`%s`" % name
+
+    if kind == "function":
+        return ":%s:func:`%s`" % (lang, name)
+
+    if kind == "enumvalue":
+        parent_name = index[record["parent"]]["name"]
+        return ":%s:`%s`" % (role, child_identifier(lang, parent_name, name))
+
+    if kind == "variable":
+        if "parent" not in record:
+            return ":%s:var:`%s`" % (lang, name)
+
+        parent_name = index[record["parent"]]["name"]
+        return ":%s:`%s`" % (role, child_identifier(lang, parent_name, name))
+
+    raise RuntimeError("Unknown link target kind: %s" % kind)
+
+
+def indent(markup, depth):
+    """
+    Indent markup to a depth level.
+
+    Like textwrap.indent() but takes an integer and works in reST indentation
+    levels for clarity."
+    """
+
+    return textwrap.indent(markup, "   " * depth)
+
+
+def heading(text, level):
+    """
+    Return a ReST heading at a given level.
+
+    Follows the style in the Python documentation guide, see
+    <https://devguide.python.org/documenting/#sections>.
+    """
+
+    assert 1 <= level <= 6
+
+    chars = ("#", "*", "=", "-", "^", '"')
+    line = chars[level] * len(text)
+
+    return "%s%s\n%s\n\n" % (line + "\n" if level < 3 else "", text, line)
+
+
+def dox_to_rst(index, lang, node):
+    """
+    Convert documentation commands (docCmdGroup) to Sphinx markup.
+
+    This is used to convert the content of descriptions in the documentation.
+    It recursively parses all children tags and raises a RuntimeError if any
+    unknown tag is encountered.
+    """
+
+    def field_value(markup):
+        """Return a value for a field as a single line or indented block."""
+        if "\n" in markup.strip():
+            return "\n" + indent(markup, 1)
+
+        return " " + markup.strip()
+
+    if node.tag == "lsquo":
+        return "‘"
+
+    if node.tag == "rsquo":
+        return "’"
+
+    if node.tag == "computeroutput":
+        return "``%s``" % plain_text(node)
+
+    if node.tag == "itemizedlist":
+        markup = ""
+        for item in node.findall("listitem"):
+            assert len(item) == 1
+            markup += "\n- %s" % dox_to_rst(index, lang, item[0])
+
+        return markup
+
+    if node.tag == "para":
+        markup = node.text if node.text is not None else ""
+        for child in node:
+            markup += dox_to_rst(index, lang, child)
+            markup += child.tail if child.tail is not None else ""
+
+        return markup.strip() + "\n\n"
+
+    if node.tag == "parameterlist":
+        markup = ""
+        for item in node.findall("parameteritem"):
+            name = item.find("parameternamelist/parametername")
+            description = item.find("parameterdescription")
+            assert len(description) == 1
+            markup += "\n\n:param %s:%s" % (
+                name.text,
+                field_value(dox_to_rst(index, lang, description[0])),
+            )
+
+        return markup + "\n"
+
+    if node.tag == "programlisting":
+        return "\n.. code-block:: %s\n\n%s" % (
+            lang,
+            indent(plain_text(node), 1),
+        )
+
+    if node.tag == "ref":
+        refid = node.get("refid")
+        if refid not in index:
+            sys.stderr.write("warning: Unresolved link: %s\n" % refid)
+            return node.text
+
+        assert len(node) == 0
+        assert len(link_markup(index, lang, refid)) > 0
+        return link_markup(index, lang, refid)
+
+    if node.tag == "simplesect":
+        assert len(node) == 1
+
+        if node.get("kind") == "return":
+            return "\n:returns:" + field_value(
+                dox_to_rst(index, lang, node[0])
+            )
+
+        if node.get("kind") == "see":
+            return dox_to_rst(index, lang, node[0])
+
+        raise RuntimeError("Unknown simplesect kind: %s" % node.get("kind"))
+
+    if node.tag == "ulink":
+        return "`%s <%s>`_" % (node.text, node.get("url"))
+
+    raise RuntimeError("Unknown documentation command: %s" % node.tag)
+
+
+def description_markup(index, lang, node):
+    """Return the markup for a brief or detailed description."""
+
+    assert node.tag == "briefdescription" or node.tag == "detaileddescription"
+    assert not (node.tag == "briefdescription" and len(node) > 1)
+    assert len(node.text.strip()) == 0
+
+    return "".join([dox_to_rst(index, lang, child) for child in node]).strip()
+
+
+def set_descriptions(index, lang, definition, record):
+    """Set a record's brief/detailed descriptions from the XML definition."""
+
+    for tag in ["briefdescription", "detaileddescription"]:
+        node = definition.find(tag)
+        if node is not None:
+            record[tag] = description_markup(index, lang, node)
+
+
+def set_template_params(node, record):
+    """Set a record's template_params from the XML definition."""
+
+    template_param_list = node.find("templateparamlist")
+    if template_param_list is not None:
+        params = []
+        for param in template_param_list.findall("param"):
+            if param.find("declname") is not None:
+                # Value parameter
+                type_text = plain_text(param.find("type"))
+                name_text = plain_text(param.find("declname"))
+
+                params += ["%s %s" % (type_text, name_text)]
+            else:
+                # Type parameter
+                params += ["%s" % (plain_text(param.find("type")))]
+
+        record["template_params"] = "%s" % ", ".join(params)
+
+
+def plain_text(node):
+    """
+    Return the plain text of a node with all tags ignored.
+
+    This is needed where Doxygen may include refs but Sphinx needs plain text
+    because it parses things itself to generate links.
+    """
+
+    if node.tag == "sp":
+        markup = " "
+    elif node.text is not None:
+        markup = node.text
+    else:
+        markup = ""
+
+    for child in node:
+        markup += plain_text(child)
+        markup += child.tail if child.tail is not None else ""
+
+    return markup
+
+
+def local_name(name):
+    """Return a name with all namespace prefixes stripped."""
+
+    return name[name.rindex("::") + 2 :] if "::" in name else name
+
+
+def read_definition_doc(index, lang, root):
+    """Walk a definition document and update described records in the index."""
+
+    # Set descriptions for the compound itself
+    compound = root.find("compounddef")
+    compound_record = index[compound.get("id")]
+    set_descriptions(index, lang, compound, compound_record)
+    set_template_params(compound, compound_record)
+
+    if compound.find("title") is not None:
+        compound_record["title"] = compound.find("title").text.strip()
+
+    # Set documentation for all children
+    for section in compound.findall("sectiondef"):
+        if section.get("kind").startswith("private"):
+            continue
+
+        for member in section.findall("memberdef"):
+            kind = member.get("kind")
+            record = index[member.get("id")]
+            set_descriptions(index, lang, member, record)
+            set_template_params(member, record)
+
+            if compound.get("kind") in ["class", "struct", "union"]:
+                assert kind in ["function", "typedef", "variable"]
+                record["type"] = plain_text(member.find("type"))
+
+            if kind == "enum":
+                for value in member.findall("enumvalue"):
+                    set_descriptions(
+                        index, lang, value, index[value.get("id")]
+                    )
+
+            elif kind == "function":
+                record["prototype"] = "%s %s%s" % (
+                    plain_text(member.find("type")),
+                    member.find("name").text,
+                    member.find("argsstring").text,
+                )
+
+            elif kind == "typedef":
+                name = local_name(record["name"])
+                args_text = member.find("argsstring").text
+                target_text = plain_text(member.find("type"))
+                if args_text is not None:  # Function pointer
+                    assert target_text[-2:] == "(*" and args_text[0] == ")"
+                    record["type"] = target_text + args_text
+                    record["definition"] = target_text + name + args_text
+                else:  # Normal named typedef
+                    assert target_text is not None
+                    record["type"] = target_text
+                    if member.find("definition").text.startswith("using"):
+                        record["definition"] = "%s = %s" % (
+                            name,
+                            target_text,
+                        )
+                    else:
+                        record["definition"] = "%s %s" % (
+                            target_text,
+                            name,
+                        )
+
+            elif kind == "variable":
+                record["type"] = plain_text(member.find("type"))
+                record["name"] = plain_text(member.find("name"))
+                record["definition"] = plain_text(member.find("definition"))
+
+            elif kind == "define":
+                record["initializer"] = member.find("initializer").text
+
+
+def declaration_string(record):
+    """
+    Return the string that describes a declaration.
+
+    This is what follows the directive, and is in C/C++ syntax, except without
+    keywords like "typedef" and "using" as expected by Sphinx.  For example,
+    "struct ThingImpl Thing" or "void run(int value)".
+    """
+
+    kind = record["kind"]
+    result = ""
+
+    if "template_params" in record:
+        result = "template <%s> " % record["template_params"]
+
+    if kind == "function":
+        result += record["prototype"]
+    elif kind == "typedef":
+        result += record["definition"]
+    elif kind == "variable":
+        if "type" in record and "name" in record:
+            result += "%s %s" % (record["type"], local_name(record["name"]))
+        else:
+            result += record["definition"]
+    elif "type" in record:
+        result += "%s %s" % (record["type"], local_name(record["name"]))
+    else:
+        result += local_name(record["name"])
+
+    return result
+
+
+def document_markup(index, lang, record):
+    """Return the complete document that describes some documented entity."""
+
+    kind = record["kind"]
+    role = sphinx_role(record, lang)
+    name = record["name"]
+    markup = ""
+
+    if name != local_name(name):
+        markup += ".. cpp:namespace:: %s\n\n" % name[0 : name.rindex("::")]
+
+    # Write top-level directive
+    markup += ".. %s:: %s\n" % (role, declaration_string(record))
+
+    # Write main description blurb
+    markup += "\n" + indent(record["briefdescription"] + "\n", 1)
+    if len(record["detaileddescription"]) > 0:
+        markup += "\n" + indent(record["detaileddescription"], 1) + "\n"
+
+    assert (
+        kind in ["class", "enum", "namespace", "struct", "union"]
+        or "children" not in record
+    )
+
+    # Sphinx C++ namespaces work by setting a scope, they have no content
+    child_indent = 0 if kind == "namespace" else 1
+
+    # Write inline children if applicable
+    markup += "\n" if "children" in record else ""
+    for child_id in record.get("children", []):
+        child_record = index[child_id]
+        child_role = sphinx_role(child_record, lang)
+
+        if not child_record["name"]:
+            continue  # Skip anonymous union member
+
+        child_header = ".. %s:: %s\n\n" % (
+            child_role,
+            declaration_string(child_record),
+        )
+
+        markup += "\n"
+        markup += indent(child_header, child_indent)
+        markup += indent(child_record["briefdescription"], child_indent + 1)
+        markup += indent(child_record["detaileddescription"], child_indent + 1)
+
+    return markup
+
+
+def symbol_filename(name):
+    """Adapt the name of a symbol to be suitable for use as a filename."""
+
+    return name.replace("::", "__")
+
+
+def emit_groups(index, lang, output_dir, force):
+    """Write a description file for every group documented in the index."""
+
+    for record in index.values():
+        if record["kind"] != "group":
+            continue
+
+        name = record["name"]
+        filename = os.path.join(output_dir, "%s.rst" % name)
+        if not force and os.path.exists(filename):
+            raise FileExistsError("File already exists: '%s'" % filename)
+
+        with open(filename, "w") as rst:
+            rst.write(".. _%s:\n\n" % name)
+            rst.write(heading(record["title"], 1))
+
+            # Get all child group and symbol names
+            child_groups = {}
+            child_symbols = {}
+            for child_id in record["children"]:
+                child = index[child_id]
+                assert child["name"][0] != "@"
+                if child["kind"] == "group":
+                    child_groups[child["name"]] = child
+                else:
+                    child_symbols[child["name"]] = child
+
+            # Emit description (document body)
+            if len(record["briefdescription"]) > 0:
+                rst.write(record["briefdescription"] + "\n\n")
+            if len(record["detaileddescription"]) > 0:
+                rst.write(record["detaileddescription"] + "\n\n")
+
+            if len(child_groups) > 0:
+                # Emit TOC for child groups
+                rst.write(".. toctree::\n\n")
+                for name, group in child_groups.items():
+                    rst.write(indent(group["name"], 1) + "\n")
+
+            # Emit symbols in sorted order
+            for name, symbol in child_symbols.items():
+                rst.write("\n")
+                rst.write(document_markup(index, lang, symbol))
+                rst.write("\n")
+
+
+def run(index_xml_path, output_dir, language, force):
+    """Write a directory of Sphinx files from a Doxygen XML directory."""
+
+    # Build skeleton index from index.xml
+    xml_dir = os.path.dirname(index_xml_path)
+    index = load_index(index_xml_path)
+
+    # Load all definition documents
+    definition_docs = []
+    for record in index.values():
+        if "xml_filename" in record:
+            xml_path = os.path.join(xml_dir, record["xml_filename"])
+            definition_docs += [xml.etree.ElementTree.parse(xml_path)]
+
+    # Do an initial pass of the definition documents to resolve the index
+    for root in definition_docs:
+        resolve_index(index, root)
+
+    # Finally read the documentation from definition documents
+    for root in definition_docs:
+        read_definition_doc(index, language, root)
+
+    # Create output directory
+    try:
+        os.makedirs(output_dir)
+    except OSError:
+        pass
+
+    # Emit output files
+    emit_groups(index, language, output_dir, force)
+
+
+if __name__ == "__main__":
+    ap = argparse.ArgumentParser(
+        usage="%(prog)s [OPTION]... XML_DIR OUTPUT_DIR",
+        description=__doc__,
+        formatter_class=argparse.RawDescriptionHelpFormatter,
+    )
+
+    ap.add_argument(
+        "-f",
+        "--force",
+        action="store_true",
+        help="overwrite files",
+    )
+
+    ap.add_argument(
+        "-l",
+        "--language",
+        default="c",
+        choices=["c", "cpp"],
+        help="language domain for output",
+    )
+
+    ap.add_argument("index_xml_path", help="path index.xml from Doxygen")
+    ap.add_argument("output_dir", help="output directory")
+
+    run(**vars(ap.parse_args(sys.argv[1:])))
diff --git a/subprojects/exess/src/.clang-tidy b/subprojects/exess/src/.clang-tidy
new file mode 100644
index 00000000..8023398e
--- /dev/null
+++ b/subprojects/exess/src/.clang-tidy
@@ -0,0 +1,12 @@
+Checks: >
+  *,
+  -*-magic-numbers,
+  -*-uppercase-literal-suffix,
+  -clang-analyzer-security.insecureAPI.DeprecatedOrUnsafeBufferHandling,
+  -hicpp-multiway-paths-covered,
+  -llvm-header-guard,
+  -llvmlibc-*,
+  -misc-no-recursion,
+WarningsAsErrors: '*'
+HeaderFilterRegex: '.*'
+FormatStyle: file
diff --git a/subprojects/exess/src/attributes.h b/subprojects/exess/src/attributes.h
new file mode 100644
index 00000000..1575113b
--- /dev/null
+++ b/subprojects/exess/src/attributes.h
@@ -0,0 +1,30 @@
+/*
+  Copyright 2019-2021 David Robillard <d@drobilla.net>
+
+  Permission to use, copy, modify, and/or distribute this software for any
+  purpose with or without fee is hereby granted, provided that the above
+  copyright notice and this permission notice appear in all copies.
+
+  THIS SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+  WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+  MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+  ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+  WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+  ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+  OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+*/
+
+#ifndef EXESS_ATTRIBUTES_H
+#define EXESS_ATTRIBUTES_H
+
+#ifdef __GNUC__
+#  define EXESS_I_PURE_FUNC __attribute__((pure))
+#  define EXESS_I_CONST_FUNC __attribute__((const))
+#  define EXESS_I_MALLOC_FUNC __attribute__((malloc))
+#else
+#  define EXESS_I_PURE_FUNC
+#  define EXESS_I_CONST_FUNC
+#  define EXESS_I_MALLOC_FUNC
+#endif
+
+#endif // EXESS_ATTRIBUTES_H
diff --git a/subprojects/exess/src/base64.c b/subprojects/exess/src/base64.c
new file mode 100644
index 00000000..de64ee68
--- /dev/null
+++ b/subprojects/exess/src/base64.c
@@ -0,0 +1,163 @@
+/*
+  Copyright 2011-2021 David Robillard <d@drobilla.net>
+
+  Permission to use, copy, modify, and/or distribute this software for any
+  purpose with or without fee is hereby granted, provided that the above
+  copyright notice and this permission notice appear in all copies.
+
+  THIS SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+  WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+  MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+  ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+  WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+  ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+  OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+*/
+
+#include "macros.h"
+#include "read_utils.h"
+#include "string_utils.h"
+#include "write_utils.h"
+
+#include "exess/exess.h"
+
+#include <assert.h>
+#include <stdint.h>
+#include <string.h>
+
+// Map a 6-bit base64 group to a base64 digit
+static inline uint8_t
+map(const unsigned group)
+{
+  assert(group < 64);
+
+  // See <http://tools.ietf.org/html/rfc3548#section-3>.
+  static const uint8_t b64_map[] =
+    "ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz0123456789+/";
+
+  return b64_map[group];
+}
+
+// Unmap a base64 digit to the numeric value used for decoding
+static inline uint8_t
+unmap(const uint8_t in)
+{
+  /* Table indexed by encoded characters that contains the numeric value used
+     for decoding, shifted up by 47 to be in the range of printable ASCII.  A
+     '$' is a placeholder for characters not in the base64 alphabet. */
+  static const uint8_t b64_unmap[] =
+    "$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$m$$$ncdefghijkl$$$$$$"
+    "$/0123456789:;<=>?@ABCDEFGH$$$$$$IJKLMNOPQRSTUVWXYZ[\\]^_`ab$$$$"
+    "$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$"
+    "$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$";
+
+  return (uint8_t)(b64_unmap[in] - 47u);
+}
+
+static char
+next_char(const char* const str, size_t* const i)
+{
+  *i += skip_whitespace(str + *i);
+
+  return str[*i];
+}
+
+size_t
+exess_base64_decoded_size(const size_t length)
+{
+  return (length * 3) / 4 + 2;
+}
+
+ExessResult
+exess_read_base64(ExessBlob* const out, const char* const str)
+{
+  uint8_t* const       uout = (uint8_t*)out->data;
+  const uint8_t* const ustr = (const uint8_t*)str;
+  size_t               i    = 0u;
+  size_t               o    = 0u;
+
+  while (str[i]) {
+    // Skip leading whitespace
+    i += skip_whitespace(str + i);
+    if (!str[i]) {
+      break;
+    }
+
+    // Read next chunk of 4 input characters
+    uint8_t in[] = "====";
+    for (size_t j = 0; j < 4; ++j) {
+      const char c = next_char(str, &i);
+      if (!is_base64(c)) {
+        return result(EXESS_EXPECTED_BASE64, i);
+      }
+
+      in[j] = ustr[i++];
+    }
+
+    if (in[0] == '=' || in[1] == '=' || (in[2] == '=' && in[3] != '=')) {
+      return result(EXESS_BAD_VALUE, i);
+    }
+
+    const size_t n_bytes = 1u + (in[2] != '=') + (in[3] != '=');
+    if (o + n_bytes > out->size) {
+      return result(EXESS_NO_SPACE, i);
+    }
+
+    const uint8_t a1 = (uint8_t)(unmap(in[0]) << 2u);
+    const uint8_t a2 = unmap(in[1]) >> 4u;
+
+    uout[o++] = a1 | a2;
+
+    if (in[2] != '=') {
+      const uint8_t b1 = (uint8_t)(unmap(in[1]) << 4u) & 0xF0u;
+      const uint8_t b2 = unmap(in[2]) >> 2u;
+
+      uout[o++] = b1 | b2;
+    }
+
+    if (in[3] != '=') {
+      const uint8_t c1 = (uint8_t)(unmap(in[2]) << 6u) & 0xC0u;
+      const uint8_t c2 = unmap(in[3]);
+
+      uout[o++] = c1 | c2;
+    }
+  }
+
+  out->size = o;
+  return result(EXESS_SUCCESS, i);
+}
+
+ExessResult
+exess_write_base64(const size_t      data_size,
+                   const void* const data,
+                   const size_t      buf_size,
+                   char* const       buf)
+{
+  const size_t length = (data_size + 2) / 3 * 4;
+  if (!buf) {
+    return result(EXESS_SUCCESS, length);
+  }
+
+  if (buf_size < length + 1) {
+    return result(EXESS_NO_SPACE, 0);
+  }
+
+  uint8_t* const out = (uint8_t*)buf;
+
+  size_t o = 0;
+  for (size_t i = 0; i < data_size; i += 3, o += 4) {
+    uint8_t      in[4] = {0, 0, 0, 0};
+    const size_t n_in  = MIN(3, data_size - i);
+    memcpy(in, (const uint8_t*)data + i, n_in);
+
+    out[o]     = map(in[0] >> 2u);
+    out[o + 1] = map(((in[0] & 0x03u) << 4u) | ((in[1] & 0xF0u) >> 4u));
+    out[o + 2] =
+      ((n_in > 1u) ? map(((in[1] & 0x0Fu) << 2u) | ((in[2] & 0xC0u) >> 6u))
+                   : '=');
+
+    out[o + 3] = ((n_in > 2u) ? map(in[2] & 0x3Fu) : '=');
+  }
+
+  return end_write(EXESS_SUCCESS, buf_size, buf, o);
+}
diff --git a/subprojects/exess/src/bigint.c b/subprojects/exess/src/bigint.c
new file mode 100644
index 00000000..8c8a95c8
--- /dev/null
+++ b/subprojects/exess/src/bigint.c
@@ -0,0 +1,605 @@
+/*
+  Copyright 2019-2021 David Robillard <d@drobilla.net>
+
+  Permission to use, copy, modify, and/or distribute this software for any
+  purpose with or without fee is hereby granted, provided that the above
+  copyright notice and this permission notice appear in all copies.
+
+  THIS SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+  WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+  MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+  ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+  WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+  ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+  OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+*/
+
+#include "bigint.h"
+#include "macros.h"
+
+#include "int_math.h"
+
+#include <assert.h>
+#include <stdbool.h>
+#include <stdlib.h>
+#include <string.h>
+
+typedef uint64_t Hugit;
+
+static const uint32_t bigit_mask = ~(uint32_t)0;
+static const uint64_t carry_mask = (uint64_t) ~(uint32_t)0 << 32u;
+
+typedef struct {
+  unsigned bigits;
+  unsigned bits;
+} Offset;
+
+static inline Offset
+make_offset(const unsigned i)
+{
+  const unsigned bigits = i / BIGINT_BIGIT_BITS;
+  const unsigned bits   = i - bigits * BIGINT_BIGIT_BITS;
+
+  const Offset offset = {bigits, bits};
+  return offset;
+}
+
+#ifndef NDEBUG
+static inline bool
+exess_bigint_is_clamped(const ExessBigint* num)
+{
+  return num->n_bigits == 0 || num->bigits[num->n_bigits - 1];
+}
+#endif
+
+void
+exess_bigint_shift_left(ExessBigint* num, const unsigned amount)
+{
+  assert(exess_bigint_is_clamped(num));
+  if (amount == 0 || num->n_bigits == 0) {
+    return;
+  }
+
+  const Offset offset = make_offset(amount);
+
+  assert(num->n_bigits + offset.bigits < BIGINT_MAX_BIGITS);
+  num->n_bigits += offset.bigits + (bool)offset.bits;
+
+  if (offset.bits == 0) { // Simple bigit-aligned shift
+    for (unsigned i = num->n_bigits - 1; i >= offset.bigits; --i) {
+      num->bigits[i] = num->bigits[i - offset.bigits];
+    }
+  } else { // Bigit + sub-bigit bit offset shift
+    const unsigned right_shift = BIGINT_BIGIT_BITS - offset.bits;
+    for (unsigned i = num->n_bigits - offset.bigits - 1; i > 0; --i) {
+      num->bigits[i + offset.bigits] =
+        (num->bigits[i] << offset.bits) | (num->bigits[i - 1] >> right_shift);
+    }
+
+    num->bigits[offset.bigits] = num->bigits[0] << offset.bits;
+  }
+
+  // Zero LSBs
+  for (unsigned i = 0; i < offset.bigits; ++i) {
+    num->bigits[i] = 0;
+  }
+
+  exess_bigint_clamp(num);
+  assert(exess_bigint_is_clamped(num));
+}
+
+void
+exess_bigint_zero(ExessBigint* num)
+{
+  static const ExessBigint zero = {{0}, 0};
+
+  *num = zero;
+}
+
+void
+exess_bigint_set(ExessBigint* num, const ExessBigint* value)
+{
+  *num = *value;
+}
+
+void
+exess_bigint_set_u32(ExessBigint* num, const uint32_t value)
+{
+  exess_bigint_zero(num);
+
+  num->bigits[0] = value;
+  num->n_bigits  = (bool)value;
+}
+
+void
+exess_bigint_clamp(ExessBigint* num)
+{
+  while (num->n_bigits > 0 && num->bigits[num->n_bigits - 1] == 0) {
+    --num->n_bigits;
+  }
+}
+
+void
+exess_bigint_set_u64(ExessBigint* num, const uint64_t value)
+{
+  exess_bigint_zero(num);
+
+  num->bigits[0] = (Bigit)(value & bigit_mask);
+  num->bigits[1] = (Bigit)(value >> BIGINT_BIGIT_BITS);
+  num->n_bigits  = num->bigits[1] ? 2u : num->bigits[0] ? 1u : 0u;
+}
+
+void
+exess_bigint_set_pow10(ExessBigint* num, const unsigned exponent)
+{
+  exess_bigint_set_u32(num, 1);
+  exess_bigint_multiply_pow10(num, exponent);
+}
+
+static uint32_t
+read_u32(const char* const str, uint32_t* result, uint32_t* n_digits)
+{
+  static const size_t uint32_digits10 = 9;
+
+  *result = *n_digits = 0;
+
+  uint32_t i = 0;
+  for (; str[i] && *n_digits < uint32_digits10; ++i) {
+    if (str[i] >= '0' && str[i] <= '9') {
+      *result = *result * 10u + (unsigned)(str[i] - '0');
+      *n_digits += 1;
+    } else if (str[i] != '.') {
+      break;
+    }
+  }
+
+  return i;
+}
+
+void
+exess_bigint_set_decimal_string(ExessBigint* num, const char* const str)
+{
+  exess_bigint_zero(num);
+
+  uint32_t pos      = 0;
+  uint32_t n_digits = 0;
+  uint32_t n_read   = 0;
+  uint32_t word     = 0;
+  while ((n_read = read_u32(str + pos, &word, &n_digits))) {
+    exess_bigint_multiply_u32(num, (uint32_t)POW10[n_digits]);
+    exess_bigint_add_u32(num, word);
+    pos += n_read;
+  }
+
+  exess_bigint_clamp(num);
+}
+
+void
+exess_bigint_set_hex_string(ExessBigint* num, const char* const str)
+{
+  exess_bigint_zero(num);
+
+  // Read digits from right to left until we run off the beginning
+  const int length       = (int)strlen(str);
+  char      digit_buf[9] = {0, 0, 0, 0, 0, 0, 0, 0, 0};
+  int       i            = length - 8;
+  for (; i >= 0; i -= 8) {
+    memcpy(digit_buf, str + i, 8);
+    num->bigits[num->n_bigits++] = (Bigit)strtoll(digit_buf, NULL, 16);
+  }
+
+  // Read leftovers into MSB if necessary
+  if (i > -8) {
+    memset(digit_buf, 0, sizeof(digit_buf));
+    memcpy(digit_buf, str, 8u + (unsigned)i);
+    num->bigits[num->n_bigits++] = (Bigit)strtoll(digit_buf, NULL, 16);
+  }
+
+  exess_bigint_clamp(num);
+}
+
+void
+exess_bigint_multiply_u32(ExessBigint* num, const uint32_t factor)
+{
+  switch (factor) {
+  case 0:
+    exess_bigint_zero(num);
+    return;
+  case 1:
+    return;
+  default:
+    break;
+  }
+
+  Hugit carry = 0;
+  for (unsigned i = 0; i < num->n_bigits; ++i) {
+    const Hugit p     = (Hugit)factor * num->bigits[i];
+    const Hugit hugit = p + (carry & bigit_mask);
+
+    num->bigits[i] = (Bigit)(hugit & bigit_mask);
+
+    carry = (hugit >> 32u) + (carry >> 32u);
+  }
+
+  for (; carry; carry >>= 32u) {
+    assert(num->n_bigits + 1 <= BIGINT_MAX_BIGITS);
+    num->bigits[num->n_bigits++] = (Bigit)carry;
+  }
+}
+
+void
+exess_bigint_multiply_u64(ExessBigint* num, const uint64_t factor)
+{
+  switch (factor) {
+  case 0:
+    exess_bigint_zero(num);
+    return;
+  case 1:
+    return;
+  default:
+    break;
+  }
+
+  const Hugit f_lo = factor & bigit_mask;
+  const Hugit f_hi = factor >> 32u;
+
+  Hugit carry = 0;
+  for (unsigned i = 0; i < num->n_bigits; ++i) {
+    const Hugit p_lo  = f_lo * num->bigits[i];
+    const Hugit p_hi  = f_hi * num->bigits[i];
+    const Hugit hugit = p_lo + (carry & bigit_mask);
+
+    num->bigits[i] = (Bigit)(hugit & bigit_mask);
+    carry          = p_hi + (hugit >> 32u) + (carry >> 32u);
+  }
+
+  for (; carry; carry >>= 32u) {
+    assert(num->n_bigits + 1 <= BIGINT_MAX_BIGITS);
+    num->bigits[num->n_bigits++] = (Bigit)(carry & bigit_mask);
+  }
+}
+
+void
+exess_bigint_multiply_pow10(ExessBigint* num, const unsigned exponent)
+{
+  /* To reduce multiplication, we exploit 10^e = (2*5)^e = 2^e * 5^e to
+     factor out an exponentiation by 5 instead of 10.  So, we first multiply
+     by 5^e (hard), then by 2^e (just a single left shift). */
+
+  // 5^27, the largest power of 5 that fits in 64 bits
+  static const uint64_t pow5_27 = 7450580596923828125ull;
+
+  // Powers of 5 up to 5^13, the largest that fits in 32 bits
+  static const uint32_t pow5[] = {
+    1,
+    5,
+    5 * 5,
+    5 * 5 * 5,
+    5 * 5 * 5 * 5,
+    5 * 5 * 5 * 5 * 5,
+    5 * 5 * 5 * 5 * 5 * 5,
+    5 * 5 * 5 * 5 * 5 * 5 * 5,
+    5 * 5 * 5 * 5 * 5 * 5 * 5 * 5,
+    5 * 5 * 5 * 5 * 5 * 5 * 5 * 5 * 5,
+    5 * 5 * 5 * 5 * 5 * 5 * 5 * 5 * 5 * 5,
+    5 * 5 * 5 * 5 * 5 * 5 * 5 * 5 * 5 * 5 * 5,
+    5 * 5 * 5 * 5 * 5 * 5 * 5 * 5 * 5 * 5 * 5 * 5,
+    5 * 5 * 5 * 5 * 5 * 5 * 5 * 5 * 5 * 5 * 5 * 5 * 5,
+  };
+
+  if (exponent == 0 || num->n_bigits == 0) {
+    return;
+  }
+
+  // Multiply by 5^27 until e < 27 so we can switch to 32 bits
+  unsigned e = exponent;
+  while (e >= 27) {
+    exess_bigint_multiply_u64(num, pow5_27);
+    e -= 27;
+  }
+
+  // Multiply by 5^13 until e < 13 so we have only one multiplication left
+  while (e >= 13) {
+    exess_bigint_multiply_u32(num, pow5[13]);
+    e -= 13;
+  }
+
+  // Multiply by the final 5^e (which may be zero, making this a noop)
+  exess_bigint_multiply_u32(num, pow5[e]);
+
+  // Finally multiply by 2^e
+  exess_bigint_shift_left(num, exponent);
+}
+
+int
+exess_bigint_compare(const ExessBigint* lhs, const ExessBigint* rhs)
+{
+  if (lhs->n_bigits < rhs->n_bigits) {
+    return -1;
+  }
+
+  if (lhs->n_bigits > rhs->n_bigits) {
+    return 1;
+  }
+
+  for (int i = (int)lhs->n_bigits - 1; i >= 0; --i) {
+    const Bigit bigit_l = lhs->bigits[i];
+    const Bigit bigit_r = rhs->bigits[i];
+    if (bigit_l < bigit_r) {
+      return -1;
+    }
+
+    if (bigit_l > bigit_r) {
+      return 1;
+    }
+  }
+
+  return 0;
+}
+
+int
+exess_bigint_plus_compare(const ExessBigint* l,
+                          const ExessBigint* p,
+                          const ExessBigint* c)
+{
+  assert(exess_bigint_is_clamped(l));
+  assert(exess_bigint_is_clamped(p));
+  assert(exess_bigint_is_clamped(c));
+
+  if (l->n_bigits < p->n_bigits) {
+    return exess_bigint_plus_compare(p, l, c);
+  }
+
+  if (l->n_bigits + 1 < c->n_bigits) {
+    return -1;
+  }
+
+  if (l->n_bigits > c->n_bigits) {
+    return 1;
+  }
+
+  if (p->n_bigits < l->n_bigits && l->n_bigits < c->n_bigits) {
+    return -1;
+  }
+
+  Hugit borrow = 0;
+  for (int i = (int)c->n_bigits - 1; i >= 0; --i) {
+    const Bigit ai  = l->bigits[i];
+    const Bigit bi  = p->bigits[i];
+    const Bigit ci  = c->bigits[i];
+    const Hugit sum = (Hugit)ai + bi;
+
+    if (sum > ci + borrow) {
+      return 1;
+    }
+
+    if ((borrow += ci - sum) > 1) {
+      return -1;
+    }
+
+    borrow <<= 32u;
+  }
+
+  return borrow ? -1 : 0;
+}
+
+void
+exess_bigint_add_u32(ExessBigint* lhs, const uint32_t rhs)
+{
+  if (lhs->n_bigits == 0) {
+    exess_bigint_set_u32(lhs, rhs);
+    return;
+  }
+
+  Hugit sum   = (Hugit)lhs->bigits[0] + rhs;
+  Bigit carry = (Bigit)(sum >> 32u);
+
+  lhs->bigits[0] = (Bigit)(sum & bigit_mask);
+
+  unsigned i = 1;
+  for (; carry; ++i) {
+    assert(carry == 0 || carry == 1);
+
+    sum            = (Hugit)carry + lhs->bigits[i];
+    lhs->bigits[i] = (Bigit)(sum & bigit_mask);
+    carry          = (Bigit)((sum & carry_mask) >> 32u);
+  }
+
+  lhs->n_bigits = MAX(i, lhs->n_bigits);
+  assert(exess_bigint_is_clamped(lhs));
+}
+
+void
+exess_bigint_add(ExessBigint* lhs, const ExessBigint* rhs)
+{
+  assert(MAX(lhs->n_bigits, rhs->n_bigits) + 1 <= BIGINT_MAX_BIGITS);
+
+  bool     carry = 0;
+  unsigned i     = 0;
+  for (; i < rhs->n_bigits; ++i) {
+    const Hugit sum = (Hugit)lhs->bigits[i] + rhs->bigits[i] + carry;
+
+    lhs->bigits[i] = (Bigit)(sum & bigit_mask);
+    carry          = (sum & carry_mask) >> 32u;
+  }
+
+  for (; carry; ++i) {
+    const Hugit sum = (Hugit)lhs->bigits[i] + carry;
+
+    lhs->bigits[i] = (Bigit)(sum & bigit_mask);
+    carry          = (sum & carry_mask) >> 32u;
+  }
+
+  lhs->n_bigits = MAX(i, lhs->n_bigits);
+  assert(exess_bigint_is_clamped(lhs));
+}
+
+void
+exess_bigint_subtract(ExessBigint* lhs, const ExessBigint* rhs)
+{
+  assert(exess_bigint_is_clamped(lhs));
+  assert(exess_bigint_is_clamped(rhs));
+  assert(exess_bigint_compare(lhs, rhs) >= 0);
+
+  bool     borrow = 0;
+  unsigned i      = 0;
+  for (i = 0; i < rhs->n_bigits; ++i) {
+    const Bigit l = lhs->bigits[i];
+    const Bigit r = rhs->bigits[i];
+
+    lhs->bigits[i] = l - r - borrow;
+    borrow         = l < r || (l == r && borrow);
+  }
+
+  for (; borrow; ++i) {
+    const Bigit l = lhs->bigits[i];
+
+    lhs->bigits[i] -= borrow;
+
+    borrow = l == 0;
+  }
+
+  exess_bigint_clamp(lhs);
+}
+
+static unsigned
+exess_bigint_leading_zeros(const ExessBigint* num)
+{
+  return 32 * (BIGINT_MAX_BIGITS - num->n_bigits) +
+         exess_clz32(num->bigits[num->n_bigits - 1]);
+}
+
+// EXESS_I_PURE_FUNC
+static Bigit
+exess_bigint_left_shifted_bigit_i(const ExessBigint* num,
+                                  const Offset       amount,
+                                  const unsigned     index)
+{
+  /* assert(exess_bigint_is_clamped(num)); */
+  if (amount.bigits == 0 && amount.bits == 0) {
+    return num->bigits[index];
+  }
+
+  if (index < amount.bigits) {
+    return 0;
+  }
+
+  if (amount.bits == 0) { // Simple bigit-aligned shift
+    return num->bigits[index - amount.bigits];
+  }
+
+  if (index == amount.bigits) { // Last non-zero bigit
+    return num->bigits[0] << amount.bits;
+  }
+
+  // Bigit + sub-bigit bit offset shift
+  const unsigned right_shift = BIGINT_BIGIT_BITS - amount.bits;
+  return (num->bigits[index - amount.bigits] << amount.bits) |
+         (num->bigits[index - amount.bigits - 1] >> right_shift);
+}
+
+Bigit
+exess_bigint_left_shifted_bigit(const ExessBigint* num,
+                                const unsigned     amount,
+                                const unsigned     index)
+{
+  return exess_bigint_left_shifted_bigit_i(num, make_offset(amount), index);
+}
+
+void
+exess_bigint_subtract_left_shifted(ExessBigint*       lhs,
+                                   const ExessBigint* rhs,
+                                   const unsigned     amount)
+{
+  assert(exess_bigint_is_clamped(lhs));
+  assert(exess_bigint_is_clamped(rhs));
+#ifndef NDEBUG
+  {
+    ExessBigint check_rhs = *rhs;
+    exess_bigint_shift_left(&check_rhs, amount);
+    assert(exess_bigint_compare(lhs, &check_rhs) >= 0);
+  }
+#endif
+
+  const Offset   offset     = make_offset(amount);
+  const unsigned r_n_bigits = rhs->n_bigits + offset.bigits + (bool)offset.bits;
+
+  bool     borrow = 0;
+  unsigned i      = 0;
+  for (i = 0; i < r_n_bigits; ++i) {
+    const Bigit l = lhs->bigits[i];
+    const Bigit r = exess_bigint_left_shifted_bigit_i(rhs, offset, i);
+
+    lhs->bigits[i] = l - r - borrow;
+    borrow         = l < r || ((l == r) && borrow);
+  }
+
+  for (; borrow; ++i) {
+    const Bigit l = lhs->bigits[i];
+
+    lhs->bigits[i] -= borrow;
+
+    borrow = l == 0;
+  }
+
+  exess_bigint_clamp(lhs);
+}
+
+uint32_t
+exess_bigint_divmod(ExessBigint* lhs, const ExessBigint* rhs)
+{
+  assert(exess_bigint_is_clamped(lhs));
+  assert(exess_bigint_is_clamped(rhs));
+  assert(rhs->n_bigits > 0);
+  if (lhs->n_bigits < rhs->n_bigits) {
+    return 0;
+  }
+
+  uint32_t       result = 0;
+  const Bigit    r0     = rhs->bigits[rhs->n_bigits - 1];
+  const unsigned rlz    = exess_bigint_leading_zeros(rhs);
+
+  // Shift and subtract until the LHS does not have more bigits
+  int big_steps = 0;
+  while (lhs->n_bigits > rhs->n_bigits) {
+    const unsigned llz   = exess_bigint_leading_zeros(lhs);
+    const unsigned shift = rlz - llz - 1;
+
+    result += 1u << shift;
+    exess_bigint_subtract_left_shifted(lhs, rhs, shift);
+    ++big_steps;
+  }
+
+  // Handle simple termination cases
+  int cmp = exess_bigint_compare(lhs, rhs);
+  if (cmp < 0) {
+    return result;
+  }
+
+  if (cmp > 0 && lhs->n_bigits == 1) {
+    assert(rhs->n_bigits == 1);
+    const Bigit l0 = lhs->bigits[lhs->n_bigits - 1];
+
+    lhs->bigits[lhs->n_bigits - 1] = l0 % r0;
+    lhs->n_bigits -= (lhs->bigits[lhs->n_bigits - 1] == 0);
+    return result + l0 / r0;
+  }
+
+  // Both now have the same number of digits, finish with subtraction
+  int final_steps = 0;
+  for (; cmp >= 0; cmp = exess_bigint_compare(lhs, rhs)) {
+    const unsigned llz = exess_bigint_leading_zeros(lhs);
+    if (rlz == llz) {
+      // Both have the same number of leading zeros, just subtract
+      exess_bigint_subtract(lhs, rhs);
+      return result + 1;
+    }
+
+    const unsigned shift = rlz - llz - 1;
+    result += 1u << shift;
+    exess_bigint_subtract_left_shifted(lhs, rhs, shift);
+    ++final_steps;
+  }
+
+  return result;
+}
diff --git a/subprojects/exess/src/bigint.h b/subprojects/exess/src/bigint.h
new file mode 100644
index 00000000..088052f6
--- /dev/null
+++ b/subprojects/exess/src/bigint.h
@@ -0,0 +1,118 @@
+/*
+  Copyright 2019-2021 David Robillard <d@drobilla.net>
+
+  Permission to use, copy, modify, and/or distribute this software for any
+  purpose with or without fee is hereby granted, provided that the above
+  copyright notice and this permission notice appear in all copies.
+
+  THIS SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+  WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+  MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+  ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+  WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+  ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+  OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+*/
+
+#ifndef EXESS_BIGINT_H
+#define EXESS_BIGINT_H
+
+#include "attributes.h"
+
+#include <stddef.h>
+#include <stdint.h>
+#include <stdio.h>
+
+typedef uint32_t Bigit;
+
+/* We need enough precision for any double, the "largest" of which (using
+   absolute exponents) is the smallest subnormal ~= 5e-324.  This is 1076 bits
+   long, but we need a bit more space for arithmetic.  This is absurd, but such
+   is decimal.  These are only used on the stack so it doesn't hurt too much.
+*/
+
+#define BIGINT_MAX_SIGNIFICANT_BITS 1280u
+#define BIGINT_BIGIT_BITS 32u
+#define BIGINT_MAX_BIGITS (BIGINT_MAX_SIGNIFICANT_BITS / BIGINT_BIGIT_BITS)
+
+typedef struct {
+  Bigit    bigits[BIGINT_MAX_BIGITS];
+  unsigned n_bigits;
+} ExessBigint;
+
+void
+exess_bigint_zero(ExessBigint* num);
+
+size_t
+exess_bigint_print_hex(FILE* stream, const ExessBigint* num);
+
+void
+exess_bigint_clamp(ExessBigint* num);
+
+void
+exess_bigint_shift_left(ExessBigint* num, unsigned amount);
+
+void
+exess_bigint_set(ExessBigint* num, const ExessBigint* value);
+
+void
+exess_bigint_set_u32(ExessBigint* num, uint32_t value);
+
+void
+exess_bigint_set_u64(ExessBigint* num, uint64_t value);
+
+void
+exess_bigint_set_pow10(ExessBigint* num, unsigned exponent);
+
+void
+exess_bigint_set_decimal_string(ExessBigint* num, const char* str);
+
+void
+exess_bigint_set_hex_string(ExessBigint* num, const char* str);
+
+void
+exess_bigint_multiply_u32(ExessBigint* num, uint32_t factor);
+
+void
+exess_bigint_multiply_u64(ExessBigint* num, uint64_t factor);
+
+void
+exess_bigint_multiply_pow10(ExessBigint* num, unsigned exponent);
+
+EXESS_I_PURE_FUNC
+int
+exess_bigint_compare(const ExessBigint* lhs, const ExessBigint* rhs);
+
+void
+exess_bigint_add_u32(ExessBigint* lhs, uint32_t rhs);
+
+void
+exess_bigint_add(ExessBigint* lhs, const ExessBigint* rhs);
+
+void
+exess_bigint_subtract(ExessBigint* lhs, const ExessBigint* rhs);
+
+EXESS_I_PURE_FUNC
+Bigit
+exess_bigint_left_shifted_bigit(const ExessBigint* num,
+                                unsigned           amount,
+                                unsigned           index);
+
+/// Faster implementation of exess_bigint_subtract(lhs, rhs << amount)
+void
+exess_bigint_subtract_left_shifted(ExessBigint*       lhs,
+                                   const ExessBigint* rhs,
+                                   unsigned           amount);
+
+/// Faster implementation of exess_bigint_compare(l + p, c)
+EXESS_I_PURE_FUNC
+int
+exess_bigint_plus_compare(const ExessBigint* l,
+                          const ExessBigint* p,
+                          const ExessBigint* c);
+
+/// Divide and set `lhs` to modulo
+uint32_t
+exess_bigint_divmod(ExessBigint* lhs, const ExessBigint* rhs);
+
+#endif // EXESS_BIGINT_H
diff --git a/subprojects/exess/src/boolean.c b/subprojects/exess/src/boolean.c
new file mode 100644
index 00000000..32657da9
--- /dev/null
+++ b/subprojects/exess/src/boolean.c
@@ -0,0 +1,69 @@
+/*
+  Copyright 2019-2021 David Robillard <d@drobilla.net>
+
+  Permission to use, copy, modify, and/or distribute this software for any
+  purpose with or without fee is hereby granted, provided that the above
+  copyright notice and this permission notice appear in all copies.
+
+  THIS SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+  WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+  MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+  ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+  WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+  ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+  OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+*/
+
+#include "read_utils.h"
+#include "write_utils.h"
+
+#include "exess/exess.h"
+
+#include <stdbool.h>
+#include <string.h>
+
+ExessResult
+exess_read_boolean(bool* const out, const char* const str)
+{
+  size_t      i = skip_whitespace(str);
+  ExessResult r = {EXESS_EXPECTED_BOOLEAN, i};
+
+  *out = false;
+
+  switch (str[i]) {
+  case '0':
+    return end_read(EXESS_SUCCESS, str, ++i);
+
+  case '1':
+    *out = true;
+    return end_read(EXESS_SUCCESS, str, ++i);
+
+  case 't':
+    if (!strncmp(str + i, "true", 4)) {
+      *out = true;
+      return end_read(EXESS_SUCCESS, str, i + 4u);
+    }
+    break;
+
+  case 'f':
+    if (!strncmp(str + i, "false", 5)) {
+      return end_read(EXESS_SUCCESS, str, i + 5u);
+    }
+    break;
+
+  default:
+    break;
+  }
+
+  return end_read(r.status, str, r.count);
+}
+
+ExessResult
+exess_write_boolean(const bool value, const size_t buf_size, char* const buf)
+{
+  return end_write(EXESS_SUCCESS,
+                   buf_size,
+                   buf,
+                   value ? write_string(4, "true", buf_size, buf, 0)
+                         : write_string(5, "false", buf_size, buf, 0));
+}
diff --git a/subprojects/exess/src/byte.c b/subprojects/exess/src/byte.c
new file mode 100644
index 00000000..4754d82d
--- /dev/null
+++ b/subprojects/exess/src/byte.c
@@ -0,0 +1,45 @@
+/*
+  Copyright 2019-2021 David Robillard <d@drobilla.net>
+
+  Permission to use, copy, modify, and/or distribute this software for any
+  purpose with or without fee is hereby granted, provided that the above
+  copyright notice and this permission notice appear in all copies.
+
+  THIS SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+  WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+  MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+  ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+  WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+  ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+  OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+*/
+
+#include "read_utils.h"
+
+#include "exess/exess.h"
+
+#include <stdint.h>
+#include <stdlib.h>
+
+ExessResult
+exess_read_byte(int8_t* const out, const char* const str)
+{
+  int64_t           long_out = 0;
+  const ExessResult r        = exess_read_long(&long_out, str);
+  if (r.status) {
+    return r;
+  }
+
+  if (long_out < INT8_MIN || long_out > INT8_MAX) {
+    return result(EXESS_OUT_OF_RANGE, r.count);
+  }
+
+  *out = (int8_t)long_out;
+  return r;
+}
+
+ExessResult
+exess_write_byte(const int8_t value, const size_t buf_size, char* const buf)
+{
+  return exess_write_long(value, buf_size, buf);
+}
diff --git a/subprojects/exess/src/canonical.c b/subprojects/exess/src/canonical.c
new file mode 100644
index 00000000..f70aaecc
--- /dev/null
+++ b/subprojects/exess/src/canonical.c
@@ -0,0 +1,309 @@
+/*
+  Copyright 2019-2021 David Robillard <d@drobilla.net>
+
+  Permission to use, copy, modify, and/or distribute this software for any
+  purpose with or without fee is hereby granted, provided that the above
+  copyright notice and this permission notice appear in all copies.
+
+  THIS SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+  WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+  MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+  ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+  WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+  ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+  OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+*/
+
+#include "read_utils.h"
+#include "string_utils.h"
+#include "write_utils.h"
+
+#include "exess/exess.h"
+
+#include <stdbool.h>
+#include <stddef.h>
+
+typedef enum {
+  EXESS_NEGATIVE,
+  EXESS_ZERO,
+  EXESS_POSITIVE,
+} ExessIntegerKind;
+
+/// Return true iff `c` is "+" or "-"
+static inline bool
+is_sign(const int c)
+{
+  return c == '+' || c == '-';
+}
+
+/// Return true iff `c` is "0"
+static inline bool
+is_zero(const int c)
+{
+  return c == '0';
+}
+
+/// Return true iff `c` is "."
+static inline bool
+is_point(const int c)
+{
+  return c == '.';
+}
+
+// Scan forwards as long as `pred` returns true for characters
+static inline size_t
+scan(bool (*pred)(const int), const char* const str, size_t i)
+{
+  while (pred(str[i])) {
+    ++i;
+  }
+
+  return i;
+}
+
+// Skip the next character if `pred` returns true for it
+static inline size_t
+skip(bool (*pred)(const int), const char* const str, const size_t i)
+{
+  return i + (pred(str[i]) ? 1 : 0);
+}
+
+static ExessResult
+write_decimal(const char* const str, const size_t buf_size, char* const buf)
+{
+  size_t i = 0;
+
+  const size_t sign    = scan(is_space, str, i);   // Sign
+  const size_t leading = skip(is_sign, str, sign); // First digit
+  if (str[leading] != '.' && !is_digit(str[leading])) {
+    return result(EXESS_EXPECTED_DIGIT, sign);
+  }
+
+  const size_t first = scan(is_zero, str, leading); // First non-zero
+  const size_t point = scan(is_digit, str, first);  // Decimal point
+  size_t last = scan(is_digit, str, skip(is_point, str, point)); // Last digit
+  const size_t end = scan(is_space, str, last); // Last non-space
+
+  const ExessStatus st = is_end(str[end]) ? EXESS_SUCCESS : EXESS_EXPECTED_END;
+
+  // Ignore trailing zeros
+  if (str[point] == '.') {
+    while (str[last - 1] == '0') {
+      --last;
+    }
+  }
+
+  // Add leading sign only if the number is negative
+  size_t o = 0;
+  if (str[sign] == '-') {
+    o += write_char('-', buf_size, buf, o);
+  }
+
+  // Handle zero as a special case (no non-zero digits to copy)
+  if (first == last) {
+    o += write_string(3, "0.0", buf_size, buf, o);
+    return result(EXESS_SUCCESS, o);
+  }
+
+  // Add leading zero if needed to have at least one digit before the point
+  if (str[first] == '.') {
+    o += write_char('0', buf_size, buf, o);
+  }
+
+  // Add digits
+  o += write_string(last - first, str + first, buf_size, buf, o);
+
+  if (str[point] != '.') {
+    // Add missing decimal suffix
+    o += write_string(2, ".0", buf_size, buf, o);
+  } else if (point == last - 1) {
+    // Add missing trailing zero after point
+    o += write_char('0', buf_size, buf, o);
+  }
+
+  return result(st, o);
+}
+
+static ExessResult
+write_integer(const char* const       str,
+              const size_t            buf_size,
+              char* const             buf,
+              ExessIntegerKind* const kind)
+{
+  const size_t sign    = scan(is_space, str, 0);   // Sign
+  const size_t leading = skip(is_sign, str, sign); // First digit
+  if (!is_digit(str[leading])) {
+    return result(EXESS_EXPECTED_DIGIT, sign);
+  }
+
+  const size_t first = scan(is_zero, str, leading); // First non-zero
+  const size_t last  = scan(is_digit, str, first);  // Last digit
+  const size_t end   = scan(is_space, str, last);   // Last non-space
+
+  const ExessStatus st = is_end(str[end]) ? EXESS_SUCCESS : EXESS_EXPECTED_END;
+
+  // Handle zero as a special case (no non-zero digits to copy)
+  size_t o = 0;
+  if (first == last) {
+    o += write_char('0', buf_size, buf, o);
+    *kind = EXESS_ZERO;
+    return result(EXESS_SUCCESS, o);
+  }
+
+  // Add leading sign only if the number is negative
+  if (str[sign] == '-') {
+    *kind = EXESS_NEGATIVE;
+    o += write_char('-', buf_size, buf, o);
+  } else {
+    *kind = EXESS_POSITIVE;
+  }
+
+  // Add digits
+  o += write_string(last - first, str + first, buf_size, buf, o);
+
+  return result(st, o);
+}
+
+static ExessResult
+write_hex(const char* const str, const size_t buf_size, char* const buf)
+{
+  size_t i = 0;
+  size_t o = 0;
+
+  for (; str[i]; ++i) {
+    if (is_hexdig(str[i])) {
+      o += write_char(str[i], buf_size, buf, o);
+    } else if (!is_space(str[i])) {
+      return result(EXESS_EXPECTED_HEX, o);
+    }
+  }
+
+  if (o == 0 || o % 2 != 0) {
+    return result(EXESS_EXPECTED_HEX, o);
+  }
+
+  return result(EXESS_SUCCESS, o);
+}
+
+static ExessResult
+write_base64(const char* const str, const size_t buf_size, char* const buf)
+{
+  size_t i = 0;
+  size_t o = 0;
+
+  for (; str[i]; ++i) {
+    if (is_base64(str[i])) {
+      o += write_char(str[i], buf_size, buf, o);
+    } else if (!is_space(str[i])) {
+      return result(EXESS_EXPECTED_BASE64, o);
+    }
+  }
+
+  if (o == 0 || o % 4 != 0) {
+    return result(EXESS_EXPECTED_BASE64, o);
+  }
+
+  return result(EXESS_SUCCESS, o);
+}
+
+static ExessResult
+write_bounded(const char* const   str,
+              const ExessDatatype datatype,
+              const size_t        buf_size,
+              char* const         buf)
+{
+  ExessVariant      variant = {EXESS_NOTHING, {EXESS_SUCCESS}};
+  const ExessResult r       = exess_read_variant(&variant, datatype, str);
+
+  return r.status ? r : exess_write_variant(variant, buf_size, buf);
+}
+
+ExessResult
+exess_write_canonical(const char* const   str,
+                      const ExessDatatype datatype,
+                      const size_t        buf_size,
+                      char* const         buf)
+{
+  ExessIntegerKind kind = EXESS_ZERO;
+  ExessResult      r    = {EXESS_UNSUPPORTED, 0};
+
+  switch (datatype) {
+  case EXESS_NOTHING:
+    break;
+
+  case EXESS_BOOLEAN:
+    r = write_bounded(str, datatype, buf_size, buf);
+    break;
+
+  case EXESS_DECIMAL:
+    r = write_decimal(str, buf_size, buf);
+    break;
+
+  case EXESS_DOUBLE:
+  case EXESS_FLOAT:
+    r = write_bounded(str, datatype, buf_size, buf);
+    break;
+
+  case EXESS_INTEGER:
+    r = write_integer(str, buf_size, buf, &kind);
+    break;
+
+  case EXESS_NON_POSITIVE_INTEGER:
+    r = write_integer(str, buf_size, buf, &kind);
+    if (kind == EXESS_POSITIVE) {
+      r.status = EXESS_BAD_VALUE;
+    }
+    break;
+
+  case EXESS_NEGATIVE_INTEGER:
+    r = write_integer(str, buf_size, buf, &kind);
+    if (kind == EXESS_ZERO || kind == EXESS_POSITIVE) {
+      r.status = EXESS_BAD_VALUE;
+    }
+    break;
+
+  case EXESS_LONG:
+  case EXESS_INT:
+  case EXESS_SHORT:
+  case EXESS_BYTE:
+    r = write_bounded(str, datatype, buf_size, buf);
+    break;
+
+  case EXESS_NON_NEGATIVE_INTEGER:
+    r = write_integer(str, buf_size, buf, &kind);
+    if (kind == EXESS_NEGATIVE) {
+      r.status = EXESS_BAD_VALUE;
+    }
+    break;
+
+  case EXESS_ULONG:
+  case EXESS_UINT:
+  case EXESS_USHORT:
+  case EXESS_UBYTE:
+    r = write_bounded(str, datatype, buf_size, buf);
+    break;
+
+  case EXESS_POSITIVE_INTEGER:
+    r = write_integer(str, buf_size, buf, &kind);
+    if (kind == EXESS_NEGATIVE || kind == EXESS_ZERO) {
+      r.status = EXESS_BAD_VALUE;
+    }
+    break;
+
+  case EXESS_DURATION:
+  case EXESS_DATETIME:
+  case EXESS_TIME:
+  case EXESS_DATE:
+    r = write_bounded(str, datatype, buf_size, buf);
+    break;
+
+  case EXESS_HEX:
+    r = write_hex(str, buf_size, buf);
+    break;
+
+  case EXESS_BASE64:
+    r = write_base64(str, buf_size, buf);
+  }
+
+  return end_write(r.status, buf_size, buf, r.count);
+}
diff --git a/subprojects/exess/src/coerce.c b/subprojects/exess/src/coerce.c
new file mode 100644
index 00000000..1a0c35dc
--- /dev/null
+++ b/subprojects/exess/src/coerce.c
@@ -0,0 +1,422 @@
+/*
+  Copyright 2019-2021 David Robillard <d@drobilla.net>
+
+  Permission to use, copy, modify, and/or distribute this software for any
+  purpose with or without fee is hereby granted, provided that the above
+  copyright notice and this permission notice appear in all copies.
+
+  THIS SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+  WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+  MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+  ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+  WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+  ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+  OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+*/
+
+#include "exess/exess.h"
+
+#include <math.h>
+#include <stdint.h>
+
+/* Limits for the range of integers that can be exactly represented in floating
+   point types.  Note that these limits are one less than the largest value,
+   since values larger than that may round to it which causes problems with
+   perfect round-tripping.  For example, 16777217 when parsed as a float will
+   result in 1.6777216E7, which a "lossless" coercion would then convert to
+   16777216. */
+
+#define MAX_FLOAT_INT 16777215
+#define MAX_DOUBLE_INT 9007199254740991L
+
+static ExessVariant
+coerce_long_in_range(const ExessVariant variant,
+                     const int64_t      min,
+                     const int64_t      max)
+{
+  const ExessVariant result = exess_coerce(variant, EXESS_LONG, EXESS_LOSSLESS);
+  if (result.datatype == EXESS_LONG) {
+    if (result.value.as_long < min || result.value.as_long > max) {
+      return exess_make_nothing(EXESS_OUT_OF_RANGE);
+    }
+  }
+
+  return result;
+}
+
+static ExessVariant
+coerce_to_long(const ExessVariant variant, const ExessCoercionFlags coercions)
+{
+  switch (variant.datatype) {
+  case EXESS_NOTHING:
+    return variant;
+
+  case EXESS_BOOLEAN:
+    return exess_make_long(variant.value.as_bool);
+
+  case EXESS_DECIMAL:
+  case EXESS_DOUBLE:
+    if (!(coercions & (ExessCoercionFlags)EXESS_ROUND) &&
+        variant.value.as_double > trunc(variant.value.as_double)) {
+      return exess_make_nothing(EXESS_WOULD_ROUND);
+    }
+
+    if (variant.value.as_double < (double)-MAX_DOUBLE_INT ||
+        variant.value.as_double > (double)MAX_DOUBLE_INT) {
+      return exess_make_nothing(EXESS_OUT_OF_RANGE);
+    }
+
+    return exess_make_long(lrint(variant.value.as_double));
+
+  case EXESS_FLOAT:
+    if (!(coercions & (ExessCoercionFlags)EXESS_ROUND) &&
+        variant.value.as_float > truncf(variant.value.as_float)) {
+      return exess_make_nothing(EXESS_WOULD_ROUND);
+    }
+
+    if (variant.value.as_float < (float)-MAX_FLOAT_INT ||
+        variant.value.as_float > (float)MAX_FLOAT_INT) {
+      return exess_make_nothing(EXESS_OUT_OF_RANGE);
+    }
+
+    return exess_make_long(lrintf(variant.value.as_float));
+
+  case EXESS_INTEGER:
+  case EXESS_NON_POSITIVE_INTEGER:
+  case EXESS_NEGATIVE_INTEGER:
+  case EXESS_LONG:
+    return exess_make_long(variant.value.as_long);
+
+  case EXESS_INT:
+    return exess_make_long(variant.value.as_int);
+
+  case EXESS_SHORT:
+    return exess_make_long(variant.value.as_short);
+
+  case EXESS_BYTE:
+    return exess_make_long(variant.value.as_byte);
+
+  case EXESS_NON_NEGATIVE_INTEGER:
+    return exess_make_long(variant.value.as_long);
+
+  case EXESS_ULONG:
+    if (variant.value.as_ulong > INT64_MAX) {
+      return exess_make_nothing(EXESS_OUT_OF_RANGE);
+    }
+
+    return exess_make_long((int64_t)variant.value.as_ulong);
+
+  case EXESS_UINT:
+    return exess_make_long(variant.value.as_uint);
+
+  case EXESS_USHORT:
+    return exess_make_long(variant.value.as_ushort);
+
+  case EXESS_UBYTE:
+    return exess_make_long(variant.value.as_ubyte);
+
+  case EXESS_POSITIVE_INTEGER:
+    if (variant.value.as_ulong > INT64_MAX) {
+      return exess_make_nothing(EXESS_OUT_OF_RANGE);
+    }
+
+    return exess_make_long((int64_t)variant.value.as_ulong);
+
+  case EXESS_DURATION:
+  case EXESS_DATETIME:
+  case EXESS_TIME:
+  case EXESS_DATE:
+  case EXESS_HEX:
+  case EXESS_BASE64:
+    break;
+  }
+
+  return exess_make_nothing(EXESS_UNSUPPORTED);
+}
+
+static ExessVariant
+coerce_ulong_in_range(const ExessVariant variant, const uint64_t max)
+{
+  const ExessVariant result =
+    exess_coerce(variant, EXESS_ULONG, EXESS_LOSSLESS);
+
+  if (result.datatype == EXESS_ULONG) {
+    if (variant.value.as_ulong > max) {
+      return exess_make_nothing(EXESS_OUT_OF_RANGE);
+    }
+  }
+
+  return result;
+}
+
+static ExessVariant
+coerce_to_ulong(const ExessVariant value, const ExessCoercionFlags coercions)
+{
+  switch (value.datatype) {
+  case EXESS_NOTHING:
+    return value;
+
+  case EXESS_BOOLEAN:
+    return exess_make_ulong(value.value.as_bool);
+
+  case EXESS_DECIMAL:
+  case EXESS_DOUBLE:
+    if (!(coercions & (ExessCoercionFlags)EXESS_ROUND) &&
+        value.value.as_double > trunc(value.value.as_double)) {
+      return exess_make_nothing(EXESS_WOULD_ROUND);
+    }
+
+    if (value.value.as_double < 0.0 ||
+        value.value.as_double > (double)MAX_DOUBLE_INT) {
+      return exess_make_nothing(EXESS_OUT_OF_RANGE);
+    }
+
+    return exess_make_ulong((uint64_t)lrint(value.value.as_double));
+
+  case EXESS_FLOAT:
+    if (!(coercions & (ExessCoercionFlags)EXESS_ROUND) &&
+        value.value.as_float > truncf(value.value.as_float)) {
+      return exess_make_nothing(EXESS_WOULD_ROUND);
+    }
+
+    if (value.value.as_float < 0.0f ||
+        value.value.as_float > (float)MAX_FLOAT_INT) {
+      return exess_make_nothing(EXESS_OUT_OF_RANGE);
+    }
+
+    return exess_make_ulong((uint64_t)lrintf(value.value.as_float));
+
+  case EXESS_INTEGER:
+  case EXESS_NON_POSITIVE_INTEGER:
+  case EXESS_NEGATIVE_INTEGER:
+  case EXESS_LONG:
+    if (value.value.as_long < 0) {
+      return exess_make_nothing(EXESS_OUT_OF_RANGE);
+    }
+
+    return exess_make_ulong((uint64_t)value.value.as_long);
+
+  case EXESS_INT:
+    if (value.value.as_int < 0) {
+      return exess_make_nothing(EXESS_OUT_OF_RANGE);
+    }
+
+    return exess_make_ulong((uint64_t)value.value.as_int);
+
+  case EXESS_SHORT:
+    if (value.value.as_short < 0) {
+      return exess_make_nothing(EXESS_OUT_OF_RANGE);
+    }
+
+    return exess_make_ulong((uint64_t)value.value.as_short);
+
+  case EXESS_BYTE:
+    if (value.value.as_byte < 0) {
+      return exess_make_nothing(EXESS_OUT_OF_RANGE);
+    }
+
+    return exess_make_ulong((uint64_t)value.value.as_byte);
+
+  case EXESS_NON_NEGATIVE_INTEGER:
+  case EXESS_ULONG:
+    return exess_make_ulong(value.value.as_ulong);
+
+  case EXESS_UINT:
+    return exess_make_ulong(value.value.as_uint);
+
+  case EXESS_USHORT:
+    return exess_make_ulong(value.value.as_ushort);
+
+  case EXESS_UBYTE:
+    return exess_make_ulong(value.value.as_ubyte);
+
+  case EXESS_POSITIVE_INTEGER:
+    return exess_make_ulong(value.value.as_ulong);
+
+  case EXESS_DURATION:
+  case EXESS_DATETIME:
+  case EXESS_TIME:
+  case EXESS_DATE:
+  case EXESS_HEX:
+  case EXESS_BASE64:
+    break;
+  }
+
+  return exess_make_nothing(EXESS_UNSUPPORTED);
+}
+
+ExessVariant
+exess_coerce(const ExessVariant       value,
+             const ExessDatatype      datatype,
+             const ExessCoercionFlags coercions)
+{
+  if (datatype == value.datatype) {
+    return value;
+  }
+
+  ExessVariant result = value;
+
+  switch (datatype) {
+  case EXESS_NOTHING:
+    break;
+
+  case EXESS_BOOLEAN:
+    result = exess_coerce(value, EXESS_LONG, coercions);
+    if (result.datatype == EXESS_LONG) {
+      if (!(coercions & (ExessCoercionFlags)EXESS_TRUNCATE) &&
+          result.value.as_long != 0 && result.value.as_long != 1) {
+        return exess_make_nothing(EXESS_WOULD_TRUNCATE);
+      }
+
+      return exess_make_boolean(result.value.as_long != 0);
+    }
+    break;
+
+  case EXESS_DECIMAL:
+    // FIXME
+
+  case EXESS_DOUBLE:
+    if (value.datatype == EXESS_DECIMAL) {
+      return exess_make_double(value.value.as_double);
+    }
+
+    if (value.datatype == EXESS_FLOAT) {
+      return exess_make_double((double)value.value.as_float);
+    }
+
+    result = coerce_long_in_range(value, -MAX_DOUBLE_INT, MAX_DOUBLE_INT);
+    if (result.datatype == EXESS_LONG) {
+      return exess_make_double((double)result.value.as_long);
+    }
+
+    break;
+
+  case EXESS_FLOAT:
+    if (value.datatype == EXESS_DECIMAL || value.datatype == EXESS_DOUBLE) {
+      if (!(coercions & (ExessCoercionFlags)EXESS_REDUCE_PRECISION)) {
+        return exess_make_nothing(EXESS_WOULD_REDUCE_PRECISION);
+      }
+
+      return exess_make_float((float)result.value.as_double);
+    } else {
+      result = coerce_long_in_range(value, -MAX_FLOAT_INT, MAX_FLOAT_INT);
+      if (result.datatype == EXESS_LONG) {
+        return exess_make_float((float)result.value.as_long);
+      }
+    }
+
+    break;
+
+  case EXESS_INTEGER:
+    result = coerce_to_long(value, coercions);
+    break;
+
+  case EXESS_NON_POSITIVE_INTEGER:
+    result = coerce_to_long(value, coercions);
+    if (result.datatype == EXESS_LONG && result.value.as_long > 0) {
+      return exess_make_nothing(EXESS_OUT_OF_RANGE);
+    }
+
+    break;
+
+  case EXESS_NEGATIVE_INTEGER:
+    result = coerce_to_long(value, coercions);
+    if (result.datatype == EXESS_LONG && result.value.as_long >= 0) {
+      return exess_make_nothing(EXESS_OUT_OF_RANGE);
+    }
+    break;
+
+  case EXESS_LONG:
+    return coerce_to_long(value, coercions);
+
+  case EXESS_INT:
+    result = coerce_long_in_range(value, INT32_MIN, INT32_MAX);
+    break;
+
+  case EXESS_SHORT:
+    result = coerce_long_in_range(value, INT16_MIN, INT16_MAX);
+    break;
+
+  case EXESS_BYTE:
+    result = coerce_long_in_range(value, INT8_MIN, INT8_MAX);
+    break;
+
+  case EXESS_NON_NEGATIVE_INTEGER:
+  case EXESS_ULONG:
+    result = coerce_to_ulong(value, coercions);
+    break;
+
+  case EXESS_UINT:
+    result = coerce_ulong_in_range(value, UINT32_MAX);
+    break;
+
+  case EXESS_USHORT:
+    result = coerce_ulong_in_range(value, UINT16_MAX);
+    break;
+
+  case EXESS_UBYTE:
+    result = coerce_ulong_in_range(value, UINT8_MAX);
+    break;
+
+  case EXESS_POSITIVE_INTEGER:
+    result = coerce_to_ulong(value, coercions);
+    if (result.datatype == EXESS_ULONG && result.value.as_ulong == 0u) {
+      return exess_make_nothing(EXESS_OUT_OF_RANGE);
+    }
+    break;
+
+  case EXESS_DURATION:
+  case EXESS_DATETIME:
+    return exess_make_nothing(EXESS_UNSUPPORTED);
+
+  case EXESS_TIME:
+    if (value.datatype != EXESS_DATETIME) {
+      return exess_make_nothing(EXESS_UNSUPPORTED);
+    }
+
+    if (coercions & (ExessCoercionFlags)EXESS_TRUNCATE) {
+      const ExessTime time = {
+        {value.value.as_datetime.is_utc ? 0 : EXESS_LOCAL},
+        value.value.as_datetime.hour,
+        value.value.as_datetime.minute,
+        value.value.as_datetime.second,
+        value.value.as_datetime.nanosecond};
+
+      return exess_make_time(time);
+    }
+
+    return exess_make_nothing(EXESS_WOULD_TRUNCATE);
+
+  case EXESS_DATE:
+    if (value.datatype != EXESS_DATETIME) {
+      return exess_make_nothing(EXESS_UNSUPPORTED);
+    }
+
+    if (coercions & (ExessCoercionFlags)EXESS_TRUNCATE) {
+      const ExessDate date = {
+        value.value.as_datetime.year,
+        value.value.as_datetime.month,
+        value.value.as_datetime.day,
+        {value.value.as_datetime.is_utc ? 0 : EXESS_LOCAL}};
+      return exess_make_date(date);
+    }
+
+    return exess_make_nothing(EXESS_WOULD_TRUNCATE);
+
+  case EXESS_HEX:
+    return (value.datatype == EXESS_BASE64)
+             ? exess_make_hex(value.value.as_blob)
+             : exess_make_nothing(EXESS_UNSUPPORTED);
+
+  case EXESS_BASE64:
+    return (value.datatype == EXESS_HEX)
+             ? exess_make_base64(value.value.as_blob)
+             : exess_make_nothing(EXESS_UNSUPPORTED);
+  }
+
+  if (result.datatype != EXESS_NOTHING) {
+    result.datatype = datatype;
+  }
+
+  return result;
+}
diff --git a/subprojects/exess/src/datatype.c b/subprojects/exess/src/datatype.c
new file mode 100644
index 00000000..c7789597
--- /dev/null
+++ b/subprojects/exess/src/datatype.c
@@ -0,0 +1,79 @@
+/*
+  Copyright 2019-2021 David Robillard <d@drobilla.net>
+
+  Permission to use, copy, modify, and/or distribute this software for any
+  purpose with or without fee is hereby granted, provided that the above
+  copyright notice and this permission notice appear in all copies.
+
+  THIS SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+  WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+  MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+  ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+  WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+  ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+  OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+*/
+
+#include "exess/exess.h"
+
+#include <stdbool.h>
+#include <string.h>
+
+#define N_DATATYPES 24
+
+static const char* EXESS_NONNULL const uris[N_DATATYPES + 1] = {
+  NULL,                               //
+  EXESS_XSD_URI "boolean",            //
+  EXESS_XSD_URI "decimal",            //
+  EXESS_XSD_URI "double",             //
+  EXESS_XSD_URI "float",              //
+  EXESS_XSD_URI "integer",            //
+  EXESS_XSD_URI "nonPositiveInteger", //
+  EXESS_XSD_URI "negativeInteger",    //
+  EXESS_XSD_URI "long",               //
+  EXESS_XSD_URI "int",                //
+  EXESS_XSD_URI "short",              //
+  EXESS_XSD_URI "byte",               //
+  EXESS_XSD_URI "nonNegativeInteger", //
+  EXESS_XSD_URI "unsignedLong",       //
+  EXESS_XSD_URI "unsignedInt",        //
+  EXESS_XSD_URI "unsignedShort",      //
+  EXESS_XSD_URI "unsignedByte",       //
+  EXESS_XSD_URI "positiveInteger",    //
+  EXESS_XSD_URI "duration",           //
+  EXESS_XSD_URI "datetime",           //
+  EXESS_XSD_URI "time",               //
+  EXESS_XSD_URI "date",               //
+  EXESS_XSD_URI "hexBinary",          //
+  EXESS_XSD_URI "base64Binary",       //
+};
+
+const char*
+exess_datatype_uri(const ExessDatatype datatype)
+{
+  return (datatype > EXESS_NOTHING && datatype <= EXESS_BASE64) ? uris[datatype]
+                                                                : NULL;
+}
+
+ExessDatatype
+exess_datatype_from_uri(const char* const uri)
+{
+  static const size_t xsd_len = sizeof(EXESS_XSD_URI) - 1;
+
+  if (!strncmp(uri, EXESS_XSD_URI, xsd_len)) {
+    const char* const name = uri + xsd_len;
+    for (size_t i = 1; i < N_DATATYPES; ++i) {
+      if (!strcmp(name, uris[i] + xsd_len)) {
+        return (ExessDatatype)i;
+      }
+    }
+  }
+
+  return EXESS_NOTHING;
+}
+
+bool
+exess_datatype_is_bounded(const ExessDatatype datatype)
+{
+  return (datatype < N_DATATYPES) ? exess_max_lengths[datatype] != 0 : false;
+}
diff --git a/subprojects/exess/src/date.c b/subprojects/exess/src/date.c
new file mode 100644
index 00000000..77b336f7
--- /dev/null
+++ b/subprojects/exess/src/date.c
@@ -0,0 +1,112 @@
+/*
+  Copyright 2019-2021 David Robillard <d@drobilla.net>
+
+  Permission to use, copy, modify, and/or distribute this software for any
+  purpose with or without fee is hereby granted, provided that the above
+  copyright notice and this permission notice appear in all copies.
+
+  THIS SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+  WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+  MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+  ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+  WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+  ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+  OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+*/
+
+#include "date_utils.h"
+#include "read_utils.h"
+#include "timezone.h"
+#include "write_utils.h"
+
+#include "exess/exess.h"
+
+#include <stdlib.h>
+#include <string.h>
+
+ExessResult
+read_date_numbers(ExessDate* const out, const char* const str)
+{
+  // Read year at the beginning
+  size_t      i = skip_whitespace(str);
+  ExessResult r = read_year_number(&out->year, str + i);
+  if (r.status) {
+    return result(r.status, i + r.count);
+  }
+
+  // Read year-month delimiter
+  i += r.count;
+  if (str[i] != '-') {
+    return result(EXESS_EXPECTED_DASH, i);
+  }
+
+  // Read month
+  ++i;
+  if ((r = read_two_digit_number(&out->month, 1, 12, str + i)).status) {
+    return result(r.status, i + r.count);
+  }
+
+  // Read month-day delimiter
+  i += r.count;
+  if (str[i] != '-') {
+    return result(EXESS_EXPECTED_DASH, i);
+  }
+
+  // Read day
+  ++i;
+  if ((r = read_two_digit_number(&out->day, 1, 31, str + i)).status) {
+    return result(r.status, i + r.count);
+  }
+
+  // Check that day is in range
+  i += r.count;
+  if (out->day > days_in_month(out->year, out->month)) {
+    return result(EXESS_OUT_OF_RANGE, i);
+  }
+
+  return result(EXESS_SUCCESS, i);
+}
+
+ExessResult
+exess_read_date(ExessDate* const out, const char* const str)
+{
+  memset(out, 0, sizeof(*out));
+
+  // Read YYYY-MM-DD numbers
+  size_t      i = skip_whitespace(str);
+  ExessResult r = read_date_numbers(out, str + i);
+
+  i += r.count;
+  if (r.status || is_end(str[i])) {
+    out->zone.quarter_hours = EXESS_LOCAL;
+    return result(r.status, i);
+  }
+
+  // Read timezone
+  r = exess_read_timezone(&out->zone, str + i);
+
+  return result(r.status, i + r.count);
+}
+
+ExessResult
+exess_write_date(const ExessDate value, const size_t buf_size, char* const buf)
+{
+  if (value.month < 1 || value.month > 12 || value.day < 1 || value.day > 31) {
+    return end_write(EXESS_BAD_VALUE, buf_size, buf, 0);
+  }
+
+  ExessResult r = write_year_number(value.year, buf_size, buf);
+  size_t      o = r.count;
+  if (r.status) {
+    return end_write(r.status, buf_size, buf, o);
+  }
+
+  o += write_char('-', buf_size, buf, o);
+  o += write_two_digit_number(value.month, buf_size, buf, o);
+  o += write_char('-', buf_size, buf, o);
+  o += write_two_digit_number(value.day, buf_size, buf, o);
+
+  r = write_timezone(value.zone, buf_size, buf, o);
+
+  return end_write(r.status, buf_size, buf, o + r.count);
+}
diff --git a/subprojects/exess/src/date_utils.h b/subprojects/exess/src/date_utils.h
new file mode 100644
index 00000000..b864925e
--- /dev/null
+++ b/subprojects/exess/src/date_utils.h
@@ -0,0 +1,65 @@
+/*
+  Copyright 2019-2021 David Robillard <d@drobilla.net>
+
+  Permission to use, copy, modify, and/or distribute this software for any
+  purpose with or without fee is hereby granted, provided that the above
+  copyright notice and this permission notice appear in all copies.
+
+  THIS SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+  WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+  MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+  ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+  WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+  ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+  OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+*/
+
+#ifndef EXESS_DATE_UTILS_H
+#define EXESS_DATE_UTILS_H
+
+#include "exess/exess.h"
+
+#include <stdbool.h>
+#include <stddef.h>
+#include <stdint.h>
+
+static inline bool
+is_leap_year(const int64_t year)
+{
+  if (year % 4) {
+    return false;
+  }
+
+  if (year % 100) {
+    return true;
+  }
+
+  if (year % 400) {
+    return false;
+  }
+
+  return true;
+}
+
+static inline uint8_t
+days_in_month(const int16_t year, const uint8_t month)
+{
+  return month == 2u ? (is_leap_year(year) ? 29u : 28u)
+                     : (uint8_t)(30u + (month + (month / 8u)) % 2u);
+}
+
+ExessResult
+read_year_number(int16_t* out, const char* str);
+
+ExessResult
+write_year_number(int16_t value, size_t buf_size, char* buf);
+
+/// Read YYYY-MM-DD date numbers without a timezone
+ExessResult
+read_date_numbers(ExessDate* out, const char* str);
+
+EXESS_CONST_FUNC
+size_t
+exess_timezone_string_length(ExessTimezone value);
+
+#endif // EXESS_DATE_UTILS_H
diff --git a/subprojects/exess/src/datetime.c b/subprojects/exess/src/datetime.c
new file mode 100644
index 00000000..0fe56afe
--- /dev/null
+++ b/subprojects/exess/src/datetime.c
@@ -0,0 +1,265 @@
+/*
+  Copyright 2019-2021 David Robillard <d@drobilla.net>
+
+  Permission to use, copy, modify, and/or distribute this software for any
+  purpose with or without fee is hereby granted, provided that the above
+  copyright notice and this permission notice appear in all copies.
+
+  THIS SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+  WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+  MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+  ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+  WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+  ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+  OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+*/
+
+#include "date_utils.h"
+#include "read_utils.h"
+#include "string_utils.h"
+#include "time_utils.h"
+#include "write_utils.h"
+
+#include "exess/exess.h"
+
+#include <stdbool.h>
+#include <stdint.h>
+#include <stdlib.h>
+
+static inline ExessDateTime
+infinite_future(const bool is_utc)
+{
+  const ExessDateTime result = {INT16_MAX,
+                                UINT8_MAX,
+                                UINT8_MAX,
+                                is_utc,
+                                UINT8_MAX,
+                                UINT8_MAX,
+                                UINT8_MAX,
+                                UINT32_MAX};
+
+  return result;
+}
+
+static inline ExessDateTime
+infinite_past(const bool is_utc)
+{
+  const ExessDateTime result = {INT16_MIN, 0, 0, is_utc, 0, 0, 0, 0};
+
+  return result;
+}
+
+static int32_t
+modulo(const int32_t a, const int32_t low, const int32_t high)
+{
+  return ((a - low) % (high - low)) + low;
+}
+
+static int32_t
+quotient(const int32_t a, const int32_t low, const int32_t high)
+{
+  return (a - low) / (high - low);
+}
+
+ExessDateTime
+exess_add_datetime_duration(const ExessDateTime s, const ExessDuration d)
+{
+  /*
+    See <https://www.w3.org/TR/xmlschema-2/#adding-durations-to-dateTimes>.
+    This algorithm is modified to support subtraction when d is negative.
+  */
+
+  const int32_t d_year       = d.months / 12;
+  const int32_t d_month      = d.months % 12;
+  const int32_t d_day        = d.seconds / (24 * 60 * 60);
+  const int32_t d_hour       = d.seconds / 60 / 60 % 24;
+  const int32_t d_minute     = d.seconds / 60 % 60;
+  const int32_t d_second     = d.seconds % 60;
+  const int32_t d_nanosecond = d.nanoseconds;
+
+  ExessDateTime e     = {0, 0u, 0u, s.is_utc, 0u, 0u, 0u, 0u};
+  int32_t       temp  = 0;
+  int32_t       carry = 0;
+
+  // Months (may be modified additionally below)
+  temp = s.month + d_month;
+  if (temp <= 0) {
+    e.month = (uint8_t)(12 + modulo(temp, 1, 13));
+    carry   = quotient(temp, 1, 13) - 1;
+  } else {
+    e.month = (uint8_t)modulo(temp, 1, 13);
+    carry   = quotient(temp, 1, 13);
+  }
+
+  // Years (may be modified additionally below)
+  temp = s.year + d_year + carry;
+  if (temp > INT16_MAX) {
+    return infinite_future(s.is_utc);
+  }
+
+  if (temp < INT16_MIN) {
+    return infinite_past(s.is_utc);
+  }
+
+  e.year = (int16_t)temp;
+
+  // Nanoseconds
+  temp = (int32_t)s.nanosecond + d_nanosecond;
+  if (temp < 0) {
+    e.nanosecond = (uint32_t)(1000000000 + (temp % 1000000000));
+    carry        = temp / 1000000000 - 1;
+  } else {
+    e.nanosecond = (uint32_t)(temp % 1000000000);
+    carry        = temp / 1000000000;
+  }
+
+  // Seconds
+  temp = s.second + d_second + carry;
+  if (temp < 0) {
+    e.second = (uint8_t)(60 + (temp % 60));
+    carry    = temp / 60 - 1;
+  } else {
+    e.second = (uint8_t)(temp % 60);
+    carry    = temp / 60;
+  }
+
+  // Minutes
+  temp = s.minute + d_minute + carry;
+  if (temp < 0) {
+    e.minute = (uint8_t)(60 + (temp % 60));
+    carry    = temp / 60 - 1;
+  } else {
+    e.minute = (uint8_t)(temp % 60);
+    carry    = temp / 60;
+  }
+
+  // Hours
+  temp = s.hour + d_hour + carry;
+  if (temp < 0) {
+    e.hour = (uint8_t)(24 + (temp % 24));
+    carry  = temp / 24 - 1;
+  } else {
+    e.hour = (uint8_t)(temp % 24);
+    carry  = temp / 24;
+  }
+
+  /*
+    Carry days into months and years as necessary.  Note that the algorithm in
+    the spec first clamps here, but we don't because no such datetime should
+    exist (exess_read_datetime refuses to read them)
+  */
+  int32_t day = s.day + d_day + carry;
+  while (day < 1 || day > days_in_month(e.year, e.month)) {
+    if (day < 1) {
+      if (e.month == 1) {
+        if (e.year == INT16_MIN) {
+          return infinite_past(s.is_utc);
+        }
+
+        --e.year;
+        e.month = 12;
+        day += days_in_month(e.year, e.month);
+      } else {
+        --e.month;
+        day += days_in_month(e.year, e.month);
+      }
+    } else {
+      day -= days_in_month(e.year, e.month);
+      if (++e.month > 12) {
+        if (e.year == INT16_MAX) {
+          return infinite_future(s.is_utc);
+        }
+
+        ++e.year;
+        e.month = (uint8_t)modulo(e.month, 1, 13);
+      }
+    }
+  }
+
+  e.day = (uint8_t)day;
+
+  return e;
+}
+
+ExessResult
+exess_read_datetime(ExessDateTime* const out, const char* const str)
+{
+  out->year  = 0;
+  out->month = 0;
+  out->day   = 0;
+
+  // Read date
+  ExessDate         date = {0, 0u, 0u, {EXESS_LOCAL}};
+  const ExessResult dr   = read_date_numbers(&date, str);
+  if (dr.status) {
+    return dr;
+  }
+
+  size_t i = dr.count;
+  if (str[i] != 'T') {
+    return result(EXESS_EXPECTED_TIME_SEP, i);
+  }
+
+  ++i;
+
+  // Read time
+  ExessTime         time = {{INT8_MAX}, 0u, 0u, 0u, 0u};
+  const ExessResult tr   = exess_read_time(&time, str + i);
+  if (tr.status) {
+    return result(tr.status, i + tr.count);
+  }
+
+  i += tr.count;
+
+  const ExessDateTime datetime = {date.year,
+                                  date.month,
+                                  date.day,
+                                  time.zone.quarter_hours != EXESS_LOCAL,
+                                  time.hour,
+                                  time.minute,
+                                  time.second,
+                                  time.nanosecond};
+
+  if (datetime.is_utc) {
+    const ExessDuration tz_duration = {
+      0u, -time.zone.quarter_hours * 15 * 60, 0};
+
+    *out = exess_add_datetime_duration(datetime, tz_duration);
+  } else {
+    *out = datetime;
+  }
+
+  return result(EXESS_SUCCESS, i);
+}
+
+ExessResult
+exess_write_datetime(const ExessDateTime value,
+                     const size_t        buf_size,
+                     char* const         buf)
+{
+  const ExessTimezone local = {EXESS_LOCAL};
+  const ExessDate     date  = {value.year, value.month, value.day, local};
+  const ExessTimezone zone  = {value.is_utc ? 0 : EXESS_LOCAL};
+  const ExessTime     time  = {
+    zone, value.hour, value.minute, value.second, value.nanosecond};
+
+  if (!in_range(value.month, 1, 12) || !in_range(value.day, 1, 31) ||
+      !in_range(value.hour, 0, 24) || !in_range(value.minute, 0, 59) ||
+      !in_range(value.second, 0, 59) || value.nanosecond > 999999999) {
+    return end_write(EXESS_BAD_VALUE, buf_size, buf, 0);
+  }
+
+  // Write date
+  ExessResult dr = exess_write_date(date, buf_size, buf);
+  if (dr.status) {
+    return end_write(dr.status, buf_size, buf, dr.count);
+  }
+
+  // Write time delimiter
+  size_t o = dr.count + write_char('T', buf_size, buf, dr.count);
+
+  // Write time with timezone
+  const ExessResult tr = write_time(time, buf_size, buf, o);
+
+  return end_write(tr.status, buf_size, buf, o + tr.count);
+}
diff --git a/subprojects/exess/src/decimal.c b/subprojects/exess/src/decimal.c
new file mode 100644
index 00000000..a8ce8ad5
--- /dev/null
+++ b/subprojects/exess/src/decimal.c
@@ -0,0 +1,267 @@
+/*
+  Copyright 2019-2021 David Robillard <d@drobilla.net>
+
+  Permission to use, copy, modify, and/or distribute this software for any
+  purpose with or without fee is hereby granted, provided that the above
+  copyright notice and this permission notice appear in all copies.
+
+  THIS SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+  WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+  MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+  ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+  WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+  ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+  OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+*/
+
+#include "decimal.h"
+#include "digits.h"
+#include "read_utils.h"
+#include "string_utils.h"
+#include "strtod.h"
+#include "warnings.h"
+#include "write_utils.h"
+
+#include "exess/exess.h"
+
+#include <math.h>
+#include <stdbool.h>
+#include <string.h>
+
+#include <assert.h>
+
+typedef enum {
+  EXESS_POINT_AFTER,   ///< Decimal point is after all significant digits
+  EXESS_POINT_BEFORE,  ///< Decimal point is before all significant digits
+  EXESS_POINT_BETWEEN, ///< Decimal point is between significant digits
+} ExessPointLocation;
+
+typedef struct {
+  ExessPointLocation point_loc;      ///< Location of decimal point
+  unsigned           n_zeros_before; ///< Number of extra zeros before point
+  unsigned           n_zeros_after;  ///< Number of extra zeros after point
+} DecimalMetrics;
+
+static DecimalMetrics
+decimal_metrics(const ExessDecimalDouble count)
+{
+  const int expt =
+    count.expt >= 0 ? (count.expt - (int)count.n_digits + 1) : count.expt;
+
+  DecimalMetrics metrics = {EXESS_POINT_AFTER, 0u, 0u};
+
+  if (count.expt >= (int)count.n_digits - 1) {
+    metrics.point_loc      = EXESS_POINT_AFTER;
+    metrics.n_zeros_before = (unsigned)count.expt - (count.n_digits - 1u);
+    metrics.n_zeros_after  = 1u;
+  } else if (count.expt < 0) {
+    metrics.point_loc      = EXESS_POINT_BEFORE;
+    metrics.n_zeros_before = 1u;
+    metrics.n_zeros_after  = (unsigned)(-expt - 1);
+  } else {
+    metrics.point_loc = EXESS_POINT_BETWEEN;
+  }
+
+  return metrics;
+}
+
+static ExessNumberKind
+number_kind(const double d)
+{
+  EXESS_DISABLE_CONVERSION_WARNINGS
+  const int  fpclass     = fpclassify(d);
+  const bool is_negative = signbit(d);
+  EXESS_RESTORE_WARNINGS
+
+  switch (fpclass) {
+  case FP_ZERO:
+    return is_negative ? EXESS_NEGATIVE_ZERO : EXESS_POSITIVE_ZERO;
+  case FP_INFINITE:
+    return is_negative ? EXESS_NEGATIVE_INFINITY : EXESS_POSITIVE_INFINITY;
+  case FP_NORMAL:
+  case FP_SUBNORMAL:
+    return is_negative ? EXESS_NEGATIVE : EXESS_POSITIVE;
+  default:
+    break;
+  }
+
+  return EXESS_NAN;
+}
+
+ExessDecimalDouble
+exess_measure_decimal(const double d, const unsigned max_precision)
+{
+  ExessDecimalDouble value = {number_kind(d), 0, 0, {0}};
+
+  if (value.kind != EXESS_NEGATIVE && value.kind != EXESS_POSITIVE) {
+    return value;
+  }
+
+  // Get decimal digits
+  const double          abs_d = fabs(d);
+  const ExessDigitCount count =
+    exess_digits(abs_d, value.digits, max_precision);
+
+  assert(count.count == 1 || value.digits[count.count - 1] != '0');
+
+  value.n_digits = count.count;
+  value.expt     = count.expt;
+
+  return value;
+}
+
+ExessDecimalDouble
+exess_measure_float(const float f)
+{
+  return exess_measure_decimal((double)f, FLT_DECIMAL_DIG);
+}
+
+ExessDecimalDouble
+exess_measure_double(const double d)
+{
+  return exess_measure_decimal(d, DBL_DECIMAL_DIG);
+}
+
+static size_t
+exess_decimal_double_string_length(const ExessDecimalDouble decimal)
+{
+  switch (decimal.kind) {
+  case EXESS_NEGATIVE:
+    break;
+  case EXESS_NEGATIVE_INFINITY:
+    return 0;
+  case EXESS_NEGATIVE_ZERO:
+    return 4;
+  case EXESS_POSITIVE_ZERO:
+    return 3;
+  case EXESS_POSITIVE:
+    break;
+  case EXESS_POSITIVE_INFINITY:
+  case EXESS_NAN:
+    return 0;
+  }
+
+  const DecimalMetrics metrics = decimal_metrics(decimal);
+  const unsigned       n_zeros = metrics.n_zeros_before + metrics.n_zeros_after;
+  const bool           is_negative = decimal.kind == EXESS_NEGATIVE;
+
+  return is_negative + decimal.n_digits + 1 + n_zeros;
+}
+
+static size_t
+copy_digits(char* const dest, const char* const src, const size_t n)
+{
+  memcpy(dest, src, n);
+  return n;
+}
+
+static size_t
+set_zeros(char* const dest, const size_t n)
+{
+  memset(dest, '0', n);
+  return n;
+}
+
+static ExessResult
+read_decimal_number(double* const out, const char* const str)
+{
+  *out = (double)NAN;
+
+  if (str[0] == '+' || str[0] == '-') {
+    if (str[1] != '.' && !is_digit(str[1])) {
+      return result(EXESS_EXPECTED_DIGIT, 1);
+    }
+  } else if (str[0] != '.' && !is_digit(str[0])) {
+    return result(EXESS_EXPECTED_DIGIT, 0);
+  }
+
+  const size_t       i  = skip_whitespace(str);
+  ExessDecimalDouble in = {EXESS_NAN, 0u, 0, {0}};
+  const ExessResult  r  = parse_decimal(&in, str + i);
+
+  if (!r.status) {
+    *out = parsed_double_to_double(in);
+  }
+
+  return result(r.status, i + r.count);
+}
+
+ExessResult
+exess_read_decimal(double* const out, const char* const str)
+{
+  const size_t      i = skip_whitespace(str);
+  const ExessResult r = read_decimal_number(out, str + i);
+
+  return end_read(r.status, str, i + r.count);
+}
+
+ExessResult
+exess_write_decimal_double(const ExessDecimalDouble decimal,
+                           const size_t             buf_size,
+                           char* const              buf)
+{
+  if (!buf) {
+    return result(EXESS_SUCCESS, exess_decimal_double_string_length(decimal));
+  }
+
+  size_t i = 0;
+  if (buf_size < 3) {
+    return end_write(EXESS_NO_SPACE, buf_size, buf, 0);
+  }
+
+  switch (decimal.kind) {
+  case EXESS_NEGATIVE:
+    buf[i++] = '-';
+    break;
+  case EXESS_NEGATIVE_INFINITY:
+    return end_write(EXESS_BAD_VALUE, buf_size, buf, 0);
+  case EXESS_NEGATIVE_ZERO:
+    return write_special(4, "-0.0", buf_size, buf);
+  case EXESS_POSITIVE_ZERO:
+    return write_special(3, "0.0", buf_size, buf);
+  case EXESS_POSITIVE:
+    break;
+  case EXESS_POSITIVE_INFINITY:
+  case EXESS_NAN:
+    return end_write(EXESS_BAD_VALUE, buf_size, buf, 0);
+  }
+
+  const DecimalMetrics metrics = decimal_metrics(decimal);
+  const unsigned       n_zeros = metrics.n_zeros_before + metrics.n_zeros_after;
+  if (buf_size - i <= decimal.n_digits + 1 + n_zeros) {
+    return end_write(EXESS_NO_SPACE, buf_size, buf, 0);
+  }
+
+  if (metrics.point_loc == EXESS_POINT_AFTER) {
+    i += copy_digits(buf + i, decimal.digits, decimal.n_digits);
+    i += set_zeros(buf + i, metrics.n_zeros_before);
+    buf[i++] = '.';
+    buf[i++] = '0';
+  } else if (metrics.point_loc == EXESS_POINT_BEFORE) {
+    buf[i++] = '0';
+    buf[i++] = '.';
+    i += set_zeros(buf + i, metrics.n_zeros_after);
+    i += copy_digits(buf + i, decimal.digits, decimal.n_digits);
+  } else {
+    assert(metrics.point_loc == EXESS_POINT_BETWEEN);
+    assert(decimal.expt >= -1);
+
+    const size_t n_before = (size_t)decimal.expt + 1u;
+    const size_t n_after  = decimal.n_digits - n_before;
+
+    i += copy_digits(buf + i, decimal.digits, n_before);
+    buf[i++] = '.';
+    memcpy(buf + i, decimal.digits + n_before, n_after);
+    i += n_after;
+  }
+
+  return end_write(EXESS_SUCCESS, buf_size, buf, i);
+}
+
+ExessResult
+exess_write_decimal(const double value, const size_t n, char* const buf)
+{
+  const ExessDecimalDouble decimal = exess_measure_double(value);
+
+  return exess_write_decimal_double(decimal, n, buf);
+}
diff --git a/subprojects/exess/src/decimal.h b/subprojects/exess/src/decimal.h
new file mode 100644
index 00000000..7c5aa963
--- /dev/null
+++ b/subprojects/exess/src/decimal.h
@@ -0,0 +1,63 @@
+/*
+  Copyright 2019-2021 David Robillard <d@drobilla.net>
+
+  Permission to use, copy, modify, and/or distribute this software for any
+  purpose with or without fee is hereby granted, provided that the above
+  copyright notice and this permission notice appear in all copies.
+
+  THIS SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+  WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+  MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+  ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+  WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+  ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+  OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+*/
+
+#ifndef EXESS_DECIMAL_H
+#define EXESS_DECIMAL_H
+
+#include "exess/exess.h"
+
+#include <stddef.h>
+
+// Define C11 numeric constants if the compiler hasn't already
+#ifndef FLT_DECIMAL_DIG
+#  define FLT_DECIMAL_DIG 9
+#endif
+#ifndef DBL_DECIMAL_DIG
+#  define DBL_DECIMAL_DIG 17
+#endif
+
+typedef enum {
+  EXESS_NEGATIVE,
+  EXESS_NEGATIVE_INFINITY,
+  EXESS_NEGATIVE_ZERO,
+  EXESS_POSITIVE_ZERO,
+  EXESS_POSITIVE,
+  EXESS_POSITIVE_INFINITY,
+  EXESS_NAN,
+} ExessNumberKind;
+
+typedef struct {
+  ExessNumberKind kind;                        ///< Kind of number
+  int             expt;                        ///< Power of 10 exponent
+  unsigned        n_digits;                    ///< Number of significant digits
+  char            digits[DBL_DECIMAL_DIG + 2]; ///< Significant digits
+} ExessDecimalDouble;
+
+ExessDecimalDouble
+exess_measure_decimal(double d, unsigned max_precision);
+
+ExessDecimalDouble
+exess_measure_float(float f);
+
+ExessDecimalDouble
+exess_measure_double(double d);
+
+ExessResult
+exess_write_decimal_double(ExessDecimalDouble decimal,
+                           size_t             buf_size,
+                           char*              buf);
+
+#endif // EXESS_DECIMAL_H
diff --git a/subprojects/exess/src/digits.c b/subprojects/exess/src/digits.c
new file mode 100644
index 00000000..dd303269
--- /dev/null
+++ b/subprojects/exess/src/digits.c
@@ -0,0 +1,243 @@
+/*
+  Copyright 2019-2021 David Robillard <d@drobilla.net>
+
+  Permission to use, copy, modify, and/or distribute this software for any
+  purpose with or without fee is hereby granted, provided that the above
+  copyright notice and this permission notice appear in all copies.
+
+  THIS SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+  WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+  MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+  ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+  WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+  ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+  OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+*/
+
+#include "digits.h"
+
+#include "bigint.h"
+#include "ieee_float.h"
+#include "soft_float.h"
+#include "warnings.h"
+
+#include <assert.h>
+#include <math.h>
+#include <stdbool.h>
+#include <stddef.h>
+#include <stdint.h>
+
+/*
+  This is more or less just an implementation of the classic rational number
+  based floating point print routine ("Dragon4").  See "How to Print
+  Floating-Point Numbers Accurately" by Guy L. Steele Jr. and Jon L White for
+  the canonical source.  The basic idea is to find a big rational between 1 and
+  10 where value = (numer / denom) * 10^e, then continuously divide it to
+  generate decimal digits.
+
+  Unfortunately, this algorithm requires pretty massive bigints to work
+  correctly for all doubles, and isn't particularly fast.  Something like
+  Grisu3 could be added to improve performance, but that has the annoying
+  property of needing a more precise fallback in some cases, meaning it would
+  only add more code, not replace any.  Since this is already a pretty
+  ridiculous amount of code, I'll hold off on this until it becomes a problem,
+  or somebody comes up with a better algorithm.
+*/
+
+/// Return true if the number is within the lower boundary
+static bool
+within_lower(const ExessBigint* const numer,
+             const ExessBigint* const d_lower,
+             const bool               is_even)
+{
+  return is_even ? exess_bigint_compare(numer, d_lower) <= 0
+                 : exess_bigint_compare(numer, d_lower) < 0;
+}
+
+/// Return true if the number is within the upper boundary
+static bool
+within_upper(const ExessBigint* const numer,
+             const ExessBigint* const denom,
+             const ExessBigint* const d_upper,
+             const bool               is_even)
+{
+  return is_even ? exess_bigint_plus_compare(numer, d_upper, denom) >= 0
+                 : exess_bigint_plus_compare(numer, d_upper, denom) > 0;
+}
+
+/**
+   Find values so that 0.1 <= numer/denom < 1 or 1 <= numer/denom < 10.
+
+   @param significand Double significand.
+   @param exponent Double exponent (base 2).
+   @param decimal_power Decimal exponent (log10 of the double).
+   @param[out] numer Numerator of rational number.
+   @param[out] denom Denominator of rational number.
+   @param[out] delta Distance to the lower and upper boundaries.
+*/
+static void
+calculate_initial_values(const uint64_t     significand,
+                         const int          exponent,
+                         const int          decimal_power,
+                         const bool         lower_is_closer,
+                         ExessBigint* const numer,
+                         ExessBigint* const denom,
+                         ExessBigint* const delta)
+{
+  /* Use a common denominator of 2^1 so that boundary distance is an integer.
+     If the lower boundary is closer, we need to scale everything but the
+     lower boundary to compensate, so add another factor of two here (this is
+     faster than shifting them again later as in the paper). */
+  const unsigned lg_denom = 1u + lower_is_closer;
+
+  if (exponent >= 0) {
+    // delta = 2^e
+    exess_bigint_set_u32(delta, 1);
+    exess_bigint_shift_left(delta, (unsigned)exponent);
+
+    // numer = f * 2^e
+    exess_bigint_set_u64(numer, significand);
+    exess_bigint_shift_left(numer, (unsigned)exponent + lg_denom);
+
+    // denom = 10^d
+    exess_bigint_set_pow10(denom, (unsigned)decimal_power);
+    exess_bigint_shift_left(denom, lg_denom);
+  } else if (decimal_power >= 0) {
+    // delta = 2^e, which is just 1 here since 2^-e is in the denominator
+    exess_bigint_set_u32(delta, 1);
+
+    // numer = f
+    exess_bigint_set_u64(numer, significand);
+    exess_bigint_shift_left(numer, lg_denom);
+
+    // denom = 10^d * 2^-e
+    exess_bigint_set_pow10(denom, (unsigned)decimal_power);
+    exess_bigint_shift_left(denom, (unsigned)-exponent + lg_denom);
+  } else {
+    // delta = 10^d
+    exess_bigint_set_pow10(delta, (unsigned)-decimal_power);
+
+    // numer = f * 10^-d
+    exess_bigint_set(numer, delta);
+    exess_bigint_multiply_u64(numer, significand);
+    exess_bigint_shift_left(numer, lg_denom);
+
+    // denom = 2^-exponent
+    exess_bigint_set_u32(denom, 1);
+    exess_bigint_shift_left(denom, (unsigned)-exponent + lg_denom);
+  }
+}
+
+#ifndef NDEBUG
+static bool
+check_initial_values(const ExessBigint* const numer,
+                     const ExessBigint* const denom,
+                     const ExessBigint* const d_upper)
+{
+  ExessBigint upper = *numer;
+  exess_bigint_add(&upper, d_upper);
+  assert(exess_bigint_compare(&upper, denom) >= 0);
+
+  const uint32_t div = exess_bigint_divmod(&upper, denom);
+  assert(div >= 1 && div < 10);
+  return true;
+}
+#endif
+
+static unsigned
+emit_digits(ExessBigint* const       numer,
+            const ExessBigint* const denom,
+            ExessBigint* const       d_lower,
+            ExessBigint* const       d_upper,
+            const bool               is_even,
+            char* const              buffer,
+            const size_t             max_digits)
+{
+  unsigned length = 0;
+  for (size_t i = 0; i < max_digits; ++i) {
+    // Emit the next digit
+    const uint32_t digit = exess_bigint_divmod(numer, denom);
+    assert(digit <= 9);
+    buffer[length++] = (char)('0' + digit);
+
+    // Check for termination
+    const bool within_low  = within_lower(numer, d_lower, is_even);
+    const bool within_high = within_upper(numer, denom, d_upper, is_even);
+    if (!within_low && !within_high) {
+      exess_bigint_multiply_u32(numer, 10);
+      exess_bigint_multiply_u32(d_lower, 10);
+      if (d_lower != d_upper) {
+        exess_bigint_multiply_u32(d_upper, 10);
+      }
+    } else {
+      if (!within_low || (within_high && exess_bigint_plus_compare(
+                                           numer, numer, denom) >= 0)) {
+        // In high only, or halfway and the next digit is > 5, round up
+        assert(buffer[length - 1] != '9');
+        buffer[length - 1]++;
+      }
+
+      break;
+    }
+  }
+
+  return length;
+}
+
+ExessDigitCount
+exess_digits(const double d, char* const buf, const unsigned max_digits)
+{
+  EXESS_DISABLE_CONVERSION_WARNINGS
+  assert(isfinite(d) && fpclassify(d) != FP_ZERO);
+  EXESS_RESTORE_WARNINGS
+
+  const ExessSoftFloat value           = soft_float_from_double(d);
+  const int            power           = (int)lrint(log10(d));
+  const bool           is_even         = !(value.f & 1u);
+  const bool           lower_is_closer = double_lower_boundary_is_closer(d);
+
+  // Calculate initial values so that v = (numer / denom) * 10^power
+  ExessBigint numer;
+  ExessBigint denom;
+  ExessBigint d_lower;
+  calculate_initial_values(
+    value.f, value.e, power, lower_is_closer, &numer, &denom, &d_lower);
+
+  ExessBigint  d_upper_storage;
+  ExessBigint* d_upper = NULL;
+  if (lower_is_closer) {
+    // Scale upper boundary to account for the closer lower boundary
+    // (the numerator and denominator were already scaled above)
+    d_upper_storage = d_lower;
+    d_upper         = &d_upper_storage;
+    exess_bigint_shift_left(d_upper, 1);
+  } else {
+    d_upper = &d_lower; // Boundaries are the same, reuse the lower
+  }
+
+  // Scale if necessary to make 1 <= (numer + delta) / denom < 10
+  ExessDigitCount count = {0, 0};
+  if (within_upper(&numer, &denom, d_upper, is_even)) {
+    count.expt = power;
+  } else {
+    count.expt = power - 1;
+    exess_bigint_multiply_u32(&numer, 10);
+    exess_bigint_multiply_u32(&d_lower, 10);
+    if (d_upper != &d_lower) {
+      exess_bigint_multiply_u32(d_upper, 10);
+    }
+  }
+
+  // Write digits to output
+  assert(check_initial_values(&numer, &denom, d_upper));
+  count.count =
+    emit_digits(&numer, &denom, &d_lower, d_upper, is_even, buf, max_digits);
+
+  // Trim trailing zeros
+  while (count.count > 1 && buf[count.count - 1] == '0') {
+    buf[--count.count] = 0;
+  }
+
+  buf[count.count] = '\0';
+  return count;
+}
diff --git a/subprojects/exess/src/digits.h b/subprojects/exess/src/digits.h
new file mode 100644
index 00000000..0753a0af
--- /dev/null
+++ b/subprojects/exess/src/digits.h
@@ -0,0 +1,38 @@
+/*
+  Copyright 2019-2021 David Robillard <d@drobilla.net>
+
+  Permission to use, copy, modify, and/or distribute this software for any
+  purpose with or without fee is hereby granted, provided that the above
+  copyright notice and this permission notice appear in all copies.
+
+  THIS SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+  WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+  MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+  ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+  WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+  ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+  OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+*/
+
+#ifndef EXESS_DIGITS_H
+#define EXESS_DIGITS_H
+
+typedef struct {
+  unsigned count; ///< Number of digits
+  int      expt;  ///< Power of 10 exponent
+} ExessDigitCount;
+
+/**
+   Write significant digits digits for `d` into `buf`.
+
+   Writes only significant digits, without any leading or trailing zeros.  The
+   actual number is given by the exponent in the return value.
+
+   @param d The number to convert to digits, must be finite and non-zero.
+   @param buf The output buffer at least `max_digits` long.
+   @param max_digits The maximum number of digits to write.
+*/
+ExessDigitCount
+exess_digits(double d, char* buf, unsigned max_digits);
+
+#endif // EXESS_DIGITS_H
diff --git a/subprojects/exess/src/double.c b/subprojects/exess/src/double.c
new file mode 100644
index 00000000..2934010d
--- /dev/null
+++ b/subprojects/exess/src/double.c
@@ -0,0 +1,53 @@
+/*
+  Copyright 2019-2021 David Robillard <d@drobilla.net>
+
+  Permission to use, copy, modify, and/or distribute this software for any
+  purpose with or without fee is hereby granted, provided that the above
+  copyright notice and this permission notice appear in all copies.
+
+  THIS SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+  WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+  MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+  ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+  WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+  ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+  OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+*/
+
+#include "decimal.h"
+#include "read_utils.h"
+#include "scientific.h"
+#include "strtod.h"
+#include "write_utils.h"
+
+#include "exess/exess.h"
+
+#include <math.h>
+#include <string.h>
+
+ExessResult
+exess_read_double(double* const out, const char* const str)
+{
+  *out = (double)NAN;
+
+  const size_t       i  = skip_whitespace(str);
+  ExessDecimalDouble in = {EXESS_NAN, 0u, 0, {0}};
+  const ExessResult  r  = parse_double(&in, str + i);
+
+  if (!r.status) {
+    *out = parsed_double_to_double(in);
+  }
+
+  return result(r.status, i + r.count);
+}
+
+ExessResult
+exess_write_double(const double value, const size_t buf_size, char* const buf)
+{
+  const ExessDecimalDouble decimal = exess_measure_double(value);
+  const ExessResult        r =
+    buf ? exess_write_scientific(decimal, buf_size, buf)
+               : result(EXESS_SUCCESS, exess_scientific_string_length(decimal));
+
+  return end_write(r.status, buf_size, buf, r.count);
+}
diff --git a/subprojects/exess/src/duration.c b/subprojects/exess/src/duration.c
new file mode 100644
index 00000000..0a753fd3
--- /dev/null
+++ b/subprojects/exess/src/duration.c
@@ -0,0 +1,322 @@
+/*
+  Copyright 2019-2021 David Robillard <d@drobilla.net>
+
+  Permission to use, copy, modify, and/or distribute this software for any
+  purpose with or without fee is hereby granted, provided that the above
+  copyright notice and this permission notice appear in all copies.
+
+  THIS SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+  WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+  MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+  ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+  WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+  ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+  OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+*/
+
+#include "read_utils.h"
+#include "string_utils.h"
+#include "time_utils.h"
+#include "write_utils.h"
+
+#include "exess/exess.h"
+
+#include <math.h>
+#include <stdbool.h>
+#include <stdint.h>
+#include <stdlib.h>
+#include <string.h>
+
+typedef enum { YEAR, MONTH, DAY, HOUR, MINUTE, SECOND } Field;
+
+static ExessStatus
+set_field(ExessDuration* const out,
+          const Field          current_field,
+          const Field          field,
+          const uint32_t       value)
+{
+  if (value >= INT32_MAX) {
+    return EXESS_OUT_OF_RANGE;
+  }
+
+  if (field < current_field) {
+    return EXESS_BAD_ORDER;
+  }
+
+  switch (field) {
+  case YEAR:
+    out->months = (int32_t)(12 * lrint(value));
+    break;
+  case MONTH:
+    out->months = (int32_t)(out->months + lrint(value));
+    break;
+  case DAY:
+    out->seconds = (int32_t)(24 * 60 * 60 * lrint(value));
+    break;
+  case HOUR:
+    out->seconds = (int32_t)(out->seconds + 60 * 60 * lrint(value));
+    break;
+  case MINUTE:
+    out->seconds = (int32_t)(out->seconds + 60 * lrint(value));
+    break;
+  case SECOND:
+    out->seconds = (int32_t)(out->seconds + lrint(value));
+    break;
+  }
+
+  return EXESS_SUCCESS;
+}
+
+static ExessResult
+read_date(ExessDuration* const out, const Field field, const char* const str)
+{
+  uint32_t    value = 0;
+  ExessResult r     = exess_read_uint(&value, str);
+  if (r.status > EXESS_EXPECTED_END) {
+    return r;
+  }
+
+  size_t i = r.count;
+  switch (str[i]) {
+  case 'Y':
+    if ((r.status = set_field(out, field, YEAR, value))) {
+      return r;
+    }
+
+    ++i;
+    if (str[i] != 'T' && !is_end(str[i])) {
+      r = read_date(out, MONTH, str + i);
+      i += r.count;
+    }
+    break;
+
+  case 'M':
+    if ((r.status = set_field(out, field, MONTH, value))) {
+      return r;
+    }
+
+    ++i;
+    if (str[i] != 'T' && !is_end(str[i])) {
+      r = read_date(out, DAY, str + i);
+      i += r.count;
+    }
+    break;
+
+  case 'D':
+    if ((r.status = set_field(out, field, DAY, value))) {
+      return r;
+    }
+
+    ++i;
+    break;
+
+  default:
+    return result(EXESS_EXPECTED_DATE_TAG, i);
+  }
+
+  return result(r.status, i);
+}
+
+static ExessResult
+read_time(ExessDuration* const out, const Field field, const char* const str)
+{
+  uint32_t    value = 0;
+  ExessResult r     = exess_read_uint(&value, str);
+  if (r.status > EXESS_EXPECTED_END) {
+    return r;
+  }
+
+  size_t      i    = r.count;
+  ExessResult next = {EXESS_SUCCESS, 0};
+  switch (str[i]) {
+  case '.': {
+    if (!is_digit(str[++i])) {
+      return result(EXESS_EXPECTED_DIGIT, i);
+    }
+
+    uint32_t nanoseconds = 0;
+
+    r = read_nanoseconds(&nanoseconds, str + i);
+    i += r.count;
+
+    if (str[i] != 'S') {
+      return result(EXESS_EXPECTED_TIME_TAG, i);
+    }
+
+    r.status         = set_field(out, field, SECOND, value);
+    out->nanoseconds = (int32_t)nanoseconds;
+
+    break;
+  }
+
+  case 'H':
+    r.status = set_field(out, field, HOUR, value);
+    if (!is_end(str[i + 1])) {
+      next = read_time(out, MINUTE, str + i + 1);
+    }
+    break;
+
+  case 'M':
+    r.status = set_field(out, field, MINUTE, value);
+    if (!is_end(str[i + 1])) {
+      next = read_time(out, SECOND, str + i + 1);
+    }
+    break;
+
+  case 'S':
+    r.status = set_field(out, field, SECOND, value);
+    break;
+
+  default:
+    return result(EXESS_EXPECTED_TIME_TAG, i);
+  }
+
+  if (r.status) {
+    return r;
+  }
+
+  return result(next.status, i + 1 + next.count);
+}
+
+ExessResult
+exess_read_duration(ExessDuration* const out, const char* const str)
+{
+  memset(out, 0, sizeof(*out));
+
+  size_t i           = skip_whitespace(str);
+  bool   is_negative = false;
+  if (str[i] == '-') {
+    is_negative = true;
+    ++i;
+  }
+
+  if (str[i] != 'P') {
+    return result(EXESS_EXPECTED_DURATION, i);
+  }
+
+  ++i;
+
+  if (str[i] != 'T') {
+    ExessResult r = read_date(out, YEAR, str + i);
+    if (r.status) {
+      return result(r.status, i + r.count);
+    }
+
+    i += r.count;
+
+    if (!is_end(str[i]) && str[i] != 'T') {
+      return result(EXESS_EXPECTED_TIME_SEP, i);
+    }
+  }
+
+  if (str[i] == 'T') {
+    ++i;
+
+    ExessResult r = read_time(out, HOUR, str + i);
+    if (r.status) {
+      return result(r.status, i + r.count);
+    }
+
+    i += r.count;
+  }
+
+  if (is_negative) {
+    out->months      = -out->months;
+    out->seconds     = -out->seconds;
+    out->nanoseconds = -out->nanoseconds;
+  }
+
+  return end_read(EXESS_SUCCESS, str, i);
+}
+
+static size_t
+write_int_field(ExessResult*   r,
+                const uint32_t value,
+                const char     tag,
+                const size_t   buf_size,
+                char* const    buf,
+                const size_t   i)
+{
+  if (!r->status) {
+    if (value == 0) {
+      *r = result(EXESS_SUCCESS, 0);
+    } else if (!buf) {
+      *r = exess_write_uint(value, buf_size, buf);
+      ++r->count;
+    } else {
+      *r = exess_write_uint(value, buf_size - i, buf + i);
+      if (!r->status) {
+        buf[i + r->count++] = tag;
+      }
+    }
+  }
+
+  return r->count;
+}
+
+ExessResult
+exess_write_duration(const ExessDuration value,
+                     const size_t        buf_size,
+                     char* const         buf)
+{
+  // Write zero as a special case
+  size_t i = 0;
+  if (value.months == 0 && value.seconds == 0 && value.nanoseconds == 0) {
+    i += write_string(3, "P0Y", buf_size, buf, i);
+    return end_write(EXESS_SUCCESS, buf_size, buf, i);
+  }
+
+  if (value.months == INT32_MIN || value.seconds == INT32_MIN) {
+    return end_write(EXESS_OUT_OF_RANGE, buf_size, buf, 0);
+  }
+
+  const bool is_negative =
+    (value.months < 0 || value.seconds < 0 || value.nanoseconds < 0);
+
+  if (is_negative &&
+      (value.months > 0 || value.seconds > 0 || value.nanoseconds > 0)) {
+    return end_write(EXESS_BAD_VALUE, buf_size, buf, 0);
+  }
+
+  // Write duration prefix
+  if (value.months < 0 || value.seconds < 0 || value.nanoseconds < 0) {
+    i += write_string(2, "-P", buf_size, buf, i);
+  } else {
+    i += write_char('P', buf_size, buf, i);
+  }
+
+  const uint32_t abs_years   = (uint32_t)(abs(value.months) / 12);
+  const uint32_t abs_months  = (uint32_t)(abs(value.months) % 12);
+  const uint32_t abs_days    = (uint32_t)(abs(value.seconds) / (24 * 60 * 60));
+  const uint32_t abs_hours   = (uint32_t)(abs(value.seconds) / 60 / 60 % 24);
+  const uint32_t abs_minutes = (uint32_t)(abs(value.seconds) / 60 % 60);
+  const uint8_t  abs_seconds = (uint8_t)(abs(value.seconds) % 60);
+  const uint32_t abs_nanoseconds = (uint32_t)abs(value.nanoseconds);
+
+  // Write date segments if present
+  ExessResult r = result(EXESS_SUCCESS, 0);
+  i += write_int_field(&r, abs_years, 'Y', buf_size, buf, i);
+  i += write_int_field(&r, abs_months, 'M', buf_size, buf, i);
+  i += write_int_field(&r, abs_days, 'D', buf_size, buf, i);
+
+  // Write time segments if present
+  const bool has_time = abs_hours + abs_minutes + abs_seconds + abs_nanoseconds;
+  if (has_time && !r.status) {
+    i += write_char('T', buf_size, buf, i);
+    i += write_int_field(&r, abs_hours, 'H', buf_size, buf, i);
+    i += write_int_field(&r, abs_minutes, 'M', buf_size, buf, i);
+
+    if (abs_seconds != 0 || abs_nanoseconds != 0) {
+      r = write_digits(abs_seconds, buf_size, buf, i);
+      i += r.count;
+
+      if (!r.status && abs_nanoseconds > 0) {
+        i += write_nanoseconds(abs_nanoseconds, buf_size, buf, i);
+      }
+
+      i += write_char('S', buf_size, buf, i);
+    }
+  }
+
+  return end_write(r.status, buf_size, buf, i);
+}
diff --git a/subprojects/exess/src/exess_config.h b/subprojects/exess/src/exess_config.h
new file mode 100644
index 00000000..4325484a
--- /dev/null
+++ b/subprojects/exess/src/exess_config.h
@@ -0,0 +1,79 @@
+/*
+  Copyright 2021 David Robillard <d@drobilla.net>
+
+  Permission to use, copy, modify, and/or distribute this software for any
+  purpose with or without fee is hereby granted, provided that the above
+  copyright notice and this permission notice appear in all copies.
+
+  THIS SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+  WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+  MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+  ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+  WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+  ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+  OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+*/
+
+/*
+  Configuration header that defines reasonable defaults at compile time.
+
+  This allows compile-time configuration from the command line (typically via
+  the build system) while still allowing the source to be built without any
+  configuration.  The build system can define EXESS_NO_DEFAULT_CONFIG to disable
+  defaults, in which case it must define things like HAVE_FEATURE to enable
+  features.  The design here ensures that compiler warnings or
+  include-what-you-use will catch any mistakes.
+*/
+
+#ifndef EXESS_CONFIG_H
+#define EXESS_CONFIG_H
+
+// Define version unconditionally so a warning will catch a mismatch
+#define EXESS_VERSION "0.0.1"
+
+#if !defined(EXESS_NO_DEFAULT_CONFIG)
+
+// GCC and clang: __builtin_clz()
+#  ifndef HAVE_BUILTIN_CLZ
+#    if defined(__has_builtin)
+#      if __has_builtin(__builtin_clz)
+#        define HAVE_BUILTIN_CLZ 1
+#      else
+#        define HAVE_BUILTIN_CLZ 0
+#      endif
+#    elif defined(__GNUC__)
+#      define HAVE_BUILTIN_CLZ 1
+#    else
+#      define HAVE_BUILTIN_CLZ 0
+#    endif
+#  endif
+
+// GCC and clang: __builtin_clz()
+#  ifndef HAVE_BUILTIN_CLZLL
+#    if defined(__has_builtin)
+#      if __has_builtin(__builtin_clzll)
+#        define HAVE_BUILTIN_CLZLL 1
+#      else
+#        define HAVE_BUILTIN_CLZLL 0
+#      endif
+#    elif defined(__GNUC__)
+#      define HAVE_BUILTIN_CLZLL 1
+#    else
+#      define HAVE_BUILTIN_CLZLL 0
+#    endif
+#  endif
+
+#endif // !defined(EXESS_NO_DEFAULT_CONFIG)
+
+/*
+  Make corresponding USE_FEATURE defines based on the HAVE_FEATURE defines from
+  above or the command line.  The code checks for these using #if (not #ifdef),
+  so there will be an undefined warning if it checks for an unknown feature,
+  and this header is always required by any code that checks for features, even
+  if the build system defines them all.
+*/
+
+#define USE_BUILTIN_CLZ HAVE_BUILTIN_CLZ
+#define USE_BUILTIN_CLZLL HAVE_BUILTIN_CLZLL
+
+#endif // EXESS_CONFIG_H
diff --git a/subprojects/exess/src/float.c b/subprojects/exess/src/float.c
new file mode 100644
index 00000000..bcc6d49a
--- /dev/null
+++ b/subprojects/exess/src/float.c
@@ -0,0 +1,44 @@
+/*
+  Copyright 2019-2021 David Robillard <d@drobilla.net>
+
+  Permission to use, copy, modify, and/or distribute this software for any
+  purpose with or without fee is hereby granted, provided that the above
+  copyright notice and this permission notice appear in all copies.
+
+  THIS SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+  WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+  MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+  ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+  WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+  ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+  OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+*/
+
+#include "decimal.h"
+#include "read_utils.h"
+#include "scientific.h"
+
+#include "exess/exess.h"
+
+#include <math.h>
+#include <string.h>
+
+ExessResult
+exess_read_float(float* const out, const char* const str)
+{
+  double            value = (double)NAN;
+  const ExessResult r     = exess_read_double(&value, str);
+
+  *out = (float)value;
+
+  return r;
+}
+
+ExessResult
+exess_write_float(const float value, const size_t buf_size, char* const buf)
+{
+  const ExessDecimalDouble decimal = exess_measure_float(value);
+
+  return buf ? exess_write_scientific(decimal, buf_size, buf)
+             : result(EXESS_SUCCESS, exess_scientific_string_length(decimal));
+}
diff --git a/subprojects/exess/src/hex.c b/subprojects/exess/src/hex.c
new file mode 100644
index 00000000..bd9de4d6
--- /dev/null
+++ b/subprojects/exess/src/hex.c
@@ -0,0 +1,131 @@
+/*
+  Copyright 2011-2021 David Robillard <d@drobilla.net>
+
+  Permission to use, copy, modify, and/or distribute this software for any
+  purpose with or without fee is hereby granted, provided that the above
+  copyright notice and this permission notice appear in all copies.
+
+  THIS SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+  WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+  MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+  ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+  WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+  ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+  OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+*/
+
+#include "read_utils.h"
+#include "string_utils.h"
+#include "write_utils.h"
+
+#include "exess/exess.h"
+
+#include <stdint.h>
+#include <string.h>
+
+/// Hex encoding table
+static const char hex_map[] = "0123456789ABCDEF";
+
+static inline uint8_t
+decode_nibble(const char c)
+{
+  if (is_digit(c)) {
+    return (uint8_t)(c - '0');
+  }
+
+  if (c >= 'A' && c <= 'F') {
+    return (uint8_t)(10 + c - 'A');
+  }
+
+  if (c >= 'a' && c <= 'f') {
+    return (uint8_t)(10 + c - 'a');
+  }
+
+  return UINT8_MAX;
+}
+
+static char
+next_char(const char* const str, size_t* const i)
+{
+  *i += skip_whitespace(str + *i);
+
+  return str[*i];
+}
+
+size_t
+exess_hex_decoded_size(const size_t length)
+{
+  return length / 2;
+}
+
+ExessResult
+exess_read_hex(ExessBlob* const out, const char* const str)
+{
+  uint8_t* const uout = (uint8_t*)out->data;
+  size_t         i    = 0u;
+  size_t         o    = 0u;
+
+  while (str[i]) {
+    const char hi_char = next_char(str, &i);
+    if (!hi_char) {
+      break;
+    }
+
+    ++i;
+
+    const uint8_t hi = decode_nibble(hi_char);
+    if (hi == UINT8_MAX) {
+      return result(EXESS_EXPECTED_HEX, i);
+    }
+
+    const char lo_char = next_char(str, &i);
+    if (!lo_char) {
+      return result(EXESS_EXPECTED_HEX, i);
+    }
+
+    ++i;
+
+    const uint8_t lo = decode_nibble(lo_char);
+    if (lo == UINT8_MAX) {
+      return result(EXESS_EXPECTED_HEX, i);
+    }
+
+    if (o >= out->size) {
+      return result(EXESS_NO_SPACE, i);
+    }
+
+    uout[o++] = (uint8_t)(hi << 4u) | lo;
+  }
+
+  out->size = o;
+  return result(EXESS_SUCCESS, i);
+}
+
+ExessResult
+exess_write_hex(const size_t      data_size,
+                const void* const data,
+                const size_t      buf_size,
+                char* const       buf)
+{
+  const size_t length = 2 * data_size;
+  if (!buf) {
+    return result(EXESS_SUCCESS, length);
+  }
+
+  if (buf_size < length + 1) {
+    return result(EXESS_NO_SPACE, 0);
+  }
+
+  const uint8_t* const in = (const uint8_t*)data;
+  size_t               o  = 0u;
+
+  for (size_t i = 0; i < data_size; ++i) {
+    const uint8_t hi = (in[i] & 0xF0u) >> 4u;
+    const uint8_t lo = (in[i] & 0x0Fu);
+
+    buf[o++] = hex_map[hi];
+    buf[o++] = hex_map[lo];
+  }
+
+  return end_write(EXESS_SUCCESS, buf_size, buf, o);
+}
diff --git a/subprojects/exess/src/ieee_float.h b/subprojects/exess/src/ieee_float.h
new file mode 100644
index 00000000..5c70bf4b
--- /dev/null
+++ b/subprojects/exess/src/ieee_float.h
@@ -0,0 +1,63 @@
+/*
+  Copyright 2019-2021 David Robillard <d@drobilla.net>
+
+  Permission to use, copy, modify, and/or distribute this software for any
+  purpose with or without fee is hereby granted, provided that the above
+  copyright notice and this permission notice appear in all copies.
+
+  THIS SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+  WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+  MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+  ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+  WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+  ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+  OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+*/
+
+#ifndef EXESS_IEEE_FLOAT_H
+#define EXESS_IEEE_FLOAT_H
+
+#include <float.h>
+#include <stdbool.h>
+#include <stdint.h>
+#include <string.h>
+
+static const unsigned dbl_physical_mant_dig = DBL_MANT_DIG - 1u;
+static const uint64_t dbl_mant_mask         = 0x000FFFFFFFFFFFFFull;
+static const uint64_t dbl_expt_mask         = 0x7FF0000000000000ul;
+static const uint64_t dbl_hidden_bit        = 0x0010000000000000ul;
+static const int      dbl_expt_bias         = 0x3FF + DBL_MANT_DIG - 1;
+static const int      dbl_subnormal_expt    = -0x3FF - DBL_MANT_DIG + 2;
+
+/// Return the raw representation of a float
+static inline uint32_t
+float_to_rep(const float d)
+{
+  uint32_t rep = 0;
+  memcpy(&rep, &d, sizeof(rep));
+  return rep;
+}
+
+/// Return the raw representation of a double
+static inline uint64_t
+double_to_rep(const double d)
+{
+  uint64_t rep = 0;
+  memcpy(&rep, &d, sizeof(rep));
+  return rep;
+}
+
+/// Return true if the lower boundary is closer than the upper boundary
+static inline bool
+double_lower_boundary_is_closer(const double d)
+{
+  const uint64_t rep          = double_to_rep(d);
+  const uint64_t mant         = rep & dbl_mant_mask;
+  const uint64_t expt         = rep & dbl_expt_mask;
+  const bool     is_subnormal = expt == 0;
+
+  // True when f = 2^(p-1) (except for the smallest normal)
+  return !is_subnormal && mant == 0;
+}
+
+#endif // EXESS_IEEE_FLOAT_H
diff --git a/subprojects/exess/src/int.c b/subprojects/exess/src/int.c
new file mode 100644
index 00000000..232cbcc4
--- /dev/null
+++ b/subprojects/exess/src/int.c
@@ -0,0 +1,45 @@
+/*
+  Copyright 2019-2021 David Robillard <d@drobilla.net>
+
+  Permission to use, copy, modify, and/or distribute this software for any
+  purpose with or without fee is hereby granted, provided that the above
+  copyright notice and this permission notice appear in all copies.
+
+  THIS SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+  WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+  MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+  ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+  WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+  ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+  OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+*/
+
+#include "read_utils.h"
+
+#include "exess/exess.h"
+
+#include <stdint.h>
+#include <stdlib.h>
+
+ExessResult
+exess_read_int(int32_t* const out, const char* const str)
+{
+  int64_t           long_out = 0;
+  const ExessResult r        = exess_read_long(&long_out, str);
+  if (r.status) {
+    return r;
+  }
+
+  if (long_out < INT32_MIN || long_out > INT32_MAX) {
+    return result(EXESS_OUT_OF_RANGE, r.count);
+  }
+
+  *out = (int32_t)long_out;
+  return r;
+}
+
+ExessResult
+exess_write_int(const int32_t value, const size_t buf_size, char* const buf)
+{
+  return exess_write_long(value, buf_size, buf);
+}
diff --git a/subprojects/exess/src/int_math.c b/subprojects/exess/src/int_math.c
new file mode 100644
index 00000000..ce263b43
--- /dev/null
+++ b/subprojects/exess/src/int_math.c
@@ -0,0 +1,78 @@
+/*
+  Copyright 2019-2021 David Robillard <d@drobilla.net>
+
+  Permission to use, copy, modify, and/or distribute this software for any
+  purpose with or without fee is hereby granted, provided that the above
+  copyright notice and this permission notice appear in all copies.
+
+  THIS SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+  WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+  MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+  ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+  WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+  ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+  OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+*/
+
+#include "int_math.h"
+
+#include "exess_config.h"
+
+#include <assert.h>
+
+unsigned
+exess_clz32(const uint32_t i)
+{
+  assert(i != 0);
+
+#if USE_BUILTIN_CLZ
+  return (unsigned)__builtin_clz(i);
+#else
+  unsigned n    = 32u;
+  uint32_t bits = i;
+  for (unsigned s = 16; s > 0; s >>= 1) {
+    const uint32_t left = bits >> s;
+    if (left) {
+      n -= s;
+      bits = left;
+    }
+  }
+  return n - bits;
+#endif
+}
+
+unsigned
+exess_clz64(const uint64_t i)
+{
+  assert(i != 0);
+
+#if USE_BUILTIN_CLZLL
+  return (unsigned)__builtin_clzll(i);
+#else
+  return i & 0xFFFFFFFF00000000 ? exess_clz32((uint32_t)(i >> 32u))
+                                : 32u + exess_clz32(i & 0xFFFFFFFF);
+#endif
+}
+
+uint64_t
+exess_ilog2(const uint64_t i)
+{
+  assert(i != 0);
+  return (64u - exess_clz64(i | 1u)) - 1u;
+}
+
+uint64_t
+exess_ilog10(const uint64_t i)
+{
+  // See https://graphics.stanford.edu/~seander/bithacks.html#IntegerLog10
+  const uint64_t log2 = exess_ilog2(i);
+  const uint64_t t    = (log2 + 1u) * 1233u >> 12u;
+
+  return t - (i < POW10[t]) + (i == 0);
+}
+
+uint8_t
+exess_num_digits(const uint64_t i)
+{
+  return i == 0u ? 1u : (uint8_t)(exess_ilog10(i) + 1u);
+}
diff --git a/subprojects/exess/src/int_math.h b/subprojects/exess/src/int_math.h
new file mode 100644
index 00000000..a4a57140
--- /dev/null
+++ b/subprojects/exess/src/int_math.h
@@ -0,0 +1,72 @@
+/*
+  Copyright 2019-2021 David Robillard <d@drobilla.net>
+
+  Permission to use, copy, modify, and/or distribute this software for any
+  purpose with or without fee is hereby granted, provided that the above
+  copyright notice and this permission notice appear in all copies.
+
+  THIS SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+  WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+  MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+  ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+  WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+  ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+  OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+*/
+
+#ifndef EXESS_INTMATH_H
+#define EXESS_INTMATH_H
+
+#include "attributes.h"
+
+#include <stdint.h>
+
+static const int uint64_digits10 = 19;
+
+static const uint64_t POW10[] = {1ull,
+                                 10ull,
+                                 100ull,
+                                 1000ull,
+                                 10000ull,
+                                 100000ull,
+                                 1000000ull,
+                                 10000000ull,
+                                 100000000ull,
+                                 1000000000ull,
+                                 10000000000ull,
+                                 100000000000ull,
+                                 1000000000000ull,
+                                 10000000000000ull,
+                                 100000000000000ull,
+                                 1000000000000000ull,
+                                 10000000000000000ull,
+                                 100000000000000000ull,
+                                 1000000000000000000ull,
+                                 10000000000000000000ull};
+
+/// Return the number of leading zeros in `i`
+EXESS_I_CONST_FUNC
+unsigned
+exess_clz32(uint32_t i);
+
+/// Return the number of leading zeros in `i`
+EXESS_I_CONST_FUNC
+unsigned
+exess_clz64(uint64_t i);
+
+/// Return the log base 2 of `i`
+EXESS_I_CONST_FUNC
+uint64_t
+exess_ilog2(uint64_t i);
+
+/// Return the log base 10 of `i`
+EXESS_I_CONST_FUNC
+uint64_t
+exess_ilog10(uint64_t i);
+
+/// Return the number of decimal digits required to represent `i`
+EXESS_I_CONST_FUNC
+uint8_t
+exess_num_digits(uint64_t i);
+
+#endif // EXESS_INTMATH_H
diff --git a/subprojects/exess/src/long.c b/subprojects/exess/src/long.c
new file mode 100644
index 00000000..2fbcdf59
--- /dev/null
+++ b/subprojects/exess/src/long.c
@@ -0,0 +1,110 @@
+/*
+  Copyright 2019-2021 David Robillard <d@drobilla.net>
+
+  Permission to use, copy, modify, and/or distribute this software for any
+  purpose with or without fee is hereby granted, provided that the above
+  copyright notice and this permission notice appear in all copies.
+
+  THIS SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+  WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+  MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+  ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+  WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+  ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+  OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+*/
+
+#include "int_math.h"
+#include "read_utils.h"
+#include "write_utils.h"
+
+#include "exess/exess.h"
+
+#include <stdbool.h>
+#include <stdint.h>
+#include <stdlib.h>
+
+ExessResult
+exess_read_long(int64_t* const out, const char* const str)
+{
+  *out = 0;
+
+  // Read leading sign if present
+  size_t i    = skip_whitespace(str);
+  int    sign = 1;
+  if (str[i] == '-') {
+    sign = -1;
+    ++i;
+  } else if (str[i] == '+') {
+    ++i;
+  }
+
+  // Read digits
+  uint64_t    magnitude = 0;
+  ExessResult r         = exess_read_ulong(&magnitude, str + i);
+  if (r.status > EXESS_EXPECTED_END) {
+    return result(r.status, i + r.count);
+  }
+
+  i += r.count;
+
+  if (sign > 0) {
+    if (magnitude > (uint64_t)INT64_MAX) {
+      return result(EXESS_OUT_OF_RANGE, i);
+    }
+
+    *out = (int64_t)magnitude;
+    return end_read(EXESS_SUCCESS, str, i);
+  }
+
+  const uint64_t min_magnitude = (uint64_t)(-(INT64_MIN + 1)) + 1;
+  if (magnitude > min_magnitude) {
+    return result(EXESS_OUT_OF_RANGE, i);
+  }
+
+  if (magnitude == min_magnitude) {
+    *out = INT64_MIN;
+  } else {
+    *out = -(int64_t)magnitude;
+  }
+
+  return end_read(r.status, str, i);
+}
+
+static size_t
+exess_long_string_length(const int64_t value)
+{
+  if (value == INT64_MIN) {
+    return 20;
+  }
+
+  if (value < 0) {
+    return 1u + exess_num_digits((uint64_t)-value);
+  }
+
+  return exess_num_digits((uint64_t)value);
+}
+
+ExessResult
+exess_write_long(const int64_t value, const size_t buf_size, char* const buf)
+{
+  if (!buf) {
+    return result(EXESS_SUCCESS, exess_long_string_length(value));
+  }
+
+  if (value == INT64_MIN) {
+    return end_write(
+      EXESS_SUCCESS,
+      buf_size,
+      buf,
+      write_string(20, "-9223372036854775808", buf_size, buf, 0));
+  }
+
+  const bool     is_negative = value < 0;
+  const uint64_t abs_value   = (uint64_t)(is_negative ? -value : value);
+
+  size_t      i = (is_negative) ? write_char('-', buf_size, buf, 0) : 0;
+  ExessResult r = write_digits(abs_value, buf_size, buf, i);
+
+  return end_write(r.status, buf_size, buf, i + r.count);
+}
diff --git a/subprojects/exess/src/macros.h b/subprojects/exess/src/macros.h
new file mode 100644
index 00000000..80ed68f5
--- /dev/null
+++ b/subprojects/exess/src/macros.h
@@ -0,0 +1,31 @@
+/*
+  Copyright 2019-2021 David Robillard <d@drobilla.net>
+
+  Permission to use, copy, modify, and/or distribute this software for any
+  purpose with or without fee is hereby granted, provided that the above
+  copyright notice and this permission notice appear in all copies.
+
+  THIS SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+  WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+  MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+  ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+  WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+  ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+  OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+*/
+
+#ifndef EXESS_MACROS_H
+#define EXESS_MACROS_H
+
+#define MIN(x, y) ((x) < (y) ? (x) : (y))
+#define MAX(x, y) ((x) > (y) ? (x) : (y))
+#define CLAMP(x, l, h) MAX(l, MIN(h, x))
+
+#define SET_IF(pointer, value) \
+  do {                         \
+    if (pointer) {             \
+      *(pointer) = (value);    \
+    }                          \
+  } while (0)
+
+#endif // EXESS_MACROS_H
diff --git a/subprojects/exess/src/read_utils.c b/subprojects/exess/src/read_utils.c
new file mode 100644
index 00000000..823fe97f
--- /dev/null
+++ b/subprojects/exess/src/read_utils.c
@@ -0,0 +1,51 @@
+/*
+  Copyright 2019-2021 David Robillard <d@drobilla.net>
+
+  Permission to use, copy, modify, and/or distribute this software for any
+  purpose with or without fee is hereby granted, provided that the above
+  copyright notice and this permission notice appear in all copies.
+
+  THIS SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+  WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+  MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+  ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+  WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+  ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+  OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+*/
+
+#include "read_utils.h"
+#include "string_utils.h"
+
+#include "exess/exess.h"
+
+ExessResult
+read_two_digit_number(uint8_t* const    out,
+                      const uint8_t     min_value,
+                      const uint8_t     max_value,
+                      const char* const str)
+{
+  size_t i = 0;
+
+  // Read digits
+  size_t d = 0;
+  for (; d < 2; ++d, ++i) {
+    if (is_digit(str[i])) {
+      *out = (uint8_t)((*out * 10) + (str[i] - '0'));
+    } else {
+      break;
+    }
+  }
+
+  // Ensure there are exactly the expected number of digits
+  if (d != 2) {
+    return result(EXESS_EXPECTED_DIGIT, i);
+  }
+
+  // Ensure value is in range
+  if (*out < min_value || *out > max_value) {
+    return result(EXESS_OUT_OF_RANGE, i);
+  }
+
+  return result(EXESS_SUCCESS, i);
+}
diff --git a/subprojects/exess/src/read_utils.h b/subprojects/exess/src/read_utils.h
new file mode 100644
index 00000000..3f7e1f56
--- /dev/null
+++ b/subprojects/exess/src/read_utils.h
@@ -0,0 +1,77 @@
+/*
+  Copyright 2019-2021 David Robillard <d@drobilla.net>
+
+  Permission to use, copy, modify, and/or distribute this software for any
+  purpose with or without fee is hereby granted, provided that the above
+  copyright notice and this permission notice appear in all copies.
+
+  THIS SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+  WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+  MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+  ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+  WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+  ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+  OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+*/
+
+#ifndef EXESS_READ_UTILS_H
+#define EXESS_READ_UTILS_H
+
+#include "string_utils.h"
+
+#include "exess/exess.h"
+
+#include <stdbool.h>
+#include <stddef.h>
+#include <stdint.h>
+
+static inline size_t
+skip_whitespace(const char* const str)
+{
+  size_t i = 0;
+  while (is_space(str[i])) {
+    ++i;
+  }
+
+  return i;
+}
+
+static inline bool
+is_end(const char c)
+{
+  switch (c) {
+  case '\0':
+  case ' ':
+  case '\f':
+  case '\n':
+  case '\r':
+  case '\t':
+  case '\v':
+    return true;
+  default:
+    break;
+  }
+
+  return false;
+}
+
+static inline ExessResult
+result(const ExessStatus status, const size_t count)
+{
+  const ExessResult r = {status, count};
+  return r;
+}
+
+ExessResult
+read_two_digit_number(uint8_t*    out,
+                      uint8_t     min_value,
+                      uint8_t     max_value,
+                      const char* str);
+
+static inline ExessResult
+end_read(const ExessStatus status, const char* str, const size_t i)
+{
+  return result((status || is_end(str[i])) ? status : EXESS_EXPECTED_END, i);
+}
+
+#endif // EXESS_READ_UTILS_H
diff --git a/subprojects/exess/src/scientific.c b/subprojects/exess/src/scientific.c
new file mode 100644
index 00000000..3e441a86
--- /dev/null
+++ b/subprojects/exess/src/scientific.c
@@ -0,0 +1,125 @@
+/*
+  Copyright 2019-2021 David Robillard <d@drobilla.net>
+
+  Permission to use, copy, modify, and/or distribute this software for any
+  purpose with or without fee is hereby granted, provided that the above
+  copyright notice and this permission notice appear in all copies.
+
+  THIS SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+  WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+  MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+  ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+  WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+  ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+  OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+*/
+
+#include "scientific.h"
+#include "decimal.h"
+#include "int_math.h"
+#include "read_utils.h"
+#include "write_utils.h"
+
+#include "exess/exess.h"
+
+#include <stdlib.h>
+#include <string.h>
+
+size_t
+exess_scientific_string_length(const ExessDecimalDouble value)
+{
+  switch (value.kind) {
+  case EXESS_NEGATIVE:
+    break;
+  case EXESS_NEGATIVE_INFINITY:
+    return 4;
+  case EXESS_NEGATIVE_ZERO:
+    return 6;
+  case EXESS_POSITIVE_ZERO:
+    return 5;
+  case EXESS_POSITIVE:
+    break;
+  case EXESS_POSITIVE_INFINITY:
+  case EXESS_NAN:
+    return 3;
+  }
+
+  const unsigned n_expt_digits =
+    (unsigned)exess_num_digits((unsigned)abs(value.expt));
+
+  return ((value.kind == EXESS_NEGATIVE) + // Sign
+          value.n_digits + 1 +             // Digits and point
+          (value.n_digits <= 1) +          // Added '0' after point
+          1 +                              // 'E'
+          (value.expt < 0) +               // Exponent sign
+          n_expt_digits);                  // Exponent digits
+}
+
+ExessResult
+exess_write_scientific(const ExessDecimalDouble value,
+                       const size_t             n,
+                       char* const              buf)
+{
+  size_t i = 0;
+
+  if (n < 4) {
+    return result(EXESS_NO_SPACE, 0);
+  }
+
+  switch (value.kind) {
+  case EXESS_NEGATIVE:
+    buf[i++] = '-';
+    break;
+  case EXESS_NEGATIVE_INFINITY:
+    return write_special(4, "-INF", n, buf);
+  case EXESS_NEGATIVE_ZERO:
+    return write_special(6, "-0.0E0", n, buf);
+  case EXESS_POSITIVE_ZERO:
+    return write_special(5, "0.0E0", n, buf);
+  case EXESS_POSITIVE:
+    break;
+  case EXESS_POSITIVE_INFINITY:
+    return write_special(3, "INF", n, buf);
+  case EXESS_NAN:
+    return write_special(3, "NaN", n, buf);
+  }
+
+  if (n - i <= value.n_digits + 1) {
+    buf[0] = '\0';
+    return result(EXESS_NO_SPACE, 0);
+  }
+
+  // Write mantissa, with decimal point after the first (normal form)
+  buf[i++] = value.digits[0];
+  buf[i++] = '.';
+  if (value.n_digits > 1) {
+    memcpy(buf + i, value.digits + 1, value.n_digits - 1);
+    i += value.n_digits - 1;
+  } else {
+    buf[i++] = '0';
+  }
+
+  // Write exponent
+
+  const unsigned n_expt_digits = exess_num_digits((unsigned)abs(value.expt));
+
+  if (n - i <= 1u + (value.expt < 0) + n_expt_digits) {
+    buf[0] = '\0';
+    return result(EXESS_NO_SPACE, 0);
+  }
+
+  buf[i++] = 'E';
+  if (value.expt < 0) {
+    buf[i++] = '-';
+  }
+
+  unsigned abs_expt = (unsigned)abs(value.expt);
+  char*    s        = buf + i + n_expt_digits;
+
+  *s-- = '\0';
+  do {
+    *s-- = (char)('0' + (abs_expt % 10));
+  } while ((abs_expt /= 10) > 0);
+
+  return result(EXESS_SUCCESS, i + n_expt_digits);
+}
diff --git a/subprojects/exess/src/scientific.h b/subprojects/exess/src/scientific.h
new file mode 100644
index 00000000..96c3096a
--- /dev/null
+++ b/subprojects/exess/src/scientific.h
@@ -0,0 +1,35 @@
+/*
+  Copyright 2019-2021 David Robillard <d@drobilla.net>
+
+  Permission to use, copy, modify, and/or distribute this software for any
+  purpose with or without fee is hereby granted, provided that the above
+  copyright notice and this permission notice appear in all copies.
+
+  THIS SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+  WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+  MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+  ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+  WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+  ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+  OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+*/
+
+#ifndef EXESS_SCIENTIFIC_H
+#define EXESS_SCIENTIFIC_H
+
+#include "decimal.h"
+
+#include "exess/exess.h"
+
+#include <stddef.h>
+
+EXESS_CONST_FUNC
+size_t
+exess_scientific_string_length(ExessDecimalDouble value);
+
+ExessResult
+exess_write_scientific(ExessDecimalDouble  value,
+                       size_t              n,
+                       char* EXESS_NONNULL buf);
+
+#endif // EXESS_SCIENTIFIC_H
diff --git a/subprojects/exess/src/short.c b/subprojects/exess/src/short.c
new file mode 100644
index 00000000..9241b75a
--- /dev/null
+++ b/subprojects/exess/src/short.c
@@ -0,0 +1,45 @@
+/*
+  Copyright 2019-2021 David Robillard <d@drobilla.net>
+
+  Permission to use, copy, modify, and/or distribute this software for any
+  purpose with or without fee is hereby granted, provided that the above
+  copyright notice and this permission notice appear in all copies.
+
+  THIS SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+  WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+  MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+  ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+  WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+  ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+  OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+*/
+
+#include "read_utils.h"
+
+#include "exess/exess.h"
+
+#include <stdint.h>
+#include <stdlib.h>
+
+ExessResult
+exess_read_short(int16_t* const out, const char* const str)
+{
+  int64_t           long_out = 0;
+  const ExessResult r        = exess_read_long(&long_out, str);
+  if (r.status) {
+    return r;
+  }
+
+  if (long_out < INT16_MIN || long_out > INT16_MAX) {
+    return result(EXESS_OUT_OF_RANGE, r.count);
+  }
+
+  *out = (int16_t)long_out;
+  return r;
+}
+
+ExessResult
+exess_write_short(const int16_t value, const size_t buf_size, char* const buf)
+{
+  return exess_write_long(value, buf_size, buf);
+}
diff --git a/subprojects/exess/src/soft_float.c b/subprojects/exess/src/soft_float.c
new file mode 100644
index 00000000..bea9a176
--- /dev/null
+++ b/subprojects/exess/src/soft_float.c
@@ -0,0 +1,161 @@
+/*
+  Copyright 2019-2021 David Robillard <d@drobilla.net>
+
+  Permission to use, copy, modify, and/or distribute this software for any
+  purpose with or without fee is hereby granted, provided that the above
+  copyright notice and this permission notice appear in all copies.
+
+  THIS SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+  WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+  MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+  ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+  WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+  ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+  OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+*/
+
+#include "soft_float.h"
+
+#include "ieee_float.h"
+#include "int_math.h"
+
+#include <assert.h>
+#include <math.h>
+#include <stdint.h>
+
+/// 10^k for k = min_dec_expt, min_dec_expt + dec_expt_step, ..., max_dec_expt
+static const ExessSoftFloat soft_pow10[] = {
+  {0xFA8FD5A0081C0288, -1220}, {0xBAAEE17FA23EBF76, -1193},
+  {0x8B16FB203055AC76, -1166}, {0xCF42894A5DCE35EA, -1140},
+  {0x9A6BB0AA55653B2D, -1113}, {0xE61ACF033D1A45DF, -1087},
+  {0xAB70FE17C79AC6CA, -1060}, {0xFF77B1FCBEBCDC4F, -1034},
+  {0xBE5691EF416BD60C, -1007}, {0x8DD01FAD907FFC3C, -980},
+  {0xD3515C2831559A83, -954},  {0x9D71AC8FADA6C9B5, -927},
+  {0xEA9C227723EE8BCB, -901},  {0xAECC49914078536D, -874},
+  {0x823C12795DB6CE57, -847},  {0xC21094364DFB5637, -821},
+  {0x9096EA6F3848984F, -794},  {0xD77485CB25823AC7, -768},
+  {0xA086CFCD97BF97F4, -741},  {0xEF340A98172AACE5, -715},
+  {0xB23867FB2A35B28E, -688},  {0x84C8D4DFD2C63F3B, -661},
+  {0xC5DD44271AD3CDBA, -635},  {0x936B9FCEBB25C996, -608},
+  {0xDBAC6C247D62A584, -582},  {0xA3AB66580D5FDAF6, -555},
+  {0xF3E2F893DEC3F126, -529},  {0xB5B5ADA8AAFF80B8, -502},
+  {0x87625F056C7C4A8B, -475},  {0xC9BCFF6034C13053, -449},
+  {0x964E858C91BA2655, -422},  {0xDFF9772470297EBD, -396},
+  {0xA6DFBD9FB8E5B88F, -369},  {0xF8A95FCF88747D94, -343},
+  {0xB94470938FA89BCF, -316},  {0x8A08F0F8BF0F156B, -289},
+  {0xCDB02555653131B6, -263},  {0x993FE2C6D07B7FAC, -236},
+  {0xE45C10C42A2B3B06, -210},  {0xAA242499697392D3, -183},
+  {0xFD87B5F28300CA0E, -157},  {0xBCE5086492111AEB, -130},
+  {0x8CBCCC096F5088CC, -103},  {0xD1B71758E219652C, -77},
+  {0x9C40000000000000, -50},   {0xE8D4A51000000000, -24},
+  {0xAD78EBC5AC620000, 3},     {0x813F3978F8940984, 30},
+  {0xC097CE7BC90715B3, 56},    {0x8F7E32CE7BEA5C70, 83},
+  {0xD5D238A4ABE98068, 109},   {0x9F4F2726179A2245, 136},
+  {0xED63A231D4C4FB27, 162},   {0xB0DE65388CC8ADA8, 189},
+  {0x83C7088E1AAB65DB, 216},   {0xC45D1DF942711D9A, 242},
+  {0x924D692CA61BE758, 269},   {0xDA01EE641A708DEA, 295},
+  {0xA26DA3999AEF774A, 322},   {0xF209787BB47D6B85, 348},
+  {0xB454E4A179DD1877, 375},   {0x865B86925B9BC5C2, 402},
+  {0xC83553C5C8965D3D, 428},   {0x952AB45CFA97A0B3, 455},
+  {0xDE469FBD99A05FE3, 481},   {0xA59BC234DB398C25, 508},
+  {0xF6C69A72A3989F5C, 534},   {0xB7DCBF5354E9BECE, 561},
+  {0x88FCF317F22241E2, 588},   {0xCC20CE9BD35C78A5, 614},
+  {0x98165AF37B2153DF, 641},   {0xE2A0B5DC971F303A, 667},
+  {0xA8D9D1535CE3B396, 694},   {0xFB9B7CD9A4A7443C, 720},
+  {0xBB764C4CA7A44410, 747},   {0x8BAB8EEFB6409C1A, 774},
+  {0xD01FEF10A657842C, 800},   {0x9B10A4E5E9913129, 827},
+  {0xE7109BFBA19C0C9D, 853},   {0xAC2820D9623BF429, 880},
+  {0x80444B5E7AA7CF85, 907},   {0xBF21E44003ACDD2D, 933},
+  {0x8E679C2F5E44FF8F, 960},   {0xD433179D9C8CB841, 986},
+  {0x9E19DB92B4E31BA9, 1013},  {0xEB96BF6EBADF77D9, 1039},
+  {0xAF87023B9BF0EE6B, 1066}};
+
+ExessSoftFloat
+soft_float_from_double(const double d)
+{
+  assert(d >= 0.0);
+
+  const uint64_t rep  = double_to_rep(d);
+  const uint64_t frac = rep & dbl_mant_mask;
+  const int      expt = (int)((rep & dbl_expt_mask) >> dbl_physical_mant_dig);
+
+  if (expt == 0) { // Subnormal
+    ExessSoftFloat v = {frac, dbl_subnormal_expt};
+    return v;
+  }
+
+  const ExessSoftFloat v = {frac + dbl_hidden_bit, expt - dbl_expt_bias};
+  return v;
+}
+
+double
+soft_float_to_double(const ExessSoftFloat v)
+{
+  return ldexp((double)v.f, v.e);
+}
+
+ExessSoftFloat
+soft_float_normalize(ExessSoftFloat value)
+{
+  const unsigned amount = exess_clz64(value.f);
+
+  value.f <<= amount;
+  value.e -= (int)amount;
+
+  return value;
+}
+
+ExessSoftFloat
+soft_float_multiply(const ExessSoftFloat lhs, const ExessSoftFloat rhs)
+{
+  static const uint64_t mask  = 0xFFFFFFFF;
+  static const uint64_t round = 1u << 31u;
+
+  const uint64_t l0   = lhs.f >> 32u;
+  const uint64_t l1   = lhs.f & mask;
+  const uint64_t r0   = rhs.f >> 32u;
+  const uint64_t r1   = rhs.f & mask;
+  const uint64_t l0r0 = l0 * r0;
+  const uint64_t l1r0 = l1 * r0;
+  const uint64_t l0r1 = l0 * r1;
+  const uint64_t l1r1 = l1 * r1;
+  const uint64_t mid  = (l1r1 >> 32u) + (l0r1 & mask) + (l1r0 & mask) + round;
+
+  const ExessSoftFloat r = {l0r0 + (l0r1 >> 32u) + (l1r0 >> 32u) + (mid >> 32u),
+                            lhs.e + rhs.e + 64};
+
+  return r;
+}
+
+ExessSoftFloat
+soft_float_exact_pow10(const int expt)
+{
+  static const ExessSoftFloat table[8] = {{0xA000000000000000, -60},
+                                          {0xC800000000000000, -57},
+                                          {0xFA00000000000000, -54},
+                                          {0x9C40000000000000, -50},
+                                          {0xC350000000000000, -47},
+                                          {0xF424000000000000, -44},
+                                          {0x9896800000000000, -40}};
+
+  assert(expt > 0);
+  assert(expt < dec_expt_step);
+  return table[expt - 1];
+}
+
+ExessSoftFloat
+soft_float_pow10_under(const int exponent, int* pow10_exponent)
+{
+  assert(exponent >= min_dec_expt);
+  assert(exponent < max_dec_expt + dec_expt_step);
+
+  const int cache_offset = -min_dec_expt;
+  const int index        = (exponent + cache_offset) / dec_expt_step;
+
+  *pow10_exponent = min_dec_expt + index * dec_expt_step;
+
+  assert(*pow10_exponent <= exponent);
+  assert(exponent < *pow10_exponent + dec_expt_step);
+
+  return soft_pow10[index];
+}
diff --git a/subprojects/exess/src/soft_float.h b/subprojects/exess/src/soft_float.h
new file mode 100644
index 00000000..b766ae5e
--- /dev/null
+++ b/subprojects/exess/src/soft_float.h
@@ -0,0 +1,71 @@
+/*
+  Copyright 2019-2021 David Robillard <d@drobilla.net>
+
+  Permission to use, copy, modify, and/or distribute this software for any
+  purpose with or without fee is hereby granted, provided that the above
+  copyright notice and this permission notice appear in all copies.
+
+  THIS SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+  WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+  MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+  ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+  WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+  ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+  OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+*/
+
+#ifndef EXESS_SOFT_FLOAT_H
+#define EXESS_SOFT_FLOAT_H
+
+#include "attributes.h"
+
+#include <stdint.h>
+
+typedef struct {
+  uint64_t f; ///< Significand
+  int      e; ///< Exponent
+} ExessSoftFloat;
+
+static const int min_dec_expt  = -348;
+static const int max_dec_expt  = 340;
+static const int dec_expt_step = 8;
+
+/// Convert `d` to a soft float
+EXESS_I_CONST_FUNC
+ExessSoftFloat
+soft_float_from_double(double d);
+
+/// Convert `v` to a double
+double
+soft_float_to_double(ExessSoftFloat v);
+
+/// Normalize `value` so the MSb of its significand is 1
+EXESS_I_CONST_FUNC
+ExessSoftFloat
+soft_float_normalize(ExessSoftFloat value);
+
+/// Multiply `lhs` by `rhs` and return the result
+EXESS_I_CONST_FUNC
+ExessSoftFloat
+soft_float_multiply(ExessSoftFloat lhs, ExessSoftFloat rhs);
+
+/// Return exactly 10^e for e in [0...dec_expt_step]
+EXESS_I_CONST_FUNC
+ExessSoftFloat
+soft_float_exact_pow10(int expt);
+
+/**
+   Return a cached power of 10 with exponent not greater than `max_exponent`.
+
+   Valid only for `max_exponent` values from min_dec_expt to max_dec_expt +
+   dec_expt_step.  The returned power's exponent is a multiple of
+   dec_expt_step.
+
+   @param max_exponent Maximum decimal exponent of the result.
+   @param[out] pow10_exponent Set to the decimal exponent of the result.
+   @return A cached power of 10 as a soft float.
+*/
+ExessSoftFloat
+soft_float_pow10_under(int max_exponent, int* pow10_exponent);
+
+#endif // EXESS_SOFT_FLOAT_H
diff --git a/subprojects/exess/src/strerror.c b/subprojects/exess/src/strerror.c
new file mode 100644
index 00000000..b2167cac
--- /dev/null
+++ b/subprojects/exess/src/strerror.c
@@ -0,0 +1,70 @@
+/*
+  Copyright 2019-2021 David Robillard <d@drobilla.net>
+
+  Permission to use, copy, modify, and/or distribute this software for any
+  purpose with or without fee is hereby granted, provided that the above
+  copyright notice and this permission notice appear in all copies.
+
+  THIS SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+  WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+  MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+  ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+  WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+  ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+  OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+*/
+
+#include "exess/exess.h"
+
+const char*
+exess_strerror(ExessStatus status)
+{
+  switch (status) {
+  case EXESS_SUCCESS:
+    return "Success";
+  case EXESS_EXPECTED_END:
+    return "Expected end of value";
+  case EXESS_EXPECTED_BOOLEAN:
+    return "Expected \"false\", \"true\", \"0\" or \"1\"";
+  case EXESS_EXPECTED_INTEGER:
+    return "Expected an integer value";
+  case EXESS_EXPECTED_DURATION:
+    return "Expected a duration starting with 'P'";
+  case EXESS_EXPECTED_SIGN:
+    return "Expected '-' or '+'";
+  case EXESS_EXPECTED_DIGIT:
+    return "Expected a digit";
+  case EXESS_EXPECTED_COLON:
+    return "Expected ':'";
+  case EXESS_EXPECTED_DASH:
+    return "Expected '-'";
+  case EXESS_EXPECTED_TIME_SEP:
+    return "Expected 'T'";
+  case EXESS_EXPECTED_TIME_TAG:
+    return "Expected 'H', 'M', or 'S'";
+  case EXESS_EXPECTED_DATE_TAG:
+    return "Expected 'Y', 'M', or 'D'";
+  case EXESS_EXPECTED_HEX:
+    return "Expected a hexadecimal character";
+  case EXESS_EXPECTED_BASE64:
+    return "Expected a base64 character";
+  case EXESS_BAD_ORDER:
+    return "Invalid field order";
+  case EXESS_BAD_VALUE:
+    return "Invalid value";
+  case EXESS_OUT_OF_RANGE:
+    return "Value outside valid range";
+  case EXESS_NO_SPACE:
+    return "Insufficient space";
+  case EXESS_WOULD_REDUCE_PRECISION:
+    return "Precision reducing coercion required";
+  case EXESS_WOULD_ROUND:
+    return "Rounding coercion required";
+  case EXESS_WOULD_TRUNCATE:
+    return "Truncating coercion required";
+  case EXESS_UNSUPPORTED:
+    return "Unsupported value";
+  }
+
+  return "Unknown error";
+}
diff --git a/subprojects/exess/src/string_utils.h b/subprojects/exess/src/string_utils.h
new file mode 100644
index 00000000..8acb2ffd
--- /dev/null
+++ b/subprojects/exess/src/string_utils.h
@@ -0,0 +1,74 @@
+/*
+  Copyright 2011-2021 David Robillard <d@drobilla.net>
+
+  Permission to use, copy, modify, and/or distribute this software for any
+  purpose with or without fee is hereby granted, provided that the above
+  copyright notice and this permission notice appear in all copies.
+
+  THIS SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+  WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+  MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+  ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+  WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+  ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+  OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+*/
+
+#ifndef EXESS_STRING_UTILS_H
+#define EXESS_STRING_UTILS_H
+
+#include <stdbool.h>
+
+/// Return true if `c` lies within [`min`...`max`] (inclusive)
+static inline bool
+in_range(const int c, const int min, const int max)
+{
+  return (c >= min && c <= max);
+}
+
+/// Return true if `c` is a whitespace character
+static inline bool
+is_space(const int c)
+{
+  switch (c) {
+  case ' ':
+  case '\f':
+  case '\n':
+  case '\r':
+  case '\t':
+  case '\v':
+    return true;
+  default:
+    return false;
+  }
+}
+
+/// ALPHA ::= [A-Za-z]
+static inline bool
+is_alpha(const int c)
+{
+  return in_range(c, 'A', 'Z') || in_range(c, 'a', 'z');
+}
+
+/// DIGIT ::= [0-9]
+static inline bool
+is_digit(const int c)
+{
+  return in_range(c, '0', '9');
+}
+
+/// HEXDIG ::= DIGIT | "A" | "B" | "C" | "D" | "E" | "F"
+static inline bool
+is_hexdig(const int c)
+{
+  return is_digit(c) || in_range(c, 'A', 'F');
+}
+
+/// BASE64 ::= ALPHA | DIGIT | "+" | "/" | "="
+static inline bool
+is_base64(const int c)
+{
+  return is_alpha(c) || is_digit(c) || c == '+' || c == '/' || c == '=';
+}
+
+#endif // EXESS_STRING_UTILS_H
diff --git a/subprojects/exess/src/strtod.c b/subprojects/exess/src/strtod.c
new file mode 100644
index 00000000..55a0b082
--- /dev/null
+++ b/subprojects/exess/src/strtod.c
@@ -0,0 +1,405 @@
+/*
+  Copyright 2019-2021 David Robillard <d@drobilla.net>
+
+  Permission to use, copy, modify, and/or distribute this software for any
+  purpose with or without fee is hereby granted, provided that the above
+  copyright notice and this permission notice appear in all copies.
+
+  THIS SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+  WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+  MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+  ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+  WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+  ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+  OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+*/
+
+#include "strtod.h"
+#include "bigint.h"
+#include "decimal.h"
+#include "ieee_float.h"
+#include "int_math.h"
+#include "macros.h"
+#include "read_utils.h"
+#include "soft_float.h"
+#include "string_utils.h"
+
+#include <assert.h>
+#include <float.h>
+#include <math.h>
+#include <stdbool.h>
+#include <stdint.h>
+#include <string.h>
+
+static inline int
+read_sign(const char** const sptr)
+{
+  if (**sptr == '-') {
+    ++(*sptr);
+    return -1;
+  }
+
+  if (**sptr == '+') {
+    ++(*sptr);
+  }
+
+  return 1;
+}
+
+ExessResult
+parse_decimal(ExessDecimalDouble* const out, const char* const str)
+{
+  memset(out, 0, sizeof(*out));
+
+  // Read leading sign if necessary
+  const char* s                = str;
+  const int   sign             = read_sign(&s);
+  int         n_leading_before = 0;
+
+  out->kind = (sign < 0) ? EXESS_NEGATIVE : EXESS_POSITIVE;
+
+  // Check that the first character is valid
+  if (*s != '.' && !is_digit(*s)) {
+    return result(EXESS_EXPECTED_DIGIT, (size_t)(s - str));
+  }
+
+  // Skip leading zeros before decimal point
+  while (*s == '0') {
+    ++s;
+    ++n_leading_before;
+  }
+
+  // Skip leading zeros after decimal point
+  int  n_leading_after = 0;     // Zeros skipped after decimal point
+  bool after_point     = false; // True if we are after the decimal point
+  if (*s == '.') {
+    after_point = true;
+    for (++s; *s == '0'; ++s) {
+      ++n_leading_after;
+    }
+  }
+
+  // Read significant digits of the mantissa into a 64-bit integer
+  uint64_t frac     = 0; // Fraction value (ignoring decimal point)
+  int      n_before = 0; // Number of digits before decimal point
+  int      n_after  = 0; // Number of digits after decimal point
+  for (; out->n_digits < DBL_DECIMAL_DIG + 1; ++s) {
+    if (is_digit(*s)) {
+      frac = (frac * 10) + (unsigned)(*s - '0');
+      n_before += !after_point;
+      n_after += after_point;
+      out->digits[out->n_digits++] = *s;
+    } else if (*s == '.' && !after_point) {
+      after_point = true;
+    } else {
+      break;
+    }
+  }
+
+  // Skip extra digits
+  int n_extra_before = 0;
+  int n_extra_after  = 0;
+  for (;; ++s) {
+    if (*s == '.' && !after_point) {
+      after_point = true;
+    } else if (is_digit(*s)) {
+      n_extra_before += !after_point;
+      n_extra_after += after_point;
+    } else {
+      break;
+    }
+  }
+
+  // Calculate final output exponent
+  out->expt = n_extra_before - n_after - n_leading_after;
+
+  if (out->n_digits == 0) {
+    out->kind =
+      out->kind == EXESS_NEGATIVE ? EXESS_NEGATIVE_ZERO : EXESS_POSITIVE_ZERO;
+  }
+
+  return result(EXESS_SUCCESS, (size_t)(s - str));
+}
+
+ExessResult
+parse_double(ExessDecimalDouble* const out, const char* const str)
+{
+  memset(out, 0, sizeof(*out));
+
+  // Handle non-numeric special cases
+
+  if (!strcmp(str, "NaN")) {
+    out->kind = EXESS_NAN;
+    return result(EXESS_SUCCESS, 3u);
+  }
+
+  if (!strcmp(str, "-INF")) {
+    out->kind = EXESS_NEGATIVE_INFINITY;
+    return result(EXESS_SUCCESS, 4u);
+  }
+
+  if (!strcmp(str, "INF")) {
+    out->kind = EXESS_POSITIVE_INFINITY;
+    return result(EXESS_SUCCESS, 3u);
+  }
+
+  if (!strcmp(str, "+INF")) {
+    out->kind = EXESS_POSITIVE_INFINITY;
+    return result(EXESS_SUCCESS, 4u);
+  }
+
+  // Read mantissa as a decimal
+  const ExessResult r = parse_decimal(out, str);
+  if (r.status) {
+    return r;
+  }
+
+  const char* s = str + r.count;
+
+  // Read exponent
+  int abs_expt  = 0;
+  int expt_sign = 1;
+  if (*s == 'e' || *s == 'E') {
+    ++s;
+
+    if (*s != '-' && *s != '+' && !is_digit(*s)) {
+      return result(EXESS_EXPECTED_DIGIT, (size_t)(s - str));
+    }
+
+    expt_sign = read_sign(&s);
+    while (is_digit(*s)) {
+      abs_expt = (abs_expt * 10) + (*s++ - '0');
+    }
+  }
+
+  // Calculate final output exponent
+  out->expt += expt_sign * abs_expt;
+
+  if (out->n_digits == 0) {
+    out->kind = out->kind < EXESS_POSITIVE_ZERO ? EXESS_NEGATIVE_ZERO
+                                                : EXESS_POSITIVE_ZERO;
+  }
+
+  return result(EXESS_SUCCESS, (size_t)(s - str));
+}
+
+static uint64_t
+normalize(ExessSoftFloat* value, const uint64_t error)
+{
+  const int original_e = value->e;
+
+  *value = soft_float_normalize(*value);
+
+  assert(value->e <= original_e);
+  return error << (unsigned)(original_e - value->e);
+}
+
+/**
+   Return the error added by floating point multiplication.
+
+   Should be l + r + l*r/(2^64) + 0.5, but we short the denominator to 63 due
+   to lack of precision, which effectively rounds up.
+*/
+static inline uint64_t
+product_error(const uint64_t lerror,
+              const uint64_t rerror,
+              const uint64_t half_ulp)
+{
+  return lerror + rerror + ((lerror * rerror) >> 63u) + half_ulp;
+}
+
+/**
+   Guess the binary floating point value for decimal input.
+
+   @param significand Significand from the input.
+   @param expt10 Decimal exponent from the input.
+   @param n_digits Number of decimal digits in the significand.
+   @param[out] guess Either the exact number, or its predecessor.
+   @return True if `guess` is correct.
+*/
+static bool
+sftod(const uint64_t        significand,
+      const int             expt10,
+      const int             n_digits,
+      ExessSoftFloat* const guess)
+{
+  assert(expt10 <= max_dec_expt);
+  assert(expt10 >= min_dec_expt);
+
+  /* The general idea here is to try and find a power of 10 that we can
+     multiply by the significand to get the number.  We get one from the
+     cache which is possibly too small, then multiply by another power of 10
+     to make up the difference if necessary.  For example, with a target
+     power of 10^70, if we get 10^68 from the cache, then we multiply again
+     by 10^2.  This, as well as normalization, accumulates error, which is
+     tracked throughout to know if we got the precise number. */
+
+  // Use a common denominator of 2^3 to avoid fractions
+  static const unsigned lg_denom = 3;
+  static const uint64_t denom    = 1u << 3u;
+  static const uint64_t half_ulp = 4u;
+
+  // Start out with just the significand, and no error
+  ExessSoftFloat input = {significand, 0};
+  uint64_t       error = normalize(&input, 0);
+
+  // Get a power of 10 that takes us most of the way without overshooting
+  int            cached_expt10 = 0;
+  ExessSoftFloat pow10         = soft_float_pow10_under(expt10, &cached_expt10);
+
+  // Get an exact fixup power if necessary
+  const int d_expt10 = expt10 - cached_expt10;
+  if (d_expt10) {
+    input = soft_float_multiply(input, soft_float_exact_pow10(d_expt10));
+    if (d_expt10 > uint64_digits10 - n_digits) {
+      error += half_ulp; // Product does not fit in an integer
+    }
+  }
+
+  // Multiply the significand by the power, normalize, and update the error
+  input = soft_float_multiply(input, pow10);
+  error = normalize(&input, product_error(error, half_ulp, half_ulp));
+
+  // Get the effective number of significant bits from the order of magnitude
+  const int      magnitude      = 64 + input.e;
+  const int      real_magnitude = magnitude - dbl_subnormal_expt;
+  const unsigned n_significant_bits =
+    (unsigned)MAX(0, MIN(real_magnitude, DBL_MANT_DIG));
+
+  // Calculate the number of "extra" bits of precision we have
+  assert(n_significant_bits <= 64);
+  unsigned n_extra_bits = 64u - n_significant_bits;
+  if (n_extra_bits + lg_denom >= 64u) {
+    // Very small subnormal where extra * denom does not fit in an integer
+    // Shift right (and accumulate some more error) to compensate
+    const unsigned amount = (n_extra_bits + lg_denom) - 63;
+
+    input.f >>= amount;
+    input.e += (int)amount;
+    error = product_error((error >> amount) + 1u, half_ulp, half_ulp);
+    n_extra_bits -= amount;
+  }
+
+  // Calculate boundaries for the extra bits (with the common denominator)
+  assert(n_extra_bits < 64);
+  const uint64_t extra_mask = (1ull << n_extra_bits) - 1u;
+  const uint64_t extra_bits = (input.f & extra_mask) * denom;
+  const uint64_t middle     = (1ull << (n_extra_bits - 1u)) * denom;
+  const uint64_t low        = middle - error;
+  const uint64_t high       = middle + error;
+
+  // Round to nearest representable double
+  guess->f = (input.f >> n_extra_bits) + (extra_bits >= high);
+  guess->e = input.e + (int)n_extra_bits;
+
+  // Too inaccurate if the extra bits are within the error around the middle
+  return extra_bits <= low || extra_bits >= high;
+}
+
+static int
+compare_buffer(const char* buf, const int expt, const ExessSoftFloat upper)
+{
+  ExessBigint buf_bigint;
+  exess_bigint_set_decimal_string(&buf_bigint, buf);
+
+  ExessBigint upper_bigint;
+  exess_bigint_set_u64(&upper_bigint, upper.f);
+
+  if (expt >= 0) {
+    exess_bigint_multiply_pow10(&buf_bigint, (unsigned)expt);
+  } else {
+    exess_bigint_multiply_pow10(&upper_bigint, (unsigned)-expt);
+  }
+
+  if (upper.e > 0) {
+    exess_bigint_shift_left(&upper_bigint, (unsigned)upper.e);
+  } else {
+    exess_bigint_shift_left(&buf_bigint, (unsigned)-upper.e);
+  }
+
+  return exess_bigint_compare(&buf_bigint, &upper_bigint);
+}
+
+double
+parsed_double_to_double(const ExessDecimalDouble in)
+{
+  static const int      n_exact_pow10        = sizeof(POW10) / sizeof(POW10[0]);
+  static const unsigned max_exact_int_digits = 15;   // Digits that fit exactly
+  static const int      max_decimal_power    = 309;  // Max finite power
+  static const int      min_decimal_power    = -324; // Min non-zero power
+
+  switch (in.kind) {
+  case EXESS_NEGATIVE:
+    break;
+  case EXESS_NEGATIVE_INFINITY:
+    return (double)-INFINITY;
+  case EXESS_NEGATIVE_ZERO:
+    return -0.0;
+  case EXESS_POSITIVE_ZERO:
+    return 0.0;
+  case EXESS_POSITIVE:
+    break;
+  case EXESS_POSITIVE_INFINITY:
+    return (double)INFINITY;
+  case EXESS_NAN:
+    return (double)NAN;
+  }
+
+  uint64_t frac = 0;
+  for (unsigned i = 0u; i < in.n_digits; ++i) {
+    if (is_digit(in.digits[i])) {
+      frac = (frac * 10) + (unsigned)(in.digits[i] - '0');
+    }
+  }
+
+  const int expt         = in.expt;
+  const int result_power = (int)in.n_digits + expt;
+
+  // Return early for simple exact cases
+
+  const int sign = in.kind >= EXESS_POSITIVE_ZERO ? 1 : -1;
+
+  if (result_power > max_decimal_power) {
+    return sign * (double)INFINITY;
+  }
+
+  if (result_power < min_decimal_power) {
+    return sign * 0.0;
+  }
+
+  if (in.n_digits < max_exact_int_digits) {
+    if (expt < 0 && -expt < n_exact_pow10) {
+      return sign * ((double)frac / (double)POW10[-expt]);
+    }
+
+    if (expt >= 0 && expt < n_exact_pow10) {
+      return sign * ((double)frac * (double)POW10[expt]);
+    }
+  }
+
+  // Try to guess the number using only soft floating point (fast path)
+  ExessSoftFloat guess = {0, 0};
+  const bool     exact = sftod(frac, expt, (int)in.n_digits, &guess);
+  const double   g     = soft_float_to_double(guess);
+  if (exact) {
+    return sign * g;
+  }
+
+  // Not sure, guess is either the number or its predecessor (rare slow path)
+  // Compare it with the buffer using bigints to find out which
+  const ExessSoftFloat upper = {guess.f * 2 + 1, guess.e - 1};
+  const int            cmp   = compare_buffer(in.digits, in.expt, upper);
+  if (cmp < 0) {
+    return sign * g;
+  }
+
+  if (cmp > 0) {
+    return sign * nextafter(g, (double)INFINITY);
+  }
+
+  if ((guess.f & 1u) == 0) {
+    return sign * g; // Round towards even
+  }
+
+  return sign * nextafter(g, (double)INFINITY); // Round odd up
+}
diff --git a/subprojects/exess/src/strtod.h b/subprojects/exess/src/strtod.h
new file mode 100644
index 00000000..7fdf7d6a
--- /dev/null
+++ b/subprojects/exess/src/strtod.h
@@ -0,0 +1,33 @@
+/*
+  Copyright 2019-2021 David Robillard <d@drobilla.net>
+
+  Permission to use, copy, modify, and/or distribute this software for any
+  purpose with or without fee is hereby granted, provided that the above
+  copyright notice and this permission notice appear in all copies.
+
+  THIS SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+  WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+  MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+  ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+  WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+  ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+  OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+*/
+
+#ifndef EXESS_STRTOD_H
+#define EXESS_STRTOD_H
+
+#include "decimal.h"
+
+#include "exess/exess.h"
+
+ExessResult
+parse_decimal(ExessDecimalDouble* out, const char* str);
+
+ExessResult
+parse_double(ExessDecimalDouble* out, const char* str);
+
+double
+parsed_double_to_double(ExessDecimalDouble in);
+
+#endif // EXESS_STRTOD_H
diff --git a/subprojects/exess/src/time.c b/subprojects/exess/src/time.c
new file mode 100644
index 00000000..8c7bc12c
--- /dev/null
+++ b/subprojects/exess/src/time.c
@@ -0,0 +1,172 @@
+/*
+  Copyright 2019-2021 David Robillard <d@drobilla.net>
+
+  Permission to use, copy, modify, and/or distribute this software for any
+  purpose with or without fee is hereby granted, provided that the above
+  copyright notice and this permission notice appear in all copies.
+
+  THIS SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+  WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+  MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+  ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+  WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+  ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+  OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+*/
+
+#include "read_utils.h"
+#include "string_utils.h"
+#include "time_utils.h"
+#include "timezone.h"
+#include "write_utils.h"
+
+#include "exess/exess.h"
+
+#include <assert.h>
+#include <stdbool.h>
+#include <stddef.h>
+#include <stdint.h>
+#include <string.h>
+
+ExessResult
+read_nanoseconds(uint32_t* const out, const char* const str)
+{
+  size_t i               = 0;
+  char   frac_digits[10] = {'0', '0', '0', '0', '0', '0', '0', '0', '0', 0};
+  for (unsigned j = 0u; j < 9u && is_digit(str[i]); ++j) {
+    frac_digits[j] = str[i++];
+  }
+
+  return result(exess_read_uint(out, frac_digits).status, i);
+}
+
+ExessResult
+exess_read_time(ExessTime* const out, const char* const str)
+{
+  memset(out, 0, sizeof(*out));
+
+  // Read hour
+  size_t      i = skip_whitespace(str);
+  ExessResult r = read_two_digit_number(&out->hour, 0, 24, str + i);
+  if (r.status) {
+    return result(r.status, i + r.count);
+  }
+
+  // Read hour-minute delimiter
+  i += r.count;
+  if (str[i] != ':') {
+    return result(EXESS_EXPECTED_COLON, i);
+  }
+
+  // Read minute
+  ++i;
+  r = read_two_digit_number(&out->minute, 0, 59, str + i);
+  if (r.status) {
+    return result(r.status, i + r.count);
+  }
+
+  // Read minute-second delimiter
+  i += r.count;
+  if (str[i] != ':') {
+    return result(EXESS_EXPECTED_COLON, i);
+  }
+
+  // Read second
+  ++i;
+  r = read_two_digit_number(&out->second, 0, 59, str + i);
+  i += r.count;
+  if (r.status) {
+    return result(r.status, i);
+  }
+
+  // Read nanoseconds if present
+  if (str[i] == '.') {
+    ++i;
+    r = read_nanoseconds(&out->nanosecond, str + i);
+    i += r.count;
+  }
+
+  // Read timezone if present
+  if (!is_end(str[i])) {
+    r = exess_read_timezone(&out->zone, str + i);
+    i += r.count;
+  } else {
+    out->zone.quarter_hours = EXESS_LOCAL;
+  }
+
+  return end_read(r.status, str, i);
+}
+
+size_t
+write_nanoseconds(const uint32_t nanosecond,
+                  const size_t   buf_size,
+                  char* const    buf,
+                  const size_t   i)
+{
+  assert(nanosecond <= 999999999);
+
+  if (nanosecond == 0) {
+    return 0;
+  }
+
+  char frac_digits[10] = {'0', '0', '0', '0', '0', '0', '0', '0', '0', 0};
+
+  // Write digits right to left, but replace trailing zeros with null
+  uint32_t remaining     = nanosecond;
+  uint32_t n_trailing    = 0;
+  bool     wrote_nonzero = false;
+  for (uint32_t j = 0; remaining > 0; ++j) {
+    const char digit = (char)('0' + (remaining % 10));
+    if (!wrote_nonzero && digit == '0') {
+      frac_digits[8 - j] = '\0';
+      ++n_trailing;
+    } else {
+      frac_digits[8 - j] = digit;
+    }
+
+    wrote_nonzero = wrote_nonzero || digit != '0';
+    remaining /= 10;
+  }
+
+  size_t n = write_char('.', buf_size, buf, i);
+
+  n += write_string(9 - n_trailing, frac_digits, buf_size, buf, i + n);
+
+  return n;
+}
+
+ExessResult
+write_time(const ExessTime value,
+           const size_t    buf_size,
+           char* const     buf,
+           const size_t    offset)
+{
+  if (value.hour > 24 || value.minute > 59 || value.second > 59 ||
+      value.nanosecond > 999999999 ||
+      (value.hour == 24 &&
+       (value.minute != 0 || value.second != 0 || value.nanosecond != 0))) {
+    return result(EXESS_BAD_VALUE, 0);
+  }
+
+  size_t o = offset;
+
+  // Write integral hour, minute, and second
+  o += write_two_digit_number(value.hour, buf_size, buf, o);
+  o += write_char(':', buf_size, buf, o);
+  o += write_two_digit_number(value.minute, buf_size, buf, o);
+  o += write_char(':', buf_size, buf, o);
+  o += write_two_digit_number(value.second, buf_size, buf, o);
+  o += write_nanoseconds(value.nanosecond, buf_size, buf, o);
+
+  const ExessResult r = write_timezone(value.zone, buf_size, buf, o);
+
+  return result(r.status, o - offset + r.count);
+}
+
+ExessResult
+exess_write_time(const ExessTime value, const size_t buf_size, char* const buf)
+{
+  const ExessResult r = write_time(value, buf_size, buf, 0);
+
+  return end_write(r.status, buf_size, buf, r.count);
+}
diff --git a/subprojects/exess/src/time_utils.h b/subprojects/exess/src/time_utils.h
new file mode 100644
index 00000000..32e4b6df
--- /dev/null
+++ b/subprojects/exess/src/time_utils.h
@@ -0,0 +1,35 @@
+/*
+  Copyright 2019-2021 David Robillard <d@drobilla.net>
+
+  Permission to use, copy, modify, and/or distribute this software for any
+  purpose with or without fee is hereby granted, provided that the above
+  copyright notice and this permission notice appear in all copies.
+
+  THIS SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+  WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+  MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+  ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+  WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+  ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+  OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+*/
+
+#ifndef EXESS_TIME_UTILS_H
+#define EXESS_TIME_UTILS_H
+
+#include <stddef.h>
+#include <stdint.h>
+
+/// Read fractional digits as an integer number of nanoseconds
+ExessResult
+read_nanoseconds(uint32_t* out, const char* str);
+
+/// Write nanoseconds as fractional digits
+size_t
+write_nanoseconds(uint32_t nanosecond, size_t buf_size, char* buf, size_t i);
+
+/// Write a complete time with timezone suffix if necessary
+ExessResult
+write_time(ExessTime value, size_t buf_size, char* buf, size_t offset);
+
+#endif // EXESS_TIME_UTILS_H
diff --git a/subprojects/exess/src/timezone.c b/subprojects/exess/src/timezone.c
new file mode 100644
index 00000000..717f0d08
--- /dev/null
+++ b/subprojects/exess/src/timezone.c
@@ -0,0 +1,150 @@
+/*
+  Copyright 2019-2021 David Robillard <d@drobilla.net>
+
+  Permission to use, copy, modify, and/or distribute this software for any
+  purpose with or without fee is hereby granted, provided that the above
+  copyright notice and this permission notice appear in all copies.
+
+  THIS SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+  WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+  MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+  ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+  WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+  ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+  OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+*/
+
+#include "timezone.h"
+#include "date_utils.h"
+#include "read_utils.h"
+#include "string_utils.h"
+#include "write_utils.h"
+
+#include "exess/exess.h"
+
+#include <stdint.h>
+#include <stdlib.h>
+
+ExessResult
+exess_read_timezone(ExessTimezone* const out, const char* const str)
+{
+  out->quarter_hours = INT8_MAX;
+
+  // Start at the beginning (no whitespace skipping here)
+  size_t i = 0;
+
+  // Handle UTC special case
+  if (str[i] == 'Z') {
+    out->quarter_hours = 0;
+    return result(EXESS_SUCCESS, i + 1);
+  }
+
+  // Read leading sign (required)
+  int sign = 1;
+  switch (str[i]) {
+  case '+':
+    ++i;
+    break;
+  case '-':
+    sign = -1;
+    ++i;
+    break;
+  default:
+    return result(EXESS_EXPECTED_SIGN, i);
+  }
+
+  const char h0 = str[i];
+  if (!is_digit(h0)) {
+    return result(EXESS_EXPECTED_DIGIT, i);
+  }
+
+  const char h1 = str[++i];
+  if (!is_digit(h1)) {
+    return result(EXESS_EXPECTED_DIGIT, i);
+  }
+
+  ++i;
+
+  const int8_t hour = (int8_t)(sign * (10 * (h0 - '0') + (h1 - '0')));
+  if (hour > 14 || hour < -14) {
+    return result(EXESS_OUT_OF_RANGE, i);
+  }
+
+  if (str[i] != ':') {
+    return result(EXESS_EXPECTED_COLON, i);
+  }
+
+  const char m0 = str[++i];
+  if (!is_digit(m0)) {
+    return result(EXESS_EXPECTED_DIGIT, i);
+  }
+
+  const char m1 = str[++i];
+  if (!is_digit(m1)) {
+    return result(EXESS_EXPECTED_DIGIT, i);
+  }
+
+  const int8_t minute = (int8_t)(sign * (10 * (m0 - '0') + (m1 - '0')));
+
+  ++i;
+
+  if (minute % 15) {
+    return result(EXESS_UNSUPPORTED, i);
+  }
+
+  if (minute > 59 || minute < -59) {
+    return result(EXESS_OUT_OF_RANGE, i);
+  }
+
+  const int8_t quarters = (int8_t)(4 * hour + minute / 15);
+  if (quarters < -56 || quarters > 56) {
+    return result(EXESS_OUT_OF_RANGE, i);
+  }
+
+  out->quarter_hours = quarters;
+
+  return result(EXESS_SUCCESS, i);
+}
+
+size_t
+exess_timezone_string_length(const ExessTimezone value)
+{
+  return value.quarter_hours != EXESS_LOCAL ? (value.quarter_hours == 0) ? 1 : 6
+                                            : 0;
+}
+
+ExessResult
+write_timezone(const ExessTimezone value,
+               const size_t        buf_size,
+               char* const         buf,
+               size_t              o)
+{
+  if (value.quarter_hours == EXESS_LOCAL) {
+    return result(EXESS_SUCCESS, 0);
+  }
+
+  if (value.quarter_hours < -56 || value.quarter_hours > 56) {
+    return result(EXESS_BAD_VALUE, 0);
+  }
+
+  if (!buf) {
+    return result(EXESS_SUCCESS, exess_timezone_string_length(value));
+  }
+
+  if (value.quarter_hours == 0) {
+    write_char('Z', buf_size, buf, o);
+    return result(EXESS_SUCCESS, 1);
+  }
+
+  const uint8_t abs_quarters = (uint8_t)abs(value.quarter_hours);
+  const uint8_t abs_hour     = abs_quarters / 4;
+  const uint8_t abs_minute   = (uint8_t)(15u * (abs_quarters % 4u));
+
+  size_t n = 0;
+  n += write_char(value.quarter_hours < 0 ? '-' : '+', buf_size, buf, o + n);
+  n += write_two_digit_number(abs_hour, buf_size, buf, o + n);
+  n += write_char(':', buf_size, buf, o + n);
+  n += write_two_digit_number(abs_minute, buf_size, buf, o + n);
+
+  return result(EXESS_SUCCESS, n);
+}
diff --git a/subprojects/exess/src/timezone.h b/subprojects/exess/src/timezone.h
new file mode 100644
index 00000000..b10c2483
--- /dev/null
+++ b/subprojects/exess/src/timezone.h
@@ -0,0 +1,59 @@
+/*
+  Copyright 2021 David Robillard <d@drobilla.net>
+
+  Permission to use, copy, modify, and/or distribute this software for any
+  purpose with or without fee is hereby granted, provided that the above
+  copyright notice and this permission notice appear in all copies.
+
+  THIS SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+  WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+  MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+  ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+  WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+  ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+  OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+*/
+
+#ifndef EXESS_TIMEZONE_H
+#define EXESS_TIMEZONE_H
+
+#include "exess/exess.h"
+
+#include <stddef.h>
+
+/// The maximum length of a canonical timezone string, 6
+#define EXESS_MAX_TIMEZONE_LENGTH 6
+
+/**
+   Read a timezone string after any leading whitespace.
+
+   @param out Set to the parsed value, or false on error.
+   @param str String input.
+   @return The `count` of characters read, and a `status` code.
+*/
+EXESS_API
+ExessResult
+exess_read_timezone(ExessTimezone* EXESS_NONNULL out,
+                    const char* EXESS_NONNULL    str);
+
+/**
+   Write a canonical timezone suffix.
+
+   The output is always in canonical form, either `Z` for UTC or a signed hour
+   and minute offset with leading zeros, like `-05:30` or `+14:00`.
+
+   @param value Value to write.
+   @param buf_size The size of `buf` in bytes.
+   @param buf Output buffer, or null to only measure.
+   @param o The current write offset in `buf`
+
+   @return #EXESS_SUCCESS on success, #EXESS_NO_SPACE if the buffer is too
+   small, or #EXESS_BAD_VALUE if the value is invalid.
+*/
+ExessResult
+write_timezone(ExessTimezone        value,
+               size_t               buf_size,
+               char* EXESS_NULLABLE buf,
+               size_t               o);
+
+#endif // EXESS_TIMEZONE_H
diff --git a/subprojects/exess/src/ubyte.c b/subprojects/exess/src/ubyte.c
new file mode 100644
index 00000000..f520832a
--- /dev/null
+++ b/subprojects/exess/src/ubyte.c
@@ -0,0 +1,45 @@
+/*
+  Copyright 2019-2021 David Robillard <d@drobilla.net>
+
+  Permission to use, copy, modify, and/or distribute this software for any
+  purpose with or without fee is hereby granted, provided that the above
+  copyright notice and this permission notice appear in all copies.
+
+  THIS SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+  WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+  MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+  ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+  WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+  ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+  OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+*/
+
+#include "read_utils.h"
+
+#include "exess/exess.h"
+
+#include <stdint.h>
+#include <stdlib.h>
+
+ExessResult
+exess_read_ubyte(uint8_t* const out, const char* const str)
+{
+  uint64_t          long_out = 0;
+  const ExessResult r        = exess_read_ulong(&long_out, str);
+  if (r.status) {
+    return r;
+  }
+
+  if (long_out > UINT8_MAX) {
+    return result(EXESS_OUT_OF_RANGE, r.count);
+  }
+
+  *out = (uint8_t)long_out;
+  return r;
+}
+
+ExessResult
+exess_write_ubyte(const uint8_t value, const size_t buf_size, char* const buf)
+{
+  return exess_write_ulong(value, buf_size, buf);
+}
diff --git a/subprojects/exess/src/uint.c b/subprojects/exess/src/uint.c
new file mode 100644
index 00000000..dad46b48
--- /dev/null
+++ b/subprojects/exess/src/uint.c
@@ -0,0 +1,45 @@
+/*
+  Copyright 2019-2021 David Robillard <d@drobilla.net>
+
+  Permission to use, copy, modify, and/or distribute this software for any
+  purpose with or without fee is hereby granted, provided that the above
+  copyright notice and this permission notice appear in all copies.
+
+  THIS SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+  WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+  MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+  ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+  WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+  ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+  OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+*/
+
+#include "read_utils.h"
+
+#include "exess/exess.h"
+
+#include <stdint.h>
+#include <stdlib.h>
+
+ExessResult
+exess_read_uint(uint32_t* const out, const char* const str)
+{
+  uint64_t          long_out = 0;
+  const ExessResult r        = exess_read_ulong(&long_out, str);
+  if (r.status && r.status != EXESS_EXPECTED_END) {
+    return r;
+  }
+
+  if (long_out > UINT32_MAX) {
+    return result(EXESS_OUT_OF_RANGE, r.count);
+  }
+
+  *out = (uint32_t)long_out;
+  return r;
+}
+
+ExessResult
+exess_write_uint(const uint32_t value, const size_t buf_size, char* const buf)
+{
+  return exess_write_ulong(value, buf_size, buf);
+}
diff --git a/subprojects/exess/src/ulong.c b/subprojects/exess/src/ulong.c
new file mode 100644
index 00000000..3d9e00c9
--- /dev/null
+++ b/subprojects/exess/src/ulong.c
@@ -0,0 +1,94 @@
+/*
+  Copyright 2019-2021 David Robillard <d@drobilla.net>
+
+  Permission to use, copy, modify, and/or distribute this software for any
+  purpose with or without fee is hereby granted, provided that the above
+  copyright notice and this permission notice appear in all copies.
+
+  THIS SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+  WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+  MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+  ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+  WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+  ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+  OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+*/
+
+#include "exess/exess.h"
+#include "int_math.h"
+#include "read_utils.h"
+#include "string_utils.h"
+#include "write_utils.h"
+
+#include <stdint.h>
+#include <stdlib.h>
+
+ExessResult
+exess_read_ulong(uint64_t* const out, const char* const str)
+{
+  *out = 0;
+
+  // Ensure the first character is a digit
+  size_t i = skip_whitespace(str);
+  if (!is_digit(str[i])) {
+    return result(EXESS_EXPECTED_DIGIT, i);
+  }
+
+  // Skip leading zeros
+  int n_zeroes = 0;
+  while (str[i] == '0') {
+    ++i;
+    ++n_zeroes;
+  }
+
+  // Read digits
+  for (; is_digit(str[i]); ++i) {
+    const uint64_t next = (*out * 10u) + (uint64_t)(str[i] - '0');
+    if (next < *out) {
+      *out = 0;
+      return result(EXESS_OUT_OF_RANGE, i);
+    }
+
+    *out = next;
+  }
+
+  return end_read(EXESS_SUCCESS, str, i);
+}
+
+ExessResult
+write_digits(const uint64_t value,
+             const size_t   buf_size,
+             char* const    buf,
+             const size_t   i)
+{
+  const uint8_t n_digits = exess_num_digits(value);
+  if (!buf) {
+    return result(EXESS_SUCCESS, n_digits);
+  }
+
+  if (i + n_digits >= buf_size) {
+    return end_write(EXESS_NO_SPACE, buf_size, buf, 0);
+  }
+
+  // Point s to the end
+  char* s = buf + i + n_digits - 1u;
+
+  // Write integer part (right to left)
+  uint64_t remaining = value;
+  do {
+    *s-- = (char)('0' + (remaining % 10));
+  } while ((remaining /= 10) > 0);
+
+  return result(EXESS_SUCCESS, n_digits);
+}
+
+ExessResult
+exess_write_ulong(const uint64_t value, const size_t buf_size, char* const buf)
+{
+  if (!buf) {
+    return result(EXESS_SUCCESS, exess_num_digits(value));
+  }
+
+  const ExessResult r = write_digits(value, buf_size, buf, 0);
+  return end_write(r.status, buf_size, buf, r.count);
+}
diff --git a/subprojects/exess/src/ushort.c b/subprojects/exess/src/ushort.c
new file mode 100644
index 00000000..fc4d2c7e
--- /dev/null
+++ b/subprojects/exess/src/ushort.c
@@ -0,0 +1,45 @@
+/*
+  Copyright 2019-2021 David Robillard <d@drobilla.net>
+
+  Permission to use, copy, modify, and/or distribute this software for any
+  purpose with or without fee is hereby granted, provided that the above
+  copyright notice and this permission notice appear in all copies.
+
+  THIS SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+  WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+  MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+  ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+  WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+  ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+  OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+*/
+
+#include "read_utils.h"
+
+#include "exess/exess.h"
+
+#include <stdint.h>
+#include <stdlib.h>
+
+ExessResult
+exess_read_ushort(uint16_t* const out, const char* const str)
+{
+  uint64_t          long_out = 0;
+  const ExessResult r        = exess_read_ulong(&long_out, str);
+  if (r.status) {
+    return r;
+  }
+
+  if (long_out > UINT16_MAX) {
+    return result(EXESS_OUT_OF_RANGE, r.count);
+  }
+
+  *out = (uint16_t)long_out;
+  return r;
+}
+
+ExessResult
+exess_write_ushort(const uint16_t value, const size_t buf_size, char* const buf)
+{
+  return exess_write_ulong(value, buf_size, buf);
+}
diff --git a/subprojects/exess/src/variant.c b/subprojects/exess/src/variant.c
new file mode 100644
index 00000000..d8977abd
--- /dev/null
+++ b/subprojects/exess/src/variant.c
@@ -0,0 +1,431 @@
+/*
+  Copyright 2019-2021 David Robillard <d@drobilla.net>
+
+  Permission to use, copy, modify, and/or distribute this software for any
+  purpose with or without fee is hereby granted, provided that the above
+  copyright notice and this permission notice appear in all copies.
+
+  THIS SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+  WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+  MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+  ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+  WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+  ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+  OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+*/
+
+#include "read_utils.h"
+#include "write_utils.h"
+
+#include "exess/exess.h"
+
+#include <stdbool.h>
+#include <stdint.h>
+#include <string.h>
+
+// Constructors
+
+ExessVariant
+exess_make_nothing(const ExessStatus status)
+{
+  ExessVariant v = {EXESS_NOTHING, .value.as_status = status};
+  return v;
+}
+
+ExessVariant
+exess_make_boolean(const bool value)
+{
+  ExessVariant v = {EXESS_BOOLEAN, .value.as_bool = value};
+  return v;
+}
+
+ExessVariant
+exess_make_decimal(const double value)
+{
+  ExessVariant v = {EXESS_DECIMAL, .value.as_double = value};
+  return v;
+}
+
+ExessVariant
+exess_make_double(const double value)
+{
+  ExessVariant v = {EXESS_DOUBLE, .value.as_double = value};
+  return v;
+}
+
+ExessVariant
+exess_make_float(const float value)
+{
+  ExessVariant v = {EXESS_FLOAT, .value.as_float = value};
+  return v;
+}
+
+ExessVariant
+exess_make_long(const int64_t value)
+{
+  ExessVariant v = {EXESS_LONG, .value.as_long = value};
+  return v;
+}
+
+ExessVariant
+exess_make_int(const int32_t value)
+{
+  ExessVariant v = {EXESS_INT, .value.as_int = value};
+  return v;
+}
+
+ExessVariant
+exess_make_short(const int16_t value)
+{
+  ExessVariant v = {EXESS_SHORT, .value.as_short = value};
+  return v;
+}
+
+ExessVariant
+exess_make_byte(const int8_t value)
+{
+  ExessVariant v = {EXESS_BYTE, .value.as_byte = value};
+  return v;
+}
+
+ExessVariant
+exess_make_ulong(const uint64_t value)
+{
+  ExessVariant v = {EXESS_ULONG, .value.as_ulong = value};
+  return v;
+}
+
+ExessVariant
+exess_make_uint(const uint32_t value)
+{
+  ExessVariant v = {EXESS_UINT, .value.as_uint = value};
+  return v;
+}
+
+ExessVariant
+exess_make_ushort(const uint16_t value)
+{
+  ExessVariant v = {EXESS_USHORT, .value.as_ushort = value};
+  return v;
+}
+
+ExessVariant
+exess_make_ubyte(const uint8_t value)
+{
+  ExessVariant v = {EXESS_UBYTE, .value.as_ubyte = value};
+  return v;
+}
+
+ExessVariant
+exess_make_duration(const ExessDuration value)
+{
+  ExessVariant v = {EXESS_DURATION, .value.as_duration = value};
+  return v;
+}
+
+ExessVariant
+exess_make_datetime(const ExessDateTime value)
+{
+  ExessVariant v = {EXESS_DATETIME, .value.as_datetime = value};
+  return v;
+}
+
+ExessVariant
+exess_make_time(const ExessTime value)
+{
+  ExessVariant v = {EXESS_TIME, .value.as_time = value};
+  return v;
+}
+
+ExessVariant
+exess_make_date(const ExessDate value)
+{
+  ExessVariant v = {EXESS_DATE, .value.as_date = value};
+  return v;
+}
+
+ExessVariant
+exess_make_hex(const ExessBlob blob)
+{
+  ExessVariant v = {EXESS_HEX, .value.as_blob = blob};
+  return v;
+}
+
+ExessVariant
+exess_make_base64(const ExessBlob blob)
+{
+  ExessVariant v = {EXESS_BASE64, .value.as_blob = blob};
+  return v;
+}
+
+// Accessors
+
+ExessStatus
+exess_get_status(const ExessVariant* const variant)
+{
+  return variant->datatype == EXESS_NOTHING ? variant->value.as_status
+                                            : EXESS_SUCCESS;
+}
+const bool*
+exess_get_boolean(const ExessVariant* const variant)
+{
+  return variant->datatype == EXESS_BOOLEAN ? &variant->value.as_bool : NULL;
+}
+
+const double*
+exess_get_double(const ExessVariant* const variant)
+{
+  return (variant->datatype == EXESS_DECIMAL ||
+          variant->datatype == EXESS_DOUBLE)
+           ? &variant->value.as_double
+           : NULL;
+}
+
+const float*
+exess_get_float(const ExessVariant* const variant)
+{
+  return variant->datatype == EXESS_FLOAT ? &variant->value.as_float : NULL;
+}
+
+const int64_t*
+exess_get_long(const ExessVariant* const variant)
+{
+  return (variant->datatype >= EXESS_INTEGER && variant->datatype <= EXESS_LONG)
+           ? &variant->value.as_long
+           : NULL;
+}
+
+const int32_t*
+exess_get_int(const ExessVariant* const variant)
+{
+  return variant->datatype == EXESS_INT ? &variant->value.as_int : NULL;
+}
+
+const int16_t*
+exess_get_short(const ExessVariant* const variant)
+{
+  return variant->datatype == EXESS_SHORT ? &variant->value.as_short : NULL;
+}
+
+const int8_t*
+exess_get_byte(const ExessVariant* const variant)
+{
+  return variant->datatype == EXESS_BYTE ? &variant->value.as_byte : NULL;
+}
+
+const uint64_t*
+exess_get_ulong(const ExessVariant* const variant)
+{
+  return (variant->datatype == EXESS_NON_NEGATIVE_INTEGER ||
+          variant->datatype == EXESS_ULONG ||
+          variant->datatype == EXESS_POSITIVE_INTEGER)
+           ? &variant->value.as_ulong
+           : NULL;
+}
+
+const uint32_t*
+exess_get_uint(const ExessVariant* const variant)
+{
+  return variant->datatype == EXESS_UINT ? &variant->value.as_uint : NULL;
+}
+
+const uint16_t*
+exess_get_ushort(const ExessVariant* const variant)
+{
+  return variant->datatype == EXESS_USHORT ? &variant->value.as_ushort : NULL;
+}
+
+const uint8_t*
+exess_get_ubyte(const ExessVariant* const variant)
+{
+  return variant->datatype == EXESS_UBYTE ? &variant->value.as_ubyte : NULL;
+}
+
+const ExessBlob*
+exess_get_blob(const ExessVariant* const variant)
+{
+  return (variant->datatype == EXESS_HEX || variant->datatype == EXESS_BASE64)
+           ? &variant->value.as_blob
+           : NULL;
+}
+
+const ExessDuration*
+exess_get_duration(const ExessVariant* const variant)
+{
+  return variant->datatype == EXESS_DURATION ? &variant->value.as_duration
+                                             : NULL;
+}
+
+const ExessDateTime*
+exess_get_datetime(const ExessVariant* const variant)
+{
+  return variant->datatype == EXESS_DATETIME ? &variant->value.as_datetime
+                                             : NULL;
+}
+
+const ExessTime*
+exess_get_time(const ExessVariant* const variant)
+{
+  return variant->datatype == EXESS_TIME ? &variant->value.as_time : NULL;
+}
+
+const ExessDate*
+exess_get_date(const ExessVariant* const variant)
+{
+  return variant->datatype == EXESS_DATE ? &variant->value.as_date : NULL;
+}
+
+// Reading and Writing
+
+ExessResult
+exess_read_variant(ExessVariant* const out,
+                   ExessDatatype       datatype,
+                   const char* const   str)
+{
+  ExessResult r = {EXESS_UNSUPPORTED, 0};
+
+  out->datatype = datatype;
+
+  switch (datatype) {
+  case EXESS_NOTHING:
+    break;
+  case EXESS_BOOLEAN:
+    return exess_read_boolean(&out->value.as_bool, str);
+  case EXESS_DECIMAL:
+    return exess_read_decimal(&out->value.as_double, str);
+  case EXESS_DOUBLE:
+    return exess_read_double(&out->value.as_double, str);
+  case EXESS_FLOAT:
+    return exess_read_float(&out->value.as_float, str);
+  case EXESS_INTEGER:
+    return exess_read_long(&out->value.as_long, str);
+
+  case EXESS_NON_POSITIVE_INTEGER:
+    if (!(r = exess_read_long(&out->value.as_long, str)).status) {
+      if (out->value.as_long > 0) {
+        return result(EXESS_OUT_OF_RANGE, r.count);
+      }
+    }
+    break;
+
+  case EXESS_NEGATIVE_INTEGER:
+    if (!(r = exess_read_long(&out->value.as_long, str)).status) {
+      if (out->value.as_long >= 0) {
+        return result(EXESS_OUT_OF_RANGE, r.count);
+      }
+    }
+    break;
+
+  case EXESS_LONG:
+    return exess_read_long(&out->value.as_long, str);
+  case EXESS_INT:
+    return exess_read_int(&out->value.as_int, str);
+  case EXESS_SHORT:
+    return exess_read_short(&out->value.as_short, str);
+  case EXESS_BYTE:
+    return exess_read_byte(&out->value.as_byte, str);
+  case EXESS_NON_NEGATIVE_INTEGER:
+  case EXESS_ULONG:
+    return exess_read_ulong(&out->value.as_ulong, str);
+  case EXESS_UINT:
+    return exess_read_uint(&out->value.as_uint, str);
+  case EXESS_USHORT:
+    return exess_read_ushort(&out->value.as_ushort, str);
+  case EXESS_UBYTE:
+    return exess_read_ubyte(&out->value.as_ubyte, str);
+
+  case EXESS_POSITIVE_INTEGER:
+    if (!(r = exess_read_ulong(&out->value.as_ulong, str)).status) {
+      if (out->value.as_ulong <= 0) {
+        return result(EXESS_OUT_OF_RANGE, r.count);
+      }
+    }
+    break;
+
+  case EXESS_DURATION:
+    return exess_read_duration(&out->value.as_duration, str);
+  case EXESS_DATETIME:
+    return exess_read_datetime(&out->value.as_datetime, str);
+  case EXESS_TIME:
+    return exess_read_time(&out->value.as_time, str);
+  case EXESS_DATE:
+    return exess_read_date(&out->value.as_date, str);
+  case EXESS_HEX:
+    return exess_read_hex(&out->value.as_blob, str);
+  case EXESS_BASE64:
+    return exess_read_base64(&out->value.as_blob, str);
+  }
+
+  return r;
+}
+
+ExessResult
+exess_write_variant(const ExessVariant variant,
+                    const size_t       buf_size,
+                    char* const        buf)
+{
+  if (buf_size > 0) {
+    buf[0] = '\0';
+  }
+
+  switch (variant.datatype) {
+  case EXESS_NOTHING:
+    break;
+  case EXESS_BOOLEAN:
+    return exess_write_boolean(variant.value.as_bool, buf_size, buf);
+  case EXESS_DECIMAL:
+    return exess_write_decimal(variant.value.as_double, buf_size, buf);
+  case EXESS_DOUBLE:
+    return exess_write_double(variant.value.as_double, buf_size, buf);
+  case EXESS_FLOAT:
+    return exess_write_float(variant.value.as_float, buf_size, buf);
+  case EXESS_INTEGER:
+  case EXESS_NON_POSITIVE_INTEGER:
+  case EXESS_NEGATIVE_INTEGER:
+  case EXESS_LONG:
+    return exess_write_long(variant.value.as_long, buf_size, buf);
+  case EXESS_INT:
+    return exess_write_int(variant.value.as_int, buf_size, buf);
+  case EXESS_SHORT:
+    return exess_write_short(variant.value.as_short, buf_size, buf);
+  case EXESS_BYTE:
+    return exess_write_byte(variant.value.as_byte, buf_size, buf);
+  case EXESS_NON_NEGATIVE_INTEGER:
+  case EXESS_ULONG:
+    return exess_write_ulong(variant.value.as_ulong, buf_size, buf);
+  case EXESS_UINT:
+    return exess_write_uint(variant.value.as_uint, buf_size, buf);
+  case EXESS_USHORT:
+    return exess_write_ushort(variant.value.as_ushort, buf_size, buf);
+  case EXESS_UBYTE:
+    return exess_write_ubyte(variant.value.as_ubyte, buf_size, buf);
+  case EXESS_POSITIVE_INTEGER:
+    return exess_write_ulong(variant.value.as_ulong, buf_size, buf);
+  case EXESS_DURATION:
+    return exess_write_duration(variant.value.as_duration, buf_size, buf);
+  case EXESS_DATETIME:
+    return exess_write_datetime(variant.value.as_datetime, buf_size, buf);
+  case EXESS_TIME:
+    return exess_write_time(variant.value.as_time, buf_size, buf);
+  case EXESS_DATE:
+    return exess_write_date(variant.value.as_date, buf_size, buf);
+  case EXESS_HEX:
+    if (variant.value.as_blob.data) {
+      return exess_write_hex(variant.value.as_blob.size,
+                             (void*)variant.value.as_blob.data,
+                             buf_size,
+                             buf);
+    }
+    break;
+  case EXESS_BASE64:
+    if (variant.value.as_blob.data) {
+      return exess_write_base64(variant.value.as_blob.size,
+                                (void*)variant.value.as_blob.data,
+                                buf_size,
+                                buf);
+    }
+    break;
+  }
+
+  return end_write(EXESS_BAD_VALUE, buf_size, buf, 0);
+}
diff --git a/subprojects/exess/src/warnings.h b/subprojects/exess/src/warnings.h
new file mode 100644
index 00000000..2b1f6587
--- /dev/null
+++ b/subprojects/exess/src/warnings.h
@@ -0,0 +1,46 @@
+/*
+  Copyright 2019-2021 David Robillard <d@drobilla.net>
+
+  Permission to use, copy, modify, and/or distribute this software for any
+  purpose with or without fee is hereby granted, provided that the above
+  copyright notice and this permission notice appear in all copies.
+
+  THIS SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+  WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+  MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+  ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+  WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+  ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+  OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+*/
+
+#ifndef EXESS_WARNINGS_H
+#define EXESS_WARNINGS_H
+
+#if defined(__clang__)
+
+#  define EXESS_DISABLE_CONVERSION_WARNINGS              \
+    _Pragma("clang diagnostic push")                     \
+    _Pragma("clang diagnostic ignored \"-Wconversion\"") \
+    _Pragma("clang diagnostic ignored \"-Wdouble-promotion\"")
+
+#  define EXESS_RESTORE_WARNINGS _Pragma("clang diagnostic pop")
+
+#elif defined(__GNUC__)
+
+#  define EXESS_DISABLE_CONVERSION_WARNINGS                  \
+    _Pragma("GCC diagnostic push")                           \
+    _Pragma("GCC diagnostic ignored \"-Wconversion\"")       \
+    _Pragma("GCC diagnostic ignored \"-Wfloat-conversion\"") \
+    _Pragma("GCC diagnostic ignored \"-Wdouble-promotion\"")
+
+#  define EXESS_RESTORE_WARNINGS _Pragma("GCC diagnostic pop")
+
+#else
+
+#  define EXESS_DISABLE_CONVERSION_WARNINGS
+#  define EXESS_RESTORE_WARNINGS
+
+#endif
+
+#endif // EXESS_WARNINGS_H
diff --git a/subprojects/exess/src/write_utils.c b/subprojects/exess/src/write_utils.c
new file mode 100644
index 00000000..fdac0754
--- /dev/null
+++ b/subprojects/exess/src/write_utils.c
@@ -0,0 +1,50 @@
+/*
+  Copyright 2019-2021 David Robillard <d@drobilla.net>
+
+  Permission to use, copy, modify, and/or distribute this software for any
+  purpose with or without fee is hereby granted, provided that the above
+  copyright notice and this permission notice appear in all copies.
+
+  THIS SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+  WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+  MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+  ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+  WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+  ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+  OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+*/
+
+#include "write_utils.h"
+#include "read_utils.h"
+
+#include "exess/exess.h"
+
+#include <string.h>
+
+size_t
+write_two_digit_number(const uint8_t value,
+                       const size_t  buf_size,
+                       char* const   buf,
+                       const size_t  i)
+{
+  if (buf_size >= i + 1) {
+    buf[i]     = (char)((value >= 10) ? ('0' + value / 10) : '0');
+    buf[i + 1] = (char)('0' + (value % 10));
+  }
+
+  return 2;
+}
+
+ExessResult
+write_special(const size_t      string_length,
+              const char* const string,
+              const size_t      buf_size,
+              char* const       buf)
+{
+  if (buf_size < string_length + 1) {
+    return end_write(EXESS_NO_SPACE, buf_size, buf, 0);
+  }
+
+  memcpy(buf, string, string_length + 1);
+  return result(EXESS_SUCCESS, string_length);
+}
diff --git a/subprojects/exess/src/write_utils.h b/subprojects/exess/src/write_utils.h
new file mode 100644
index 00000000..ea941743
--- /dev/null
+++ b/subprojects/exess/src/write_utils.h
@@ -0,0 +1,85 @@
+/*
+  Copyright 2019-2021 David Robillard <d@drobilla.net>
+
+  Permission to use, copy, modify, and/or distribute this software for any
+  purpose with or without fee is hereby granted, provided that the above
+  copyright notice and this permission notice appear in all copies.
+
+  THIS SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+  WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+  MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+  ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+  WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+  ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+  OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+*/
+
+#ifndef EXESS_WRITE_UTILS_H
+#define EXESS_WRITE_UTILS_H
+
+#include "exess/exess.h"
+
+#include <stddef.h>
+#include <stdint.h>
+#include <string.h>
+
+static inline size_t
+write_char(const char c, size_t buf_size, char* const buf, const size_t i)
+{
+  if (buf && buf_size >= i + 1) {
+    buf[i] = c;
+  }
+
+  return 1;
+}
+
+static inline size_t
+write_string(const size_t len,
+             const char*  str,
+             const size_t buf_size,
+             char* const  buf,
+             const size_t i)
+{
+  if (buf && buf_size >= i + len + 1) {
+    memcpy(buf + i, str, len);
+    buf[i + len] = 0;
+  }
+
+  return len;
+}
+
+static inline ExessResult
+end_write(const ExessStatus status,
+          const size_t      buf_size,
+          char* const       buf,
+          const size_t      i)
+{
+  ExessResult r = {status, status > EXESS_EXPECTED_END ? 0 : i};
+
+  if (buf) {
+    if (!status && i >= buf_size) {
+      r.status = EXESS_NO_SPACE;
+      r.count  = 0;
+    }
+
+    if (r.count < buf_size) {
+      buf[r.count] = '\0';
+    }
+  }
+
+  return r;
+}
+
+ExessResult
+write_digits(uint64_t value, size_t buf_size, char* buf, size_t i);
+
+size_t
+write_two_digit_number(uint8_t value, size_t buf_size, char* buf, size_t i);
+
+ExessResult
+write_special(size_t      string_length,
+              const char* string,
+              size_t      buf_size,
+              char*       buf);
+
+#endif // EXESS_WRITE_UTILS_H
diff --git a/subprojects/exess/src/year.c b/subprojects/exess/src/year.c
new file mode 100644
index 00000000..e268ff80
--- /dev/null
+++ b/subprojects/exess/src/year.c
@@ -0,0 +1,98 @@
+/*
+  Copyright 2019-2021 David Robillard <d@drobilla.net>
+
+  Permission to use, copy, modify, and/or distribute this software for any
+  purpose with or without fee is hereby granted, provided that the above
+  copyright notice and this permission notice appear in all copies.
+
+  THIS SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+  WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+  MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+  ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+  WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+  ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+  OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+*/
+
+#include "date_utils.h"
+#include "int_math.h"
+#include "macros.h"
+#include "read_utils.h"
+#include "write_utils.h"
+
+#include "exess/exess.h"
+
+#include <stdbool.h>
+#include <stdint.h>
+#include <stdlib.h>
+
+ExessResult
+read_year_number(int16_t* const out, const char* const str)
+{
+  *out = 0;
+
+  // Read leading sign if present
+  size_t i    = 0;
+  int    sign = 1;
+  if (str[i] == '-') {
+    sign = -1;
+    ++i;
+  }
+
+  // Read digits
+  uint64_t    magnitude = 0;
+  ExessResult r         = exess_read_ulong(&magnitude, str + i);
+  if (r.status > EXESS_EXPECTED_END) {
+    return result(r.status, i + r.count);
+  }
+
+  i += r.count;
+
+  if (sign > 0) {
+    if (magnitude > (uint16_t)INT16_MAX) {
+      return result(EXESS_OUT_OF_RANGE, i);
+    }
+
+    *out = (int16_t)magnitude;
+  } else {
+    const uint16_t min_magnitude = (uint16_t)(-(INT16_MIN + 1)) + 1;
+    if (magnitude > min_magnitude) {
+      return result(EXESS_OUT_OF_RANGE, i);
+    }
+
+    if (magnitude == min_magnitude) {
+      *out = INT16_MIN;
+    } else {
+      *out = (int16_t)-magnitude;
+    }
+  }
+
+  return result(r.count >= 4 ? EXESS_SUCCESS : EXESS_EXPECTED_DIGIT, i);
+}
+
+ExessResult
+write_year_number(const int16_t value, const size_t buf_size, char* const buf)
+{
+  const uint32_t abs_year    = (uint32_t)abs(value);
+  const uint8_t  n_digits    = exess_num_digits(abs_year);
+  const bool     is_negative = value < 0;
+
+  if (!buf) {
+    return result(EXESS_SUCCESS, is_negative + MAX(4u, n_digits));
+  }
+
+  // Write sign
+  size_t i = 0;
+  if (is_negative) {
+    i += write_char('-', buf_size, buf, i);
+  }
+
+  // Write leading zeros to ensure we have at least 4 year digits
+  for (size_t j = n_digits; j < 4; ++j) {
+    i += write_char('0', buf_size, buf, i);
+  }
+
+  const ExessResult yr = exess_write_uint(abs_year, buf_size - i, buf + i);
+
+  return end_write(yr.status, buf_size, buf, i + yr.count);
+}
diff --git a/subprojects/exess/test/.clang-tidy b/subprojects/exess/test/.clang-tidy
new file mode 100644
index 00000000..98645dc0
--- /dev/null
+++ b/subprojects/exess/test/.clang-tidy
@@ -0,0 +1,11 @@
+Checks: >
+  *,
+  -*-magic-numbers,
+  -*-uppercase-literal-suffix,
+  -clang-analyzer-nullability.NullableDereferenced,
+  -clang-analyzer-security.insecureAPI.DeprecatedOrUnsafeBufferHandling,
+  -llvm-header-guard,
+  -llvmlibc-*,
+WarningsAsErrors: '*'
+HeaderFilterRegex: '.*'
+FormatStyle: file
diff --git a/subprojects/exess/test/float_test_data.h b/subprojects/exess/test/float_test_data.h
new file mode 100644
index 00000000..eaea65c4
--- /dev/null
+++ b/subprojects/exess/test/float_test_data.h
@@ -0,0 +1,130 @@
+/*
+  Copyright 2019-2021 David Robillard <d@drobilla.net>
+
+  Permission to use, copy, modify, and/or distribute this software for any
+  purpose with or without fee is hereby granted, provided that the above
+  copyright notice and this permission notice appear in all copies.
+
+  THIS SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+  WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+  MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+  ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+  WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+  ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+  OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+*/
+
+#include "attributes.h"
+#include "ieee_float.h"
+#include "warnings.h"
+
+#include <assert.h>
+#include <math.h>
+#include <stdint.h>
+#include <string.h>
+
+/// Return the float with representation `rep`
+static inline float
+float_from_rep(const uint32_t rep)
+{
+  float f = 0.0f;
+  memcpy(&f, &rep, sizeof(f));
+  return f;
+}
+
+/// Return the double with representation `rep`
+static inline double
+double_from_rep(const uint64_t rep)
+{
+  double d = 0.0;
+  memcpy(&d, &rep, sizeof(d));
+  return d;
+}
+
+/// Return the distance between two doubles in ULPs
+static EXESS_I_PURE_FUNC uint64_t
+double_ulp_distance(const double a, const double b)
+{
+  assert(a >= 0.0);
+  assert(b >= 0.0);
+
+  const uint64_t ia = double_to_rep(a);
+  const uint64_t ib = double_to_rep(b);
+  if (ia == ib) {
+    return 0;
+  }
+
+  EXESS_DISABLE_CONVERSION_WARNINGS
+  if (isnan(a) || isnan(b) || isinf(a) || isinf(b)) {
+    return UINT64_MAX;
+  }
+  EXESS_RESTORE_WARNINGS
+
+  return ia > ib ? ia - ib : ib - ia;
+}
+
+/// Return the distance between two floats in ULPs
+static EXESS_I_PURE_FUNC uint32_t
+float_ulp_distance(const float a, const float b)
+{
+  assert(a >= 0.0f);
+  assert(b >= 0.0f);
+
+  const uint32_t ia = float_to_rep(a);
+  const uint32_t ib = float_to_rep(b);
+  if (ia == ib) {
+    return 0;
+  }
+
+  EXESS_DISABLE_CONVERSION_WARNINGS
+  if (isnan(a) || isnan(b) || isinf(a) || isinf(b)) {
+    return UINT32_MAX;
+  }
+  EXESS_RESTORE_WARNINGS
+
+  return ia > ib ? ia - ib : ib - ia;
+}
+
+static inline bool
+float_matches(const float a, const float b)
+{
+  EXESS_DISABLE_CONVERSION_WARNINGS
+  const bool a_is_nan      = isnan(a);
+  const bool a_is_negative = signbit(a);
+  const bool b_is_nan      = isnan(b);
+  const bool b_is_negative = signbit(b);
+  EXESS_RESTORE_WARNINGS
+
+  if (a_is_nan && b_is_nan) {
+    return true;
+  }
+
+  if (a_is_nan || b_is_nan || a_is_negative != b_is_negative) {
+    return false;
+  }
+
+  return a_is_negative ? float_ulp_distance(-a, -b) == 0
+                       : float_ulp_distance(a, b) == 0;
+}
+
+static inline bool
+double_matches(const double a, const double b)
+{
+  EXESS_DISABLE_CONVERSION_WARNINGS
+  const bool a_is_nan      = isnan(a);
+  const bool a_is_negative = signbit(a);
+  const bool b_is_nan      = isnan(b);
+  const bool b_is_negative = signbit(b);
+  EXESS_RESTORE_WARNINGS
+
+  if (a_is_nan && b_is_nan) {
+    return true;
+  }
+
+  if (a_is_nan || b_is_nan || a_is_negative != b_is_negative) {
+    return false;
+  }
+
+  return a_is_negative ? double_ulp_distance(-a, -b) == 0
+                       : double_ulp_distance(a, b) == 0;
+}
diff --git a/subprojects/exess/test/int_test_data.h b/subprojects/exess/test/int_test_data.h
new file mode 100644
index 00000000..978f863b
--- /dev/null
+++ b/subprojects/exess/test/int_test_data.h
@@ -0,0 +1,42 @@
+/*
+  Copyright 2019-2021 David Robillard <d@drobilla.net>
+
+  Permission to use, copy, modify, and/or distribute this software for any
+  purpose with or without fee is hereby granted, provided that the above
+  copyright notice and this permission notice appear in all copies.
+
+  THIS SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+  WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+  MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+  ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+  WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+  ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+  OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+*/
+
+#ifndef INT_TEST_DATA_H
+#define INT_TEST_DATA_H
+
+#include <stdint.h>
+
+/// Linear Congruential Generator for making random 32-bit integers
+static inline uint32_t
+lcg32(const uint32_t i)
+{
+  static const uint32_t a = 134775813u;
+  static const uint32_t c = 1u;
+
+  return (a * i) + c;
+}
+
+/// Linear Congruential Generator for making random 64-bit integers
+static inline uint64_t
+lcg64(const uint64_t i)
+{
+  static const uint64_t a = 6364136223846793005ull;
+  static const uint64_t c = 1ull;
+
+  return (a * i) + c;
+}
+
+#endif // INT_TEST_DATA_H
diff --git a/subprojects/exess/test/meson.build b/subprojects/exess/test/meson.build
new file mode 100644
index 00000000..972793f0
--- /dev/null
+++ b/subprojects/exess/test/meson.build
@@ -0,0 +1,76 @@
+autoship = find_program('autoship', required: false)
+
+test_args = []
+
+if get_option('strict')
+  if cc.get_id() == 'clang'
+    test_args = [
+      '-Wno-float-equal',
+    ]
+  elif cc.get_id() == 'gcc'
+    test_args = [
+      '-Wno-float-equal',
+      '-Wno-suggest-attribute=pure',
+    ]
+  elif cc.get_id() == 'msvc'
+    test_args = [
+      '/wd4996',  # POSIX name is deprecated
+    ]
+  endif
+endif
+
+private_tests = [
+  'bigint',
+  'int_math',
+]
+
+foreach unit : private_tests
+  test(unit,
+       executable('test_@0@'.format(unit),
+                  'test_@0@.c'.format(unit),
+                  c_args: exess_c_args + prog_args + test_args,
+                  dependencies: exess_static_dep,
+                  include_directories: include_directories('../src')),
+       suite: 'private')
+endforeach
+
+public_tests = [
+  'base64',
+  'boolean',
+  'byte',
+  'canonical',
+  'coerce',
+  'datatype',
+  'date',
+  'datetime',
+  'decimal',
+  'double',
+  'duration',
+  'float',
+  'hex',
+  'int',
+  'long',
+  'short',
+  'strerror',
+  'time',
+  'timezone',
+  'ubyte',
+  'uint',
+  'ulong',
+  'ushort',
+  'variant',
+]
+
+foreach unit : public_tests
+  test(unit,
+       executable('test_@0@'.format(unit),
+                  'test_@0@.c'.format(unit),
+                  c_args: exess_c_args + prog_args + test_args,
+                  dependencies: [m_dep, exess_dep],
+                  include_directories: include_directories('../src')),
+       suite: 'public')
+endforeach
+
+if autoship.found()
+  test('autoship', autoship, args: ['test', exess_src_root], suite: 'data')
+endif
diff --git a/subprojects/exess/test/num_test_utils.h b/subprojects/exess/test/num_test_utils.h
new file mode 100644
index 00000000..ca29a479
--- /dev/null
+++ b/subprojects/exess/test/num_test_utils.h
@@ -0,0 +1,85 @@
+/*
+  Copyright 2011-2021 David Robillard <d@drobilla.net>
+
+  Permission to use, copy, modify, and/or distribute this software for any
+  purpose with or without fee is hereby granted, provided that the above
+  copyright notice and this permission notice appear in all copies.
+
+  THIS SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+  WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+  MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+  ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+  WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+  ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+  OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+*/
+
+#include <stdbool.h>
+#include <stdint.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <time.h>
+
+#ifdef _WIN32
+#  include <process.h>
+#else
+#  include <unistd.h>
+#endif
+
+typedef struct {
+  size_t   n_tests;
+  uint32_t seed;
+  bool     exhaustive;
+  bool     error;
+} ExessNumTestOptions;
+
+static bool
+print_num_test_usage(const char* const name)
+{
+  fprintf(stderr, "Usage: %s [OPTION]...\n", name);
+  fprintf(stderr, "  -n NUM_TESTS Number of random tests to run.\n");
+  fprintf(stderr, "  -s SEED      Use random seed.\n");
+  fprintf(stderr, "  -x           Exhaustively test numbers.\n");
+  return true;
+}
+
+static ExessNumTestOptions
+parse_num_test_options(const int argc, char* const* const argv)
+{
+  ExessNumTestOptions opts = {
+    16384u, (uint32_t)time(NULL) + (uint32_t)getpid(), false, false};
+
+  int a = 1;
+  for (; a < argc && argv[a][0] == '-'; ++a) {
+    if (argv[a][1] == 'x') {
+      opts.exhaustive = true;
+    } else if (argv[a][1] == 's') {
+      if (++a == argc) {
+        opts.error = print_num_test_usage(argv[0]);
+        break;
+      }
+
+      opts.seed = (uint32_t)strtol(argv[a], NULL, 10);
+    } else if (argv[a][1] == 'n') {
+      if (++a == argc) {
+        opts.error = print_num_test_usage(argv[0]);
+        break;
+      }
+
+      opts.n_tests = (uint32_t)strtol(argv[a], NULL, 10);
+    } else {
+      opts.error = print_num_test_usage(argv[0]);
+      break;
+    }
+  }
+
+  return opts;
+}
+
+static void
+print_num_test_progress(const uint64_t i, const uint64_t n_tests)
+{
+  if (i % (n_tests / 20) == 1) {
+    fprintf(stderr, "%f%%\n", (double)i / (double)n_tests * 100.0);
+  }
+}
diff --git a/subprojects/exess/test/test_base64.c b/subprojects/exess/test/test_base64.c
new file mode 100644
index 00000000..760f44fe
--- /dev/null
+++ b/subprojects/exess/test/test_base64.c
@@ -0,0 +1,196 @@
+/*
+  Copyright 2011-2021 David Robillard <d@drobilla.net>
+
+  Permission to use, copy, modify, and/or distribute this software for any
+  purpose with or without fee is hereby granted, provided that the above
+  copyright notice and this permission notice appear in all copies.
+
+  THIS SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+  WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+  MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+  ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+  WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+  ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+  OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+*/
+
+#undef NDEBUG
+
+#include "exess/exess.h"
+
+#include <assert.h>
+#include <stddef.h>
+#include <stdint.h>
+#include <stdlib.h>
+#include <string.h>
+
+static void
+check_read(const char* const string,
+           const ExessStatus expected_status,
+           const size_t      expected_value_length,
+           const char* const expected_value,
+           const size_t      expected_value_size,
+           const size_t      expected_count)
+{
+  char      buf[9] = {0, 0, 0, 0, 0, 0, 0, 0, 0};
+  ExessBlob blob   = {9, buf};
+
+  ExessResult r = exess_read_base64(&blob, string);
+  assert(r.status == expected_status);
+  assert(r.count == expected_count);
+  assert(r.status || blob.size == expected_value_size);
+  if (expected_value_length > 0) {
+    assert(!strncmp(buf, expected_value, expected_value_length));
+    assert(blob.size <= exess_base64_decoded_size(strlen(string)));
+  }
+}
+
+static void
+test_rfc4648_cases(void)
+{
+  char buf[9] = {1, 2, 3, 4, 5, 6, 7, 8, 9};
+
+  assert(!exess_write_base64(6, "foobar", sizeof(buf), buf).status);
+  assert(!strcmp(buf, "Zm9vYmFy"));
+
+  assert(!exess_write_base64(5, "fooba", sizeof(buf), buf).status);
+  assert(!strcmp(buf, "Zm9vYmE="));
+
+  assert(!exess_write_base64(4, "foob", sizeof(buf), buf).status);
+  assert(!strcmp(buf, "Zm9vYg=="));
+
+  assert(!exess_write_base64(3, "foo", sizeof(buf), buf).status);
+  assert(!strcmp(buf, "Zm9v"));
+
+  assert(!exess_write_base64(2, "fo", sizeof(buf), buf).status);
+  assert(!strcmp(buf, "Zm8="));
+
+  assert(!exess_write_base64(1, "f", sizeof(buf), buf).status);
+  assert(!strcmp(buf, "Zg=="));
+}
+
+static void
+test_whitespace(void)
+{
+  check_read("Zm9vYmFy", EXESS_SUCCESS, 6, "foobar", 6, 8);
+  check_read(" Zm9vYmFy", EXESS_SUCCESS, 6, "foobar", 6, 9);
+  check_read("Z\fm9vYmFy", EXESS_SUCCESS, 6, "foobar", 6, 9);
+  check_read("Zm\n9vYmFy", EXESS_SUCCESS, 6, "foobar", 6, 9);
+  check_read("Zm9\rvYmFy", EXESS_SUCCESS, 6, "foobar", 6, 9);
+  check_read("Zm9v\tYmFy", EXESS_SUCCESS, 6, "foobar", 6, 9);
+  check_read("Zm9vY\vmFy", EXESS_SUCCESS, 6, "foobar", 6, 9);
+  check_read(" \f\n\r\t\vZm9vYmFy", EXESS_SUCCESS, 6, "foobar", 6, 14);
+  check_read("Zm9vYmFy \f\n\r\t\v", EXESS_SUCCESS, 6, "foobar", 6, 14);
+}
+
+static void
+test_syntax_errors(void)
+{
+  check_read("Z", EXESS_EXPECTED_BASE64, 0, NULL, 0, 1);
+  check_read("ZZ", EXESS_EXPECTED_BASE64, 0, NULL, 0, 2);
+  check_read("ZZZ", EXESS_EXPECTED_BASE64, 0, NULL, 0, 3);
+
+  check_read("=ZZZ", EXESS_BAD_VALUE, 0, NULL, 0, 4);
+  check_read("Z=ZZ", EXESS_BAD_VALUE, 0, NULL, 0, 4);
+  check_read("ZZ=Z", EXESS_BAD_VALUE, 0, NULL, 0, 4);
+
+  check_read("!m9vYmFy", EXESS_EXPECTED_BASE64, 0, NULL, 0, 0);
+  check_read("Z!9vYmFy", EXESS_EXPECTED_BASE64, 0, NULL, 0, 1);
+  check_read("Zm!vYmFy", EXESS_EXPECTED_BASE64, 0, NULL, 0, 2);
+  check_read("Zm9!YmFy", EXESS_EXPECTED_BASE64, 0, NULL, 0, 3);
+  check_read("Zm9v!mFy", EXESS_EXPECTED_BASE64, 0, NULL, 3, 4);
+  check_read("Zm9vY!Fy", EXESS_EXPECTED_BASE64, 0, NULL, 3, 5);
+  check_read("Zm9vYm!y", EXESS_EXPECTED_BASE64, 0, NULL, 3, 6);
+  check_read("Zm9vYmF!", EXESS_EXPECTED_BASE64, 0, NULL, 3, 7);
+}
+
+static void
+test_read_overflow(void)
+{
+  char      buf[3] = {0, 0, 0};
+  ExessBlob blob0  = {0, buf};
+  ExessBlob blob1  = {1, buf};
+  ExessBlob blob2  = {2, buf};
+  ExessBlob blob3  = {3, buf};
+
+  ExessResult r = exess_read_base64(&blob0, "Zm9v");
+  assert(r.status == EXESS_NO_SPACE);
+  assert(r.count == 4);
+  assert(blob0.size == 0);
+
+  r = exess_read_base64(&blob1, "Zm9v");
+  assert(r.status == EXESS_NO_SPACE);
+  assert(r.count == 4);
+  assert(!buf[0]);
+
+  r = exess_read_base64(&blob2, "Zm9v");
+  assert(r.status == EXESS_NO_SPACE);
+  assert(r.count == 4);
+  assert(!buf[0]);
+
+  r = exess_read_base64(&blob3, "Zm9v");
+  assert(r.status == EXESS_SUCCESS);
+  assert(r.count == 4);
+  assert(blob3.size == 3);
+  assert(!strncmp(buf, "foo", 3));
+}
+
+static void
+test_write_overflow(void)
+{
+  char buf[5] = {1, 2, 3, 4, 5};
+
+  assert(exess_write_base64(3, "foo", 0, buf).status == EXESS_NO_SPACE);
+  assert(exess_write_base64(3, "foo", 1, buf).status == EXESS_NO_SPACE);
+  assert(exess_write_base64(3, "foo", 2, buf).status == EXESS_NO_SPACE);
+  assert(exess_write_base64(3, "foo", 3, buf).status == EXESS_NO_SPACE);
+  assert(exess_write_base64(3, "foo", 4, buf).status == EXESS_NO_SPACE);
+  assert(exess_write_base64(3, "foo", 5, buf).status == EXESS_SUCCESS);
+}
+
+static void
+test_round_trip(void)
+{
+  for (size_t size = 1; size < 256; ++size) {
+    // Allocate and generate data
+    uint8_t* const data = (uint8_t*)malloc(size);
+    for (size_t i = 0; i < size; ++i) {
+      data[i] = (uint8_t)((size + i) % 256);
+    }
+
+    // Allocate buffer for encoding with minimum required size
+    const size_t str_len = exess_write_base64(size, data, 0, NULL).count;
+    char* const  str     = (char*)malloc(str_len + 1);
+
+    // Encode data to string buffer
+    assert(!exess_write_base64(size, data, str_len + 1, str).status);
+    assert(strlen(str) == str_len);
+    assert(str_len % 4 == 0);
+
+    // Allocate buffer for decoded data with the same size as the input
+    uint8_t* const decoded      = (uint8_t*)malloc(size);
+    ExessBlob      decoded_blob = {size, decoded};
+
+    // Decode and check that data matches the original input
+    assert(!exess_read_base64(&decoded_blob, str).status);
+    assert(decoded_blob.size == size);
+    assert(!memcmp(decoded, data, size));
+
+    free(decoded);
+    free(str);
+    free(data);
+  }
+}
+
+int
+main(void)
+{
+  test_rfc4648_cases();
+  test_whitespace();
+  test_syntax_errors();
+  test_read_overflow();
+  test_write_overflow();
+  test_round_trip();
+
+  return 0;
+}
diff --git a/subprojects/exess/test/test_bigint.c b/subprojects/exess/test/test_bigint.c
new file mode 100644
index 00000000..3a74bb92
--- /dev/null
+++ b/subprojects/exess/test/test_bigint.c
@@ -0,0 +1,841 @@
+/*
+  Copyright 2011-2021 David Robillard <d@drobilla.net>
+
+  Permission to use, copy, modify, and/or distribute this software for any
+  purpose with or without fee is hereby granted, provided that the above
+  copyright notice and this permission notice appear in all copies.
+
+  THIS SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+  WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+  MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+  ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+  WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+  ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+  OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+*/
+
+#undef NDEBUG
+
+#include "bigint.h"
+
+#include <assert.h>
+#include <stdbool.h>
+#include <stdint.h>
+
+/* Some test data borrowed from http://github.com/google/double-conversion
+   which uses a completely different bigint representation (so if these agree,
+   everything is probably fine).  Others cases are either made up to hit the
+   edges of the implementation, or interesting cases collected from testing the
+   decimal implementation.  Almost everything here uses the hex representation
+   so it is easy to dump into Python as a sanity check. */
+
+static ExessBigint
+bigint_from_hex(const char* str)
+{
+  ExessBigint num;
+  exess_bigint_set_hex_string(&num, str);
+  return num;
+}
+
+static bool
+check_hex_equals(const char* const str, const ExessBigint* const num)
+{
+  const ExessBigint expected = bigint_from_hex(str);
+
+  return exess_bigint_compare(&expected, num) == 0;
+}
+
+#define CHECK_HEXEQ(str, num) assert(check_hex_equals(str, num))
+
+#define CHECK_SET(setter, value, expected)  \
+  do {                                      \
+    ExessBigint num;                        \
+    exess_bigint_set_##setter(&num, value); \
+    CHECK_HEXEQ(expected, &num);            \
+  } while (0)
+
+static void
+test_set(void)
+{
+  CHECK_SET(u32, 0, "0");
+  CHECK_SET(u32, 0xA, "A");
+  CHECK_SET(u32, 0x20, "20");
+  CHECK_SET(u32, 0x12345678, "12345678");
+
+  CHECK_SET(u64, 0, "0");
+  CHECK_SET(u64, 0xA, "A");
+  CHECK_SET(u64, 0x20, "20");
+  CHECK_SET(u64, 0x12345678, "12345678");
+  CHECK_SET(u64, 0xFFFFFFFFFFFFFFFFull, "FFFFFFFFFFFFFFFF");
+  CHECK_SET(u64, 0x123456789ABCDEF0ull, "123456789ABCDEF0");
+  CHECK_SET(u64, 0x123456789ABCDEF0ull, "123456789ABCDEF0");
+
+  CHECK_SET(decimal_string, "0", "0");
+  CHECK_SET(decimal_string, "1", "1");
+  CHECK_SET(decimal_string, "01234567890", "499602D2");
+  CHECK_SET(decimal_string, "12345.67890", "499602D2");
+  CHECK_SET(decimal_string, "12345.67890EOF", "499602D2");
+  CHECK_SET(decimal_string, "012345678901", "2DFDC1C35");
+  CHECK_SET(decimal_string, "12345.678901", "2DFDC1C35");
+  CHECK_SET(decimal_string,
+            "340282366920938463463374607431768211456",
+            "100000000000000000000000000000000");
+
+  CHECK_SET(hex_string, "0", "0");
+  CHECK_SET(hex_string, "123456789ABCDEF0", "123456789ABCDEF0");
+
+  const ExessBigint orig = bigint_from_hex("123456789ABCDEF01");
+  ExessBigint       copy;
+  exess_bigint_set(&copy, &orig);
+  CHECK_HEXEQ("123456789ABCDEF01", &copy);
+}
+
+static void
+check_left_shifted_bigit(const char*    value,
+                         const unsigned amount,
+                         const unsigned index,
+                         const Bigit    expected)
+{
+  const ExessBigint num = bigint_from_hex(value);
+  const Bigit actual    = exess_bigint_left_shifted_bigit(&num, amount, index);
+
+  assert(expected == actual);
+}
+
+static void
+test_left_shifted_bigit(void)
+{
+  check_left_shifted_bigit("0", 100, 1, 0x0);
+  check_left_shifted_bigit("1", 0, 0, 0x1);
+  check_left_shifted_bigit("1", 1, 0, 0x2);
+  check_left_shifted_bigit("1", 4, 0, 0x10);
+  check_left_shifted_bigit("1", 32, 0, 0x0);
+  check_left_shifted_bigit("1", 32, 1, 0x1);
+  check_left_shifted_bigit("1", 64, 0, 0x0);
+  check_left_shifted_bigit("1", 64, 1, 0x0);
+  check_left_shifted_bigit("1", 64, 2, 0x1);
+  check_left_shifted_bigit("123456789ABCDEF", 64, 0, 0x0);
+  check_left_shifted_bigit("123456789ABCDEF", 64, 1, 0x0);
+  check_left_shifted_bigit("123456789ABCDEF", 64, 2, 0x89ABCDEF);
+  check_left_shifted_bigit("123456789ABCDEF", 64, 3, 0x1234567);
+  check_left_shifted_bigit("123456789ABCDEF", 64, 4, 0x0);
+  check_left_shifted_bigit("123456789ABCDEF", 65, 0, 0x0);
+  check_left_shifted_bigit("123456789ABCDEF", 65, 1, 0x0);
+  check_left_shifted_bigit("123456789ABCDEF", 65, 2, 0x13579BDE);
+  check_left_shifted_bigit("123456789ABCDEF", 65, 3, 0x2468ACF);
+  check_left_shifted_bigit("123456789ABCDEF", 65, 4, 0x0);
+}
+
+static void
+check_shift_left(const char* value, const unsigned amount, const char* expected)
+{
+  ExessBigint num = bigint_from_hex(value);
+  exess_bigint_shift_left(&num, amount);
+  CHECK_HEXEQ(expected, &num);
+}
+
+static void
+test_shift_left(void)
+{
+  check_shift_left("0", 100, "0");
+  check_shift_left("1", 1, "2");
+  check_shift_left("1", 4, "10");
+  check_shift_left("1", 32, "100000000");
+  check_shift_left("1", 64, "10000000000000000");
+  check_shift_left("123456789ABCDEF", 0, "123456789ABCDEF");
+  check_shift_left("123456789ABCDEF", 64, "123456789ABCDEF0000000000000000");
+  check_shift_left("123456789ABCDEF", 65, "2468ACF13579BDE0000000000000000");
+  check_shift_left("16B8B5E06EDC79", 23, "B5C5AF0376E3C800000");
+}
+
+static void
+check_add_u32(const char* value, const uint32_t rhs, const char* expected)
+{
+  ExessBigint num = bigint_from_hex(value);
+  exess_bigint_add_u32(&num, rhs);
+  CHECK_HEXEQ(expected, &num);
+}
+
+static void
+test_add_u32(void)
+{
+  check_add_u32("0", 1, "1");
+  check_add_u32("1", 1, "2");
+  check_add_u32("FFFFFFF", 1, "10000000");
+  check_add_u32("FFFFFFFFFFFFFF", 1, "100000000000000");
+
+  check_add_u32("10000000000000000000000000000000000080000000",
+                0x80000000,
+                "10000000000000000000000000000000000100000000");
+
+  check_add_u32("10000000000000000000000000000000000000000000",
+                0x1,
+                "10000000000000000000000000000000000000000001");
+}
+
+static void
+check_add(const char* lhs_hex, const char* rhs_hex, const char* expected)
+{
+  ExessBigint       lhs = bigint_from_hex(lhs_hex);
+  const ExessBigint rhs = bigint_from_hex(rhs_hex);
+
+  exess_bigint_add(&lhs, &rhs);
+  CHECK_HEXEQ(expected, &lhs);
+}
+
+static void
+test_add(void)
+{
+  check_add("1", "0", "1");
+  check_add("1", "1", "2");
+  check_add("FFFFFFF", "1", "10000000");
+  check_add("FFFFFFFFFFFFFF", "1", "100000000000000");
+  check_add("1", "1000000000000", "1000000000001");
+  check_add("FFFFFFF", "1000000000000", "100000FFFFFFF");
+
+  check_add("10000000000000000000000000000000000000000000",
+            "1",
+            "10000000000000000000000000000000000000000001");
+
+  check_add("10000000000000000000000000000000000000000000",
+            "1000000000000",
+            "10000000000000000000000000000001000000000000");
+
+  check_add("FFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFF",
+            "1000000000000",
+            "1000000000000000000000000000000FFFFFFFFFFFF");
+
+  check_add("10000000000000000000000000",
+            "1000000000000",
+            "10000000000001000000000000");
+
+  check_add(
+    "1", "10000000000000000000000000000", "10000000000000000000000000001");
+
+  check_add("FFFFFFF",
+            "10000000000000000000000000000",
+            "1000000000000000000000FFFFFFF");
+
+  check_add("10000000000000000000000000000000000000000000",
+            "10000000000000000000000000000",
+            "10000000000000010000000000000000000000000000");
+
+  check_add("FFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFF",
+            "10000000000000000000000000000",
+            "100000000000000FFFFFFFFFFFFFFFFFFFFFFFFFFFF");
+
+  check_add("10000000000000000000000000",
+            "10000000000000000000000000000",
+            "10010000000000000000000000000");
+}
+
+static void
+check_subtract(const char* lhs_hex, const char* rhs_hex, const char* expected)
+{
+  ExessBigint       lhs = bigint_from_hex(lhs_hex);
+  const ExessBigint rhs = bigint_from_hex(rhs_hex);
+
+  exess_bigint_subtract(&lhs, &rhs);
+  CHECK_HEXEQ(expected, &lhs);
+}
+
+static void
+test_subtract(void)
+{
+  check_subtract("1", "0", "1");
+  check_subtract("2", "0", "2");
+  check_subtract("10000000", "1", "FFFFFFF");
+  check_subtract("1FFFFFFFF00000000", "FFFFFFFF", "1FFFFFFFE00000001");
+  check_subtract("100000000000000", "1", "FFFFFFFFFFFFFF");
+  check_subtract("1000000000001", "1000000000000", "1");
+  check_subtract("100000FFFFFFF", "1000000000000", "FFFFFFF");
+
+  check_subtract(
+    "11F2678326EA00000000", "0878678326EAC9000000", "979FFFFFFFF37000000");
+
+  check_subtract("10000000000000000000000000000000000000000001",
+                 "00000000000000000000000000000000000000000001",
+                 "10000000000000000000000000000000000000000000");
+
+  check_subtract("10000000000000000000000000000001000000000000",
+                 "00000000000000000000000000000001000000000000",
+                 "10000000000000000000000000000000000000000000");
+
+  check_subtract("1000000000000000000000000000000FFFFFFFFFFFF",
+                 "0000000000000000000000000000001000000000000",
+                 " FFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFF");
+
+  check_subtract("10000000000000000000000000",
+                 "00000000000001000000000000",
+                 " FFFFFFFFFFFFF000000000000");
+
+  check_subtract("10000000000000000000000000",
+                 "1000000000000000000000000",
+                 "F000000000000000000000000");
+
+  check_subtract("FFFFFFF000000000000000",
+                 "0000000000000800000000",
+                 "FFFFFFEFFFFFF800000000");
+
+  check_subtract("10000000000000000000000000000000000000000000",
+                 "00000000000000000000000000000000000800000000",
+                 "FFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFF800000000");
+
+  check_subtract("FFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFF",
+                 "000000000000000000000000000000000800000000",
+                 "FFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFF7FFFFFFFF");
+}
+
+static void
+check_subtract_left_shifted(const char*    lhs_hex,
+                            const char*    rhs_hex,
+                            const unsigned amount,
+                            const char*    expected)
+{
+  ExessBigint       lhs = bigint_from_hex(lhs_hex);
+  const ExessBigint rhs = bigint_from_hex(rhs_hex);
+
+  exess_bigint_subtract_left_shifted(&lhs, &rhs, amount);
+  CHECK_HEXEQ(expected, &lhs);
+}
+
+static void
+test_subtract_left_shifted(void)
+{
+  check_subtract_left_shifted("1", "0", 1, "1");
+  check_subtract_left_shifted("10000000", "1", 1, "FFFFFFE");
+  check_subtract_left_shifted("100000000", "40000000", 2, "0");
+  check_subtract_left_shifted("1000000000000000", "400000000000000", 2, "0");
+  check_subtract_left_shifted("1000000000000000", "800000000000000", 1, "0");
+  check_subtract_left_shifted("1000000000000000", "F", 16, "FFFFFFFFFF10000");
+  check_subtract_left_shifted("1000000000000000", "F", 24, "FFFFFFFF1000000");
+  check_subtract_left_shifted("100000000000000", "1", 0, "FFFFFFFFFFFFFF");
+  check_subtract_left_shifted("100000000000000", "1", 56, "0");
+
+  check_subtract_left_shifted(
+    "11F2678326EA00000000", "43C33C1937564800000", 1, "979FFFFFFFF37000000");
+}
+
+static void
+check_multiply_u32(const char* value, const uint32_t rhs, const char* expected)
+{
+  ExessBigint num = bigint_from_hex(value);
+  exess_bigint_multiply_u32(&num, rhs);
+  CHECK_HEXEQ(expected, &num);
+}
+
+static void
+test_multiply_u32(void)
+{
+  check_multiply_u32("0", 0x25, "0");
+  check_multiply_u32("123456789ABCDEF", 0, "0");
+  check_multiply_u32("2", 0x5, "A");
+  check_multiply_u32("10000000", 0x9, "90000000");
+  check_multiply_u32("100000000000000", 0xFFFF, "FFFF00000000000000");
+  check_multiply_u32("100000000000000", 0xFFFFFFFF, "FFFFFFFF00000000000000");
+  check_multiply_u32("1234567ABCD", 0xFFF, "12333335552433");
+  check_multiply_u32("1234567ABCD", 0xFFFFFFF, "12345679998A985433");
+  check_multiply_u32("FFFFFFFFFFFFFFFF", 0x2, "1FFFFFFFFFFFFFFFE");
+  check_multiply_u32("FFFFFFFFFFFFFFFF", 0x4, "3FFFFFFFFFFFFFFFC");
+  check_multiply_u32("FFFFFFFFFFFFFFFF", 0xF, "EFFFFFFFFFFFFFFF1");
+  check_multiply_u32("FFFFFFFFFFFFFFFF", 0xFFFFFF, "FFFFFEFFFFFFFFFF000001");
+  check_multiply_u32("377654D193A171", 10000000, "210EDD6D4CDD2580EE80");
+  check_multiply_u32("2E3F36D108373C00000", 10, "1CE78242A52285800000");
+
+  check_multiply_u32(
+    "10000000000000000000000000", 0x00000002, "20000000000000000000000000");
+
+  check_multiply_u32(
+    "10000000000000000000000000", 0x0000000F, "F0000000000000000000000000");
+
+  check_multiply_u32("FFFF0000000000000000000000000",
+                     0xFFFF,
+                     "FFFE00010000000000000000000000000");
+
+  check_multiply_u32("FFFF0000000000000000000000000",
+                     0xFFFFFFFF,
+                     "FFFEFFFF00010000000000000000000000000");
+
+  check_multiply_u32("FFFF0000000000000000000000000",
+                     0xFFFFFFFF,
+                     "FFFEFFFF00010000000000000000000000000");
+}
+
+static void
+check_multiply_u64(const char* value, const uint64_t rhs, const char* expected)
+{
+  ExessBigint num = bigint_from_hex(value);
+  exess_bigint_multiply_u64(&num, rhs);
+  CHECK_HEXEQ(expected, &num);
+}
+
+static void
+test_multiply_u64(void)
+{
+  check_multiply_u64("0", 0x25, "0");
+  check_multiply_u64("123456789ABCDEF", 0, "0");
+  check_multiply_u64("123456789ABCDEF", 1, "123456789ABCDEF");
+  check_multiply_u64("2", 0x5, "A");
+  check_multiply_u64("10000000", 0x9, "90000000");
+  check_multiply_u64("100000000000000", 0xFFFF, "FFFF00000000000000");
+  check_multiply_u64("1234567ABCD", 0xFFF, "12333335552433");
+  check_multiply_u64("1234567ABCD", 0xFFFFFFFFFFull, "1234567ABCBDCBA985433");
+  check_multiply_u64("FFFFFFFFFFFFFFFF", 0x2, "1FFFFFFFFFFFFFFFE");
+  check_multiply_u64("FFFFFFFFFFFFFFFF", 0x4, "3FFFFFFFFFFFFFFFC");
+  check_multiply_u64("FFFFFFFFFFFFFFFF", 0xF, "EFFFFFFFFFFFFFFF1");
+
+  check_multiply_u64(
+    "100000000000000", 0xFFFFFFFFFFFFFFFFull, "FFFFFFFFFFFFFFFF00000000000000");
+
+  check_multiply_u64("FFFFFFFFFFFFFFFF",
+                     0xFFFFFFFFFFFFFFFFull,
+                     "FFFFFFFFFFFFFFFE0000000000000001");
+
+  check_multiply_u64(
+    "10000000000000000000000000", 0x00000002, "20000000000000000000000000");
+
+  check_multiply_u64(
+    "10000000000000000000000000", 0x0000000F, "F0000000000000000000000000");
+
+  check_multiply_u64("FFFF0000000000000000000000000",
+                     0xFFFF,
+                     "FFFE00010000000000000000000000000");
+
+  check_multiply_u64("FFFF0000000000000000000000000",
+                     0xFFFFFFFF,
+                     "FFFEFFFF00010000000000000000000000000");
+
+  check_multiply_u64("FFFF0000000000000000000000000",
+                     0xFFFFFFFFFFFFFFFFull,
+                     "FFFEFFFFFFFFFFFF00010000000000000000000000000");
+
+  check_multiply_u64(
+    "377654D193A171", 0x8AC7230489E80000ull, "1E10EE4B11D15A7F3DE7F3C7680000");
+}
+
+static void
+check_multiply_pow10(const char*    value,
+                     const unsigned exponent,
+                     const char*    expected)
+{
+  ExessBigint num = bigint_from_hex(value);
+  exess_bigint_multiply_pow10(&num, exponent);
+  CHECK_HEXEQ(expected, &num);
+}
+
+static void
+test_multiply_pow10(void)
+{
+  check_multiply_pow10("0", 10, "0");
+  check_multiply_pow10("1234", 0, "1234");
+  check_multiply_pow10("4D2", 1, "3034");
+  check_multiply_pow10("4D2", 2, "1E208");
+  check_multiply_pow10("4D2", 3, "12D450");
+  check_multiply_pow10("4D2", 4, "BC4B20");
+  check_multiply_pow10("4D2", 5, "75AEF40");
+  check_multiply_pow10("4D2", 6, "498D5880");
+  check_multiply_pow10("4D2", 7, "2DF857500");
+  check_multiply_pow10("4D2", 8, "1CBB369200");
+  check_multiply_pow10("4D2", 9, "11F5021B400");
+  check_multiply_pow10("4D2", 10, "B3921510800");
+  check_multiply_pow10("4D2", 11, "703B4D2A5000");
+  check_multiply_pow10("4D2", 12, "4625103A72000");
+  check_multiply_pow10("4D2", 13, "2BD72A24874000");
+  check_multiply_pow10("4D2", 14, "1B667A56D488000");
+  check_multiply_pow10("4D2", 15, "11200C7644D50000");
+  check_multiply_pow10("4D2", 16, "AB407C9EB0520000");
+  check_multiply_pow10("4D2", 17, "6B084DE32E3340000");
+  check_multiply_pow10("4D2", 18, "42E530ADFCE0080000");
+  check_multiply_pow10("4D2", 19, "29CF3E6CBE0C0500000");
+  check_multiply_pow10("4D2", 20, "1A218703F6C783200000");
+  check_multiply_pow10("4D2", 21, "1054F4627A3CB1F400000");
+  check_multiply_pow10("4D2", 22, "A3518BD8C65EF38800000");
+  check_multiply_pow10("4D2", 23, "6612F7677BFB5835000000");
+  check_multiply_pow10("4D2", 24, "3FCBDAA0AD7D17212000000");
+  check_multiply_pow10("4D2", 25, "27DF68A46C6E2E74B4000000");
+  check_multiply_pow10("4D2", 26, "18EBA166C3C4DD08F08000000");
+  check_multiply_pow10("4D2", 27, "F9344E03A5B0A259650000000");
+  check_multiply_pow10("4D2", 28, "9BC0B0C2478E6577DF20000000");
+  check_multiply_pow10("4D2", 29, "61586E796CB8FF6AEB740000000");
+  check_multiply_pow10("4D2", 30, "3CD7450BE3F39FA2D32880000000");
+  check_multiply_pow10("4D2", 31, "26068B276E7843C5C3F9500000000");
+
+  check_multiply_pow10("4D2",
+                       50,
+                       "149D1B4CFED03B23AB5F4E1196EF45C0"
+                       "8000000000000");
+
+  check_multiply_pow10("4D2",
+                       100,
+                       "5827249F27165024FBC47DFCA9359BF3"
+                       "16332D1B91ACEECF471FBAB06D9B2000"
+                       "0000000000000000000000");
+
+  check_multiply_pow10("4D2",
+                       305,
+                       "AFBA390D657B0829339F5B98DC852A89"
+                       "682758E01829EADFD016D1528D4D548B"
+                       "80894B9ED9C2EC6A9CABB4881302A637"
+                       "9FF3058908FEAC310C52FCA009799718"
+                       "8260B0B2E2EC96E471B7892AD9B4F9F9"
+                       "A448CBF150D2E87F3934000000000000"
+                       "00000000000000000000000000000000"
+                       "00000000000000000000000000000000");
+
+  check_multiply_pow10("123456789ABCDEF0", 0, "123456789ABCDEF0");
+  check_multiply_pow10("123456789ABCDEF0",
+                       44,
+                       "51A1AD66ACE4E5C79209330F58F52DE3"
+                       "7CEFFF1F000000000000");
+  check_multiply_pow10("123456789ABCDEF0",
+                       88,
+                       "16E0C6D18F4BFA7D0289B88382F56151"
+                       "EB9DA5DB09D56C9BA5D8305619CEE057"
+                       "4F00000000000000000000000");
+  check_multiply_pow10("123456789ABCDEF0",
+                       132,
+                       "6696B1DA27BEA173B5EFCAABBB8492A9"
+                       "2AE3D97F7EE3C7314FB7E2FF8AEFD329"
+                       "F5F8202C22650BB79A7D9F3867F00000"
+                       "00000000000000000000000000000");
+  check_multiply_pow10("123456789ABCDEF0",
+                       176,
+                       "1CC05FF0499D8BC7D8EBE0C6DC2FDC09"
+                       "E93765F3448235FB16AD09D98BBB3A0A"
+                       "843372D33A318EE63DAE6998DA59EF34"
+                       "B15C40A65B9B65ABF3CAF00000000000"
+                       "00000000000000000000000000000000"
+                       "00");
+  check_multiply_pow10("123456789ABCDEF0",
+                       220,
+                       "80ED0FD9A6C0F56A495F466320D34E22"
+                       "507FAA83F0519E7FF909FDDBDA184682"
+                       "BB70D38D43284C828A3681540722E550"
+                       "960567BAB1C25389C1BE7705228BE8CC"
+                       "AF3EBD382829DF000000000000000000"
+                       "00000000000000000000000000000000"
+                       "000000");
+  check_multiply_pow10("123456789ABCDEF0",
+                       264,
+                       "2421FD0F55C486D05211339D45EC2DC4"
+                       "12AE7A64DDFE619DA81B73C069088D3E"
+                       "83D7AA9F99B571815DE939A5275FB4A6"
+                       "9D8930798C01FB96781B9D633BB59AD5"
+                       "A7F322A7EC14154D1B8B5DF1718779A5"
+                       "2291FE0F000000000000000000000000"
+                       "00000000000000000000000000000000"
+                       "00000000000");
+  check_multiply_pow10("123456789ABCDEF0",
+                       308,
+                       "A206620F35C83E9E780ECC07DCAF13BB"
+                       "0A7EE2E213747914340BC172D783BA56"
+                       "661E8DCFFD03C398BD66F5570F445AC6"
+                       "737126283C64AE1A289B9D8BB4531033"
+                       "8C3E34DE2D534187092ABA1F4706100E"
+                       "ECF66D14059461A05A9BEBBCCBA0F693"
+                       "F0000000000000000000000000000000"
+                       "00000000000000000000000000000000"
+                       "000000000000000");
+}
+
+static void
+check_divmod(const char*    lhs_hex,
+             const char*    rhs_hex,
+             const uint32_t expected_divisor,
+             const char*    expected_mod_hex)
+{
+  ExessBigint       lhs     = bigint_from_hex(lhs_hex);
+  const ExessBigint rhs     = bigint_from_hex(rhs_hex);
+  const uint32_t    divisor = exess_bigint_divmod(&lhs, &rhs);
+
+  assert(divisor == expected_divisor);
+  CHECK_HEXEQ(expected_mod_hex, &lhs);
+}
+
+static void
+test_divmod(void)
+{
+  check_divmod("A", "2", 5, "0");
+  check_divmod("B", "2", 5, "1");
+  check_divmod("C", "2", 6, "0");
+  check_divmod("A", "1234567890", 0, "A");
+  check_divmod("FFFFFFFF", "3", 0x55555555, "0");
+  check_divmod("12345678", "3789012", 5, "D9861E");
+  check_divmod("70000001", "1FFFFFFF", 3, "10000004");
+  check_divmod("28000000", "12A05F20", 2, "2BF41C0");
+  check_divmod("FFFFFFFFF", "FFFFFFFF", 16, "F");
+  check_divmod("100000000000001", "FFFFFFF", 0x10000001, "2");
+  check_divmod("40000000000002", "2FAF0800000000", 1, "1050F800000002");
+  check_divmod("40000000000000", "40000000000000", 1, "0");
+
+  check_divmod("43DE72C3DF858FC278A361EEB5A000000",
+               "80000000000000000000000000000000",
+               8,
+               "3DE72C3DF858FC278A361EEB5A000000");
+
+  check_divmod(
+    "B5C5AF0376E3C800000", "43C33C1937564800000", 2, "2E3F36D108373800000");
+
+  check_divmod("A0000000000000000000000000000000"
+               "00000000000000000000000000000000"
+               "00000000000000000000000000000000"
+               "000000000000000000000000000000",
+               "20000000000000000000000000000000"
+               "00000000000000000000000000000000"
+               "00000000000000000000000000000000"
+               "000000000000000000000000000000",
+               5,
+               "0");
+
+  check_divmod("A0000000000000000000000000000000"
+               "00000000000000000000000000000000"
+               "00000000000000000000000000000000"
+               "000000000000000000000000000001",
+               "20000000000000000000000000000000"
+               "00000000000000000000000000000000"
+               "00000000000000000000000000000000"
+               "000000000000000000000000000000",
+               5,
+               "1");
+
+  check_divmod("B6080000000000000000000000000000"
+               "00000000000000000000000000000000"
+               "00000000000000000000000000000000"
+               "000000000000000000000000000000FF"
+               "F",
+               "A0000000000000000000000000000000"
+               "00000000000000000000000000000000"
+               "00000000000000000000000000000000"
+               "000000000000000000000000000000",
+               0x1234,
+               "FFF");
+
+  check_divmod("B6080000000000000000000000000000"
+               "00000000000000000000000000000000"
+               "00000000000000000000000000000000"
+               "00000000000000000000000000000000"
+               "0",
+               "9FFFFFFFFFFFFFFFFFFFFFFFFFFFFFFF"
+               "FFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFF"
+               "FFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFF"
+               "FFFFFFFFFFFFFFFFFFFFFFFFFFF001",
+               0x1234,
+               "1232DCC");
+}
+
+static void
+check_compare(const char* lhs_hex, const char* rhs_hex, const int expected_cmp)
+{
+  const ExessBigint lhs = bigint_from_hex(lhs_hex);
+  const ExessBigint rhs = bigint_from_hex(rhs_hex);
+  const int         cmp = exess_bigint_compare(&lhs, &rhs);
+
+  assert(cmp == expected_cmp);
+
+  if (cmp) {
+    const int rcmp = exess_bigint_compare(&rhs, &lhs);
+    assert(rcmp == -cmp);
+  }
+}
+
+static void
+test_compare(void)
+{
+  check_compare("1", "1", 0);
+  check_compare("0", "1", -1);
+  check_compare("F", "FF", -1);
+  check_compare("F", "FFFFFFFFF", -1);
+  check_compare("10000000000000000", "10000000000000000", 0);
+  check_compare("10000000000000000", "10000000000000001", -1);
+  check_compare("FFFFFFFFFFFFFFFFF", "100000000000000000", -1);
+  check_compare("10000000000000000", "10000000000000001", -1);
+  check_compare("1234567890ABCDEF12345", "1234567890ABCDEF12345", 0);
+  check_compare("1234567890ABCDEF12345", "1234567890ABCDEF12346", -1);
+
+  const char* const huge = "123456789ABCDEF0123456789ABCDEF0"
+                           "123456789ABCDEF0123456789ABCDEF0"
+                           "123456789ABCDEF0123456789ABCDEF0"
+                           "123456789ABCDEF0123456789ABCDEF0";
+
+  const char* const huger = "123456789ABCDEF0123456789ABCDEF0"
+                            "123456789ABCDEF0123456789ABCDEF0"
+                            "123456789ABCDEF0123456789ABCDEF0"
+                            "123456789ABCDEF0123456789ABCDEF1";
+
+  check_compare(huge, huge, 0);
+  check_compare(huger, huger, 0);
+  check_compare(huge, huger, -1);
+}
+
+static void
+check_plus_compare(const char* l_hex,
+                   const char* p_hex,
+                   const char* c_hex,
+                   const int   expected_cmp)
+{
+  const ExessBigint l    = bigint_from_hex(l_hex);
+  const ExessBigint p    = bigint_from_hex(p_hex);
+  const ExessBigint c    = bigint_from_hex(c_hex);
+  const int         cmp  = exess_bigint_plus_compare(&l, &p, &c);
+  const int         rcmp = exess_bigint_plus_compare(&p, &l, &c);
+
+  assert(cmp == expected_cmp);
+  assert(rcmp == expected_cmp);
+}
+
+static void
+test_plus_compare(void)
+{
+  check_plus_compare("1", "0", "1", 0);
+  check_plus_compare("0", "0", "1", -1);
+  check_plus_compare("FFFFFFFFF", "F", "F", 1);
+  check_plus_compare("F", "F", "800000000", -1);
+  check_plus_compare("F", "F", "80000000000000000", -1);
+  check_plus_compare("800000000", "F", "80000000000000000", -1);
+  check_plus_compare("2D79883D20000", "2D79883D20000", "5AF3107A40000", 0);
+  check_plus_compare("20000000000000", "1", "20000000000000", +1);
+
+  check_plus_compare(
+    "0588A503282FE00000", "0588A503282FE00000", "0AD78EBC5AC6200000", +1);
+
+  check_plus_compare("2F06018572BEADD1280000000",
+                     "0204FCE5E3E25026110000000",
+                     "4000000000000000000000000",
+                     -1);
+
+  check_plus_compare("1234567890ABCDEF12345",
+                     "000000000000000000001",
+                     "1234567890ABCDEF12345",
+                     +1);
+
+  check_plus_compare("1234567890ABCDEF12344",
+                     "000000000000000000001",
+                     "1234567890ABCDEF12345",
+                     0);
+
+  check_plus_compare("123456789000000000000",
+                     "0000000000ABCDEF12345",
+                     "1234567890ABCDEF12345",
+                     0);
+
+  check_plus_compare("123456789000000000000",
+                     "0000000000ABCDEF12344",
+                     "1234567890ABCDEF12345",
+                     -1);
+
+  check_plus_compare("123456789000000000000",
+                     "0000000000ABCDEF12346",
+                     "1234567890ABCDEF12345",
+                     1);
+
+  check_plus_compare("123456789100000000000",
+                     "0000000000ABCDEF12345",
+                     "1234567890ABCDEF12345",
+                     1);
+
+  check_plus_compare("123456788900000000000",
+                     "0000000000ABCDEF12345",
+                     "1234567890ABCDEF12345",
+                     -1);
+
+  check_plus_compare("12345678900000000000000000000",
+                     "0000000000ABCDEF1234500000000",
+                     "1234567890ABCDEF1234500000000",
+                     0);
+
+  check_plus_compare("12345678900000000000000000000",
+                     "0000000000ABCDEF1234400000000",
+                     "1234567890ABCDEF1234500000000",
+                     -1);
+
+  check_plus_compare("12345678900000000000000000000",
+                     "0000000000ABCDEF1234600000000",
+                     "1234567890ABCDEF1234500000000",
+                     1);
+
+  check_plus_compare("12345678910000000000000000000",
+                     "0000000000ABCDEF1234500000000",
+                     "1234567890ABCDEF1234500000000",
+                     1);
+  check_plus_compare("12345678890000000000000000000",
+                     "0000000000ABCDEF1234500000000",
+                     "1234567890ABCDEF1234500000000",
+                     -1);
+
+  check_plus_compare("12345678900000000000000000000",
+                     "000000000000000000ABCDEF12345",
+                     "123456789000000000ABCDEF12345",
+                     0);
+
+  check_plus_compare("12345678900000000000000000000",
+                     "000000000000000000ABCDEF12346",
+                     "123456789000000000ABCDEF12345",
+                     1);
+
+  check_plus_compare("12345678900000000000000000000",
+                     "000000000000000000ABCDEF12344",
+                     "123456789000000000ABCDEF12345",
+                     -1);
+
+  check_plus_compare("12345678900000000000000000000",
+                     "000000000000000000ABCDEF12345",
+                     "12345678900000ABCDEF123450000",
+                     -1);
+
+  check_plus_compare("12345678900000000000000000000",
+                     "000000000000000000ABCDEF12344",
+                     "12345678900000ABCDEF123450000",
+                     -1);
+
+  check_plus_compare("12345678900000000000000000000",
+                     "000000000000000000ABCDEF12345",
+                     "12345678900000ABCDEF123450001",
+                     -1);
+
+  check_plus_compare("12345678900000000000000000000",
+                     "00000000000000ABCDEF123460000",
+                     "12345678900000ABCDEF123450000",
+                     1);
+}
+
+static void
+check_pow10(const unsigned exponent, const char* expected)
+{
+  ExessBigint num;
+  exess_bigint_set_pow10(&num, exponent);
+  CHECK_HEXEQ(expected, &num);
+}
+
+static void
+test_set_pow10(void)
+{
+  check_pow10(0, "1");
+  check_pow10(1, "A");
+  check_pow10(2, "64");
+  check_pow10(5, "186A0");
+  check_pow10(8, "5F5E100");
+  check_pow10(16, "2386F26FC10000");
+  check_pow10(30, "C9F2C9CD04674EDEA40000000");
+  check_pow10(31, "7E37BE2022C0914B2680000000");
+}
+
+int
+main(void)
+{
+  test_set();
+  test_left_shifted_bigit();
+  test_shift_left();
+  test_add_u32();
+  test_add();
+  test_subtract();
+  test_subtract_left_shifted();
+  test_multiply_u32();
+  test_multiply_u64();
+  test_multiply_pow10();
+  test_divmod();
+  test_compare();
+  test_plus_compare();
+  test_set_pow10();
+
+  return 0;
+}
diff --git a/subprojects/exess/test/test_boolean.c b/subprojects/exess/test/test_boolean.c
new file mode 100644
index 00000000..65eb9b24
--- /dev/null
+++ b/subprojects/exess/test/test_boolean.c
@@ -0,0 +1,114 @@
+/*
+  Copyright 2011-2021 David Robillard <d@drobilla.net>
+
+  Permission to use, copy, modify, and/or distribute this software for any
+  purpose with or without fee is hereby granted, provided that the above
+  copyright notice and this permission notice appear in all copies.
+
+  THIS SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+  WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+  MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+  ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+  WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+  ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+  OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+*/
+
+#undef NDEBUG
+
+#include "exess/exess.h"
+
+#include <assert.h>
+#include <stdbool.h>
+#include <stddef.h>
+#include <string.h>
+
+static void
+check_read(const char* const string,
+           const ExessStatus expected_status,
+           const bool        expected_value,
+           const size_t      expected_count)
+{
+  bool value = false;
+
+  const ExessResult r = exess_read_boolean(&value, string);
+  assert(value == expected_value);
+  assert(r.status == expected_status);
+  assert(r.count == expected_count);
+}
+
+static void
+test_read_boolean(void)
+{
+  // No input
+  check_read("", EXESS_EXPECTED_BOOLEAN, false, 0);
+  check_read(" \f\n\r\t\v", EXESS_EXPECTED_BOOLEAN, false, 6);
+
+  // Canonical form
+  check_read("false", EXESS_SUCCESS, false, 5);
+  check_read("true", EXESS_SUCCESS, true, 4);
+
+  // Non-canonical form
+  check_read("0", EXESS_SUCCESS, false, 1);
+  check_read("1", EXESS_SUCCESS, true, 1);
+  check_read(" \f\n\r\t\vfalse ", EXESS_SUCCESS, false, 11);
+  check_read(" \f\n\r\t\vtrue ", EXESS_SUCCESS, true, 10);
+  check_read(" \f\n\r\t\v0 ", EXESS_SUCCESS, false, 7);
+  check_read(" \f\n\r\t\v1 ", EXESS_SUCCESS, true, 7);
+
+  // Trailing garbage
+  check_read("falsely", EXESS_EXPECTED_END, false, 5);
+  check_read("truely", EXESS_EXPECTED_END, true, 4);
+  check_read("0no", EXESS_EXPECTED_END, false, 1);
+  check_read("1yes", EXESS_EXPECTED_END, true, 1);
+
+  // Garbage
+  check_read("twue", EXESS_EXPECTED_BOOLEAN, false, 0);
+  check_read("fawse", EXESS_EXPECTED_BOOLEAN, false, 0);
+  check_read("tr", EXESS_EXPECTED_BOOLEAN, false, 0);
+  check_read("fa", EXESS_EXPECTED_BOOLEAN, false, 0);
+  check_read("yes", EXESS_EXPECTED_BOOLEAN, false, 0);
+  check_read("no", EXESS_EXPECTED_BOOLEAN, false, 0);
+}
+
+static void
+check_write(const bool        value,
+            const ExessStatus expected_status,
+            const size_t      buf_size,
+            const char* const expected_string)
+{
+  char buf[EXESS_MAX_BOOLEAN_LENGTH + 1] = {1, 2, 3, 4, 5, 6};
+
+  assert(buf_size <= sizeof(buf));
+
+  const ExessResult r = exess_write_boolean(value, buf_size, buf);
+  assert(!strcmp(buf, expected_string));
+  assert(r.status == expected_status);
+  assert(r.count == strlen(buf));
+  assert(r.status || exess_write_boolean(value, 0, NULL).count == r.count);
+}
+
+static void
+test_write_boolean(void)
+{
+  check_write(true, EXESS_SUCCESS, 5, "true");
+  check_write(false, EXESS_SUCCESS, 6, "false");
+
+  check_write(true, EXESS_NO_SPACE, 4, "");
+  check_write(false, EXESS_NO_SPACE, 5, "");
+
+  // Check that nothing is written when there isn't enough space
+  char              c = 42;
+  const ExessResult r = exess_write_boolean(false, 0, &c);
+  assert(c == 42);
+  assert(r.status == EXESS_NO_SPACE);
+  assert(r.count == 0);
+}
+
+int
+main(void)
+{
+  test_read_boolean();
+  test_write_boolean();
+  return 0;
+}
diff --git a/subprojects/exess/test/test_byte.c b/subprojects/exess/test/test_byte.c
new file mode 100644
index 00000000..a67fd779
--- /dev/null
+++ b/subprojects/exess/test/test_byte.c
@@ -0,0 +1,98 @@
+/*
+  Copyright 2011-2021 David Robillard <d@drobilla.net>
+
+  Permission to use, copy, modify, and/or distribute this software for any
+  purpose with or without fee is hereby granted, provided that the above
+  copyright notice and this permission notice appear in all copies.
+
+  THIS SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+  WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+  MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+  ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+  WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+  ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+  OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+*/
+
+#undef NDEBUG
+
+#include "exess/exess.h"
+
+#include <assert.h>
+#include <stddef.h>
+#include <stdint.h>
+#include <string.h>
+
+static void
+check_read(const char* const string,
+           const ExessStatus expected_status,
+           const int8_t      expected_value,
+           const size_t      expected_count)
+{
+  int8_t            value = 0;
+  const ExessResult r     = exess_read_byte(&value, string);
+
+  assert(r.status == expected_status);
+  assert(r.count == expected_count);
+  assert(value == expected_value);
+}
+
+static void
+test_read_byte(void)
+{
+  // Limits
+  check_read("-128", EXESS_SUCCESS, INT8_MIN, EXESS_MAX_BYTE_LENGTH);
+  check_read("127", EXESS_SUCCESS, INT8_MAX, 3);
+
+  // Out of range
+  check_read("-129", EXESS_OUT_OF_RANGE, 0, 4);
+  check_read("128", EXESS_OUT_OF_RANGE, 0, 3);
+
+  // Garbage
+  check_read("+", EXESS_EXPECTED_DIGIT, 0, 1);
+}
+
+static void
+check_write(const int8_t      value,
+            const ExessStatus expected_status,
+            const size_t      buf_size,
+            const char* const expected_string)
+{
+  char buf[EXESS_MAX_BYTE_LENGTH + 1] = {1, 2, 3, 4, 5};
+  assert(buf_size <= sizeof(buf));
+
+  const ExessResult r = exess_write_byte(value, buf_size, buf);
+  assert(r.status == expected_status);
+  assert(!strcmp(buf, expected_string));
+  assert(r.count == strlen(buf));
+  assert(r.status || exess_write_byte(value, 0, NULL).count == r.count);
+}
+
+static void
+test_write_byte(void)
+{
+  check_write(INT8_MIN, EXESS_SUCCESS, 5, "-128");
+  check_write(INT8_MAX, EXESS_SUCCESS, 4, "127");
+}
+
+static void
+test_round_trip(void)
+{
+  int8_t value                          = 0;
+  char   buf[EXESS_MAX_BYTE_LENGTH + 1] = {1, 2, 3, 4, 5};
+
+  for (int16_t i = INT8_MIN; i <= INT8_MAX; ++i) {
+    assert(!exess_write_byte((int8_t)i, sizeof(buf), buf).status);
+    assert(!exess_read_byte(&value, buf).status);
+    assert(value == i);
+  }
+}
+
+int
+main(void)
+{
+  test_read_byte();
+  test_write_byte();
+  test_round_trip();
+  return 0;
+}
diff --git a/subprojects/exess/test/test_canonical.c b/subprojects/exess/test/test_canonical.c
new file mode 100644
index 00000000..77167fcf
--- /dev/null
+++ b/subprojects/exess/test/test_canonical.c
@@ -0,0 +1,412 @@
+/*
+  Copyright 2011-2021 David Robillard <d@drobilla.net>
+
+  Permission to use, copy, modify, and/or distribute this software for any
+  purpose with or without fee is hereby granted, provided that the above
+  copyright notice and this permission notice appear in all copies.
+
+  THIS SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+  WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+  MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+  ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+  WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+  ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+  OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+*/
+
+#undef NDEBUG
+
+#include "exess/exess.h"
+
+#include <assert.h>
+#include <stddef.h>
+#include <string.h>
+
+static void
+check_write(const ExessDatatype datatype,
+            const char* const   value,
+            const ExessStatus   expected_status,
+            const size_t        buf_size,
+            const char* const   expected_string)
+{
+  char buf[328] = {42};
+
+  assert(buf_size <= sizeof(buf));
+
+  const ExessResult r = exess_write_canonical(value, datatype, buf_size, buf);
+  assert(r.status == expected_status);
+  assert(r.count == strlen(buf));
+  assert(!strcmp(buf, expected_string));
+  assert((r.status && r.status != EXESS_EXPECTED_END) ||
+         exess_write_canonical(value, datatype, 0, NULL).count == r.count);
+}
+
+static void
+test_decimal(void)
+{
+  check_write(EXESS_DECIMAL, "", EXESS_EXPECTED_DIGIT, 1, "");
+  check_write(EXESS_DECIMAL, " \f\n\r\t\v", EXESS_EXPECTED_DIGIT, 1, "");
+
+  check_write(EXESS_DECIMAL, " -001 ", EXESS_SUCCESS, 5, "-1.0");
+  check_write(EXESS_DECIMAL, " -000 ", EXESS_SUCCESS, 5, "-0.0");
+  check_write(EXESS_DECIMAL, "  000 ", EXESS_SUCCESS, 4, "0.0");
+  check_write(EXESS_DECIMAL, "  001 ", EXESS_SUCCESS, 4, "1.0");
+  check_write(EXESS_DECIMAL, " +001 ", EXESS_SUCCESS, 4, "1.0");
+
+  check_write(EXESS_DECIMAL, " -.123", EXESS_SUCCESS, 7, "-0.123");
+  check_write(EXESS_DECIMAL, " +.123", EXESS_SUCCESS, 6, "0.123");
+  check_write(EXESS_DECIMAL, " -0 ", EXESS_SUCCESS, 5, "-0.0");
+  check_write(EXESS_DECIMAL, " +0 ", EXESS_SUCCESS, 4, "0.0");
+
+  check_write(EXESS_DECIMAL, " +00.10 ", EXESS_SUCCESS, 4, "0.1");
+  check_write(EXESS_DECIMAL, " +01 ", EXESS_SUCCESS, 4, "1.0");
+
+  check_write(EXESS_DECIMAL,
+              " 36893488147419103232 ",
+              EXESS_SUCCESS,
+              23,
+              "36893488147419103232.0");
+
+  check_write(EXESS_DECIMAL,
+              " 0036893488147419103232 ",
+              EXESS_SUCCESS,
+              23,
+              "36893488147419103232.0");
+
+  check_write(EXESS_DECIMAL,
+              " +36893488147419103232 ",
+              EXESS_SUCCESS,
+              23,
+              "36893488147419103232.0");
+
+  check_write(EXESS_DECIMAL,
+              " -36893488147419103232 ",
+              EXESS_SUCCESS,
+              24,
+              "-36893488147419103232.0");
+
+  check_write(EXESS_DECIMAL,
+              " +0036893488147419103232 ",
+              EXESS_SUCCESS,
+              23,
+              "36893488147419103232.0");
+
+  check_write(EXESS_DECIMAL,
+              " +0036893488147419103232. ",
+              EXESS_SUCCESS,
+              23,
+              "36893488147419103232.0");
+
+  check_write(EXESS_DECIMAL,
+              " +0036893488147419103232.00 ",
+              EXESS_SUCCESS,
+              23,
+              "36893488147419103232.0");
+
+  check_write(EXESS_DECIMAL,
+              " +0036893488147419103232.12300 ",
+              EXESS_SUCCESS,
+              25,
+              "36893488147419103232.123");
+
+  check_write(EXESS_DECIMAL,
+              " -0036893488147419103232 ",
+              EXESS_SUCCESS,
+              24,
+              "-36893488147419103232.0");
+
+  check_write(EXESS_DECIMAL,
+              " -0036893488147419103232. ",
+              EXESS_SUCCESS,
+              24,
+              "-36893488147419103232.0");
+
+  check_write(EXESS_DECIMAL,
+              " -0036893488147419103232.00 ",
+              EXESS_SUCCESS,
+              24,
+              "-36893488147419103232.0");
+
+  check_write(EXESS_DECIMAL,
+              " -0036893488147419103232.12300 ",
+              EXESS_SUCCESS,
+              26,
+              "-36893488147419103232.123");
+
+  check_write(EXESS_DECIMAL, " -1234extra", EXESS_EXPECTED_END, 8, "-1234.0");
+  check_write(EXESS_DECIMAL, " 1234extra", EXESS_EXPECTED_END, 7, "1234.0");
+
+  check_write(EXESS_DECIMAL, " f", EXESS_EXPECTED_DIGIT, 2, "");
+  check_write(EXESS_DECIMAL, "", EXESS_EXPECTED_DIGIT, 1, "");
+}
+
+static void
+test_integer(void)
+{
+  check_write(EXESS_INTEGER, "", EXESS_EXPECTED_DIGIT, 1, "");
+  check_write(EXESS_INTEGER, " \f\n\r\t\v", EXESS_EXPECTED_DIGIT, 1, "");
+
+  // Integer
+
+  check_write(EXESS_INTEGER, " -001 ", EXESS_SUCCESS, 3, "-1");
+  check_write(EXESS_INTEGER, "  000 ", EXESS_SUCCESS, 2, "0");
+  check_write(EXESS_INTEGER, "  001 ", EXESS_SUCCESS, 2, "1");
+  check_write(EXESS_INTEGER, " +001 ", EXESS_SUCCESS, 2, "1");
+
+  check_write(EXESS_INTEGER, " junk 987654321 ", EXESS_EXPECTED_DIGIT, 2, "");
+  check_write(
+    EXESS_INTEGER, " 987654321 junk ", EXESS_EXPECTED_END, 10, "987654321");
+
+  check_write(EXESS_INTEGER,
+              " 36893488147419103232 ",
+              EXESS_SUCCESS,
+              21,
+              "36893488147419103232");
+
+  check_write(EXESS_INTEGER,
+              " 0036893488147419103232 ",
+              EXESS_SUCCESS,
+              21,
+              "36893488147419103232");
+
+  check_write(EXESS_INTEGER,
+              " +36893488147419103232 ",
+              EXESS_SUCCESS,
+              21,
+              "36893488147419103232");
+
+  check_write(EXESS_INTEGER,
+              " +0036893488147419103232 ",
+              EXESS_SUCCESS,
+              21,
+              "36893488147419103232");
+
+  check_write(EXESS_INTEGER,
+              " -36893488147419103232 ",
+              EXESS_SUCCESS,
+              22,
+              "-36893488147419103232");
+
+  check_write(EXESS_INTEGER,
+              " -0036893488147419103232 ",
+              EXESS_SUCCESS,
+              22,
+              "-36893488147419103232");
+
+  // NonPositiveInteger
+
+  check_write(EXESS_NON_POSITIVE_INTEGER, " -001 ", EXESS_SUCCESS, 3, "-1");
+  check_write(EXESS_NON_POSITIVE_INTEGER, "  000 ", EXESS_SUCCESS, 2, "0");
+  check_write(EXESS_NON_POSITIVE_INTEGER, "  001 ", EXESS_BAD_VALUE, 3, "");
+  check_write(EXESS_NON_POSITIVE_INTEGER, " +001 ", EXESS_BAD_VALUE, 3, "");
+
+  check_write(EXESS_NON_POSITIVE_INTEGER,
+              " -36893488147419103232 ",
+              EXESS_SUCCESS,
+              22,
+              "-36893488147419103232");
+
+  check_write(EXESS_NON_POSITIVE_INTEGER,
+              " -0036893488147419103232 ",
+              EXESS_SUCCESS,
+              22,
+              "-36893488147419103232");
+
+  // NegativeInteger
+
+  check_write(EXESS_NEGATIVE_INTEGER, " -001 ", EXESS_SUCCESS, 3, "-1");
+  check_write(EXESS_NEGATIVE_INTEGER, "  000 ", EXESS_BAD_VALUE, 3, "");
+  check_write(EXESS_NEGATIVE_INTEGER, "  001 ", EXESS_BAD_VALUE, 3, "");
+  check_write(EXESS_NEGATIVE_INTEGER, " +001 ", EXESS_BAD_VALUE, 3, "");
+
+  check_write(EXESS_NEGATIVE_INTEGER,
+              " -36893488147419103232 ",
+              EXESS_SUCCESS,
+              22,
+              "-36893488147419103232");
+
+  check_write(EXESS_NEGATIVE_INTEGER,
+              " -0036893488147419103232 ",
+              EXESS_SUCCESS,
+              22,
+              "-36893488147419103232");
+
+  // NonNegativeInteger
+
+  check_write(EXESS_NON_NEGATIVE_INTEGER, " -001 ", EXESS_BAD_VALUE, 3, "");
+  check_write(EXESS_NON_NEGATIVE_INTEGER, "  000 ", EXESS_SUCCESS, 2, "0");
+  check_write(EXESS_NON_NEGATIVE_INTEGER, "  001 ", EXESS_SUCCESS, 2, "1");
+  check_write(EXESS_NON_NEGATIVE_INTEGER, " +001 ", EXESS_SUCCESS, 2, "1");
+
+  check_write(EXESS_NON_NEGATIVE_INTEGER,
+              " 36893488147419103232 ",
+              EXESS_SUCCESS,
+              21,
+              "36893488147419103232");
+
+  check_write(EXESS_NON_NEGATIVE_INTEGER,
+              " 0036893488147419103232 ",
+              EXESS_SUCCESS,
+              21,
+              "36893488147419103232");
+
+  // PositiveInteger
+
+  check_write(EXESS_POSITIVE_INTEGER, " -001 ", EXESS_BAD_VALUE, 3, "");
+  check_write(EXESS_POSITIVE_INTEGER, "  000 ", EXESS_BAD_VALUE, 3, "");
+  check_write(EXESS_POSITIVE_INTEGER, "  001 ", EXESS_SUCCESS, 2, "1");
+  check_write(EXESS_POSITIVE_INTEGER, " +001 ", EXESS_SUCCESS, 2, "1");
+
+  check_write(EXESS_POSITIVE_INTEGER,
+              " 36893488147419103232 ",
+              EXESS_SUCCESS,
+              21,
+              "36893488147419103232");
+
+  check_write(EXESS_POSITIVE_INTEGER,
+              " 0036893488147419103232 ",
+              EXESS_SUCCESS,
+              21,
+              "36893488147419103232");
+}
+
+static void
+test_fixed_numbers(void)
+{
+  check_write(EXESS_DOUBLE, " +00.10 ", EXESS_SUCCESS, 7, "1.0E-1");
+  check_write(EXESS_FLOAT, " +00.10 ", EXESS_SUCCESS, 14, "1.00000001E-1");
+  check_write(EXESS_BOOLEAN, " 0 ", EXESS_SUCCESS, 6, "false");
+  check_write(EXESS_BOOLEAN, " 1 ", EXESS_SUCCESS, 5, "true");
+  check_write(EXESS_LONG, " +012 ", EXESS_SUCCESS, 3, "12");
+  check_write(EXESS_INT, " +012 ", EXESS_SUCCESS, 3, "12");
+  check_write(EXESS_SHORT, " +012 ", EXESS_SUCCESS, 3, "12");
+  check_write(EXESS_BYTE, " +012 ", EXESS_SUCCESS, 3, "12");
+  check_write(EXESS_ULONG, " 012 ", EXESS_SUCCESS, 3, "12");
+  check_write(EXESS_UINT, " 012 ", EXESS_SUCCESS, 3, "12");
+  check_write(EXESS_USHORT, " 012 ", EXESS_SUCCESS, 3, "12");
+  check_write(EXESS_UBYTE, " 012 ", EXESS_SUCCESS, 3, "12");
+}
+
+static void
+test_time(void)
+{
+  check_write(EXESS_DURATION, " P0Y6M ", EXESS_SUCCESS, 4, "P6M");
+  check_write(EXESS_DURATION, " P1Y6M0D ", EXESS_SUCCESS, 6, "P1Y6M");
+  check_write(EXESS_TIME, " 12:15:01+00:00 ", EXESS_SUCCESS, 14, "12:15:01Z");
+  check_write(
+    EXESS_DATE, " 02004-04-12+00:00 ", EXESS_SUCCESS, 12, "2004-04-12Z");
+}
+
+static void
+test_datetime(void)
+{
+  // Local
+  check_write(EXESS_DATETIME,
+              " 02001-02-03T04:05:06.007 ",
+              EXESS_SUCCESS,
+              26,
+              "2001-02-03T04:05:06.007");
+
+  // Positive carry: minute => hour
+  check_write(EXESS_DATETIME,
+              " 02001-02-03T04:46:59-00:15 ",
+              EXESS_SUCCESS,
+              21,
+              "2001-02-03T05:01:59Z");
+
+  // Positive carry: minute => hour => day
+  check_write(EXESS_DATETIME,
+              " 02001-02-03T23:46:59-00:15 ",
+              EXESS_SUCCESS,
+              21,
+              "2001-02-04T00:01:59Z");
+
+  // Positive carry: minute => hour => day => month (common year)
+  check_write(EXESS_DATETIME,
+              " 02001-02-28T23:46:59-00:15 ",
+              EXESS_SUCCESS,
+              21,
+              "2001-03-01T00:01:59Z");
+
+  // Positive carry: minute => hour => day => month (leap year)
+  check_write(EXESS_DATETIME,
+              " 02000-02-29T23:46:59-00:15 ",
+              EXESS_SUCCESS,
+              21,
+              "2000-03-01T00:01:59Z");
+
+  // Positive carry: minute => hour => day => month => year
+  check_write(EXESS_DATETIME,
+              " 02001-12-31T23:46:59-00:15 ",
+              EXESS_SUCCESS,
+              21,
+              "2002-01-01T00:01:59Z");
+
+  // Negative carry: minute => hour
+  check_write(EXESS_DATETIME,
+              " 02001-02-03T04:14:59+00:15 ",
+              EXESS_SUCCESS,
+              21,
+              "2001-02-03T03:59:59Z");
+
+  // Negative carry: minute => hour => day
+  check_write(EXESS_DATETIME,
+              " 02001-02-02T00:14:59+00:15 ",
+              EXESS_SUCCESS,
+              21,
+              "2001-02-01T23:59:59Z");
+
+  // Negative carry: minute => hour => day => month (common year)
+  check_write(EXESS_DATETIME,
+              " 02001-03-01T00:14:59+00:15 ",
+              EXESS_SUCCESS,
+              21,
+              "2001-02-28T23:59:59Z");
+
+  // Negative carry: minute => hour => day => month (leap year)
+  check_write(EXESS_DATETIME,
+              " 02000-03-01T00:14:59+00:15 ",
+              EXESS_SUCCESS,
+              21,
+              "2000-02-29T23:59:59Z");
+
+  // Negative carry: minute => hour => day => month => year
+  check_write(EXESS_DATETIME,
+              " 02001-01-01T00:14:59+00:15 ",
+              EXESS_SUCCESS,
+              21,
+              "2000-12-31T23:59:59Z");
+}
+
+static void
+test_binary(void)
+{
+  check_write(EXESS_HEX, " D EA  D B3 3F", EXESS_SUCCESS, 9, "DEADB33F");
+  check_write(EXESS_HEX, "invalid", EXESS_EXPECTED_HEX, 1, "");
+  check_write(EXESS_HEX, "1A2B3", EXESS_EXPECTED_HEX, 5, "");
+  check_write(EXESS_HEX, "1", EXESS_EXPECTED_HEX, 1, "");
+  check_write(EXESS_HEX, "", EXESS_EXPECTED_HEX, 5, "");
+
+  check_write(
+    EXESS_BASE64, " Z\fm\n9\rv\tY\vmFy", EXESS_SUCCESS, 9, "Zm9vYmFy");
+  check_write(EXESS_BASE64, "!nvalid", EXESS_EXPECTED_BASE64, 1, "");
+  check_write(EXESS_BASE64, "Z", EXESS_EXPECTED_BASE64, 1, "");
+  check_write(EXESS_BASE64, "Zm", EXESS_EXPECTED_BASE64, 2, "");
+  check_write(EXESS_BASE64, "Zm9", EXESS_EXPECTED_BASE64, 3, "");
+  check_write(EXESS_BASE64, "", EXESS_EXPECTED_BASE64, 5, "");
+}
+
+int
+main(void)
+{
+  check_write(EXESS_NOTHING, "?", EXESS_UNSUPPORTED, 1, "");
+
+  test_decimal();
+  test_integer();
+  test_fixed_numbers();
+  test_time();
+  test_datetime();
+  test_binary();
+
+  return 0;
+}
diff --git a/subprojects/exess/test/test_coerce.c b/subprojects/exess/test/test_coerce.c
new file mode 100644
index 00000000..dc5ad060
--- /dev/null
+++ b/subprojects/exess/test/test_coerce.c
@@ -0,0 +1,519 @@
+/*
+  Copyright 2011-2021 David Robillard <d@drobilla.net>
+
+  Permission to use, copy, modify, and/or distribute this software for any
+  purpose with or without fee is hereby granted, provided that the above
+  copyright notice and this permission notice appear in all copies.
+
+  THIS SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+  WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+  MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+  ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+  WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+  ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+  OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+*/
+
+#undef NDEBUG
+
+#include "exess/exess.h"
+
+#include <assert.h>
+#include <string.h>
+
+static const char* const min_long_str  = "-9223372036854775808";
+static const char* const max_long_str  = "9223372036854775807";
+static const char* const max_ulong_str = "18446744073709551615";
+
+static void
+check_from_to(const ExessDatatype from_datatype,
+              const char* const   from_string,
+              const ExessDatatype to_datatype,
+              const char* const   to_string)
+{
+  char value_buf[4] = {0, 0, 0, 0};
+  char buf[328]     = {42};
+
+  // Read original value
+  ExessVariant value = {EXESS_HEX, {.as_blob = {sizeof(value_buf), value_buf}}};
+  assert(!exess_read_variant(&value, from_datatype, from_string).status);
+
+  // Coerce to target datatype
+  const ExessVariant coerced = exess_coerce(value, to_datatype, EXESS_LOSSLESS);
+  assert(coerced.datatype == to_datatype);
+
+  // Write coerced value and check string against expectation
+  assert(!exess_write_variant(coerced, sizeof(buf), buf).status);
+  assert(!strcmp(buf, to_string));
+
+  // Coerce the value back to the original type
+  const ExessVariant tripped =
+    exess_coerce(coerced, from_datatype, EXESS_LOSSLESS);
+  assert(tripped.datatype == from_datatype);
+
+  // Write round-tripped value and check string against the original
+  assert(!exess_write_variant(tripped, sizeof(buf), buf).status);
+  assert(!strcmp(buf, from_string));
+}
+
+static void
+check_one_way(const ExessDatatype      from_datatype,
+              const char* const        from_string,
+              const ExessCoercionFlags coercions,
+              const ExessDatatype      to_datatype,
+              const char* const        to_string)
+{
+  char buf[328] = {42};
+
+  // Read original value
+  ExessVariant value = {EXESS_NOTHING, {EXESS_SUCCESS}};
+  assert(!exess_read_variant(&value, from_datatype, from_string).status);
+
+  // Coerce to target datatype
+  ExessVariant coerced = exess_coerce(value, to_datatype, coercions);
+  assert(coerced.datatype == to_datatype);
+
+  // Write coerced value and check string against expectation
+  assert(!exess_write_variant(coerced, sizeof(buf), buf).status);
+  assert(!strcmp(buf, to_string));
+}
+
+static void
+check_failure(const ExessDatatype from_datatype,
+              const char* const   from_string,
+              const ExessDatatype to_datatype,
+              const ExessStatus   expected_status)
+{
+  // Read original value
+  ExessVariant value = {EXESS_NOTHING, {EXESS_SUCCESS}};
+  assert(!exess_read_variant(&value, from_datatype, from_string).status);
+
+  // Try to coerce to target datatype
+  const ExessVariant coerced = exess_coerce(value, to_datatype, EXESS_LOSSLESS);
+  assert(coerced.datatype == EXESS_NOTHING);
+  assert(exess_get_status(&coerced) == expected_status);
+}
+
+static void
+test_unknown(void)
+{
+  ExessVariant long_value    = exess_make_long(1);
+  ExessVariant ulong_value   = exess_make_ulong(1u);
+  ExessVariant unknown_value = exess_make_nothing(EXESS_SUCCESS);
+
+  assert(exess_coerce(unknown_value, EXESS_LONG, EXESS_LOSSLESS).datatype ==
+         EXESS_NOTHING);
+
+  assert(exess_coerce(unknown_value, EXESS_ULONG, EXESS_LOSSLESS).datatype ==
+         EXESS_NOTHING);
+
+  assert(exess_coerce(long_value, EXESS_NOTHING, EXESS_LOSSLESS).datatype ==
+         EXESS_NOTHING);
+
+  assert(exess_coerce(ulong_value, EXESS_NOTHING, EXESS_LOSSLESS).datatype ==
+         EXESS_NOTHING);
+}
+
+static void
+test_boolean(void)
+{
+  // Exactly from 0 or 1, lossy from 0 or non-zero
+  check_from_to(EXESS_BOOLEAN, "false", EXESS_FLOAT, "0.0E0");
+  check_from_to(EXESS_BOOLEAN, "true", EXESS_FLOAT, "1.0E0");
+  check_from_to(EXESS_BOOLEAN, "false", EXESS_DOUBLE, "0.0E0");
+  check_from_to(EXESS_BOOLEAN, "true", EXESS_DOUBLE, "1.0E0");
+  check_from_to(EXESS_BOOLEAN, "false", EXESS_LONG, "0");
+  check_from_to(EXESS_BOOLEAN, "true", EXESS_LONG, "1");
+  check_failure(EXESS_LONG, "-1", EXESS_BOOLEAN, EXESS_WOULD_TRUNCATE);
+  check_failure(EXESS_LONG, "2", EXESS_BOOLEAN, EXESS_WOULD_TRUNCATE);
+  check_one_way(EXESS_LONG, "42", EXESS_TRUNCATE, EXESS_BOOLEAN, "true");
+  check_one_way(EXESS_LONG, "-1", EXESS_TRUNCATE, EXESS_BOOLEAN, "true");
+  check_from_to(EXESS_BOOLEAN, "false", EXESS_ULONG, "0");
+  check_from_to(EXESS_BOOLEAN, "true", EXESS_ULONG, "1");
+  check_failure(EXESS_LONG, "2", EXESS_BOOLEAN, EXESS_WOULD_TRUNCATE);
+  check_one_way(EXESS_ULONG, "42", EXESS_TRUNCATE, EXESS_BOOLEAN, "true");
+
+  // Not convertible to any time types
+  check_failure(EXESS_BOOLEAN, "true", EXESS_DURATION, EXESS_UNSUPPORTED);
+  check_failure(EXESS_BOOLEAN, "true", EXESS_DATETIME, EXESS_UNSUPPORTED);
+  check_failure(EXESS_BOOLEAN, "true", EXESS_TIME, EXESS_UNSUPPORTED);
+  check_failure(EXESS_BOOLEAN, "true", EXESS_DATE, EXESS_UNSUPPORTED);
+}
+
+static void
+test_long(void)
+{
+  // Truncating boolean conversion
+  check_one_way(EXESS_LONG, "42", EXESS_TRUNCATE, EXESS_BOOLEAN, "true");
+  check_one_way(EXESS_LONG, "-1", EXESS_TRUNCATE, EXESS_BOOLEAN, "true");
+
+  // All smaller integer types
+  check_from_to(EXESS_LONG, "-2147483648", EXESS_INT, "-2147483648");
+  check_failure(EXESS_LONG, "-2147483649", EXESS_INT, EXESS_OUT_OF_RANGE);
+  check_from_to(EXESS_LONG, "-32768", EXESS_SHORT, "-32768");
+  check_failure(EXESS_LONG, "-32769", EXESS_SHORT, EXESS_OUT_OF_RANGE);
+  check_from_to(EXESS_LONG, "-128", EXESS_BYTE, "-128");
+  check_failure(EXESS_LONG, "-129", EXESS_BYTE, EXESS_OUT_OF_RANGE);
+
+  // Positive values to/from all unsigned types
+  check_from_to(EXESS_LONG, max_long_str, EXESS_ULONG, max_long_str);
+  check_failure(EXESS_LONG, "-1", EXESS_ULONG, EXESS_OUT_OF_RANGE);
+  check_from_to(EXESS_LONG, "4294967295", EXESS_UINT, "4294967295");
+  check_failure(EXESS_LONG, "-1", EXESS_UINT, EXESS_OUT_OF_RANGE);
+  check_from_to(EXESS_LONG, "65535", EXESS_USHORT, "65535");
+  check_failure(EXESS_LONG, "-1", EXESS_USHORT, EXESS_OUT_OF_RANGE);
+  check_from_to(EXESS_LONG, "255", EXESS_UBYTE, "255");
+  check_failure(EXESS_LONG, "-1", EXESS_UBYTE, EXESS_OUT_OF_RANGE);
+
+  // Any value to/from integer
+  check_from_to(EXESS_LONG, min_long_str, EXESS_INTEGER, min_long_str);
+  check_from_to(EXESS_LONG, max_long_str, EXESS_INTEGER, max_long_str);
+
+  // Non-positive values to/from nonPositiveInteger
+  check_from_to(
+    EXESS_LONG, min_long_str, EXESS_NON_POSITIVE_INTEGER, min_long_str);
+  check_from_to(EXESS_LONG, "0", EXESS_NON_POSITIVE_INTEGER, "0");
+  check_failure(
+    EXESS_LONG, "1", EXESS_NON_POSITIVE_INTEGER, EXESS_OUT_OF_RANGE);
+
+  // Negative values to/from negativeInteger
+  check_from_to(EXESS_LONG, min_long_str, EXESS_NEGATIVE_INTEGER, min_long_str);
+  check_from_to(EXESS_LONG, "-1", EXESS_NEGATIVE_INTEGER, "-1");
+  check_failure(EXESS_LONG, "0", EXESS_NEGATIVE_INTEGER, EXESS_OUT_OF_RANGE);
+
+  // Non-negative values to/from nonNegativeInteger
+  check_failure(
+    EXESS_LONG, "-1", EXESS_NON_NEGATIVE_INTEGER, EXESS_OUT_OF_RANGE);
+  check_from_to(EXESS_LONG, "0", EXESS_NON_NEGATIVE_INTEGER, "0");
+  check_from_to(
+    EXESS_LONG, max_long_str, EXESS_NON_NEGATIVE_INTEGER, max_long_str);
+
+  // Positive values to/from positiveInteger
+  check_failure(EXESS_LONG, "-1", EXESS_POSITIVE_INTEGER, EXESS_OUT_OF_RANGE);
+  check_failure(EXESS_LONG, "0", EXESS_POSITIVE_INTEGER, EXESS_OUT_OF_RANGE);
+  check_from_to(EXESS_LONG, max_long_str, EXESS_POSITIVE_INTEGER, max_long_str);
+  check_failure(EXESS_POSITIVE_INTEGER,
+                "9223372036854775808",
+                EXESS_LONG,
+                EXESS_OUT_OF_RANGE);
+
+  // Float
+  check_failure(EXESS_FLOAT, "1.5", EXESS_LONG, EXESS_WOULD_ROUND);
+  check_from_to(EXESS_LONG, "-16777215", EXESS_FLOAT, "-1.6777215E7");
+  check_failure(EXESS_LONG, "-16777216", EXESS_FLOAT, EXESS_OUT_OF_RANGE);
+  check_failure(EXESS_FLOAT, "-16777216", EXESS_LONG, EXESS_OUT_OF_RANGE);
+  check_from_to(EXESS_LONG, "16777215", EXESS_FLOAT, "1.6777215E7");
+  check_failure(EXESS_LONG, "16777216", EXESS_FLOAT, EXESS_OUT_OF_RANGE);
+  check_failure(EXESS_FLOAT, "16777216", EXESS_LONG, EXESS_OUT_OF_RANGE);
+  check_one_way(EXESS_FLOAT, "1.0", EXESS_LOSSLESS, EXESS_LONG, "1");
+  check_one_way(EXESS_FLOAT, "1.5", EXESS_ROUND, EXESS_LONG, "2");
+  check_one_way(EXESS_FLOAT, "2.5", EXESS_ROUND, EXESS_LONG, "2");
+  check_one_way(EXESS_FLOAT, "3.5", EXESS_ROUND, EXESS_LONG, "4");
+
+  // Double
+  check_failure(EXESS_DOUBLE, "1.5", EXESS_LONG, EXESS_WOULD_ROUND);
+  check_from_to(
+    EXESS_LONG, "-9007199254740991", EXESS_DOUBLE, "-9.007199254740991E15");
+  check_failure(
+    EXESS_LONG, "-9007199254740992", EXESS_DOUBLE, EXESS_OUT_OF_RANGE);
+  check_failure(
+    EXESS_DOUBLE, "-9007199254740992", EXESS_LONG, EXESS_OUT_OF_RANGE);
+  check_from_to(
+    EXESS_LONG, "9007199254740991", EXESS_DOUBLE, "9.007199254740991E15");
+  check_failure(
+    EXESS_LONG, "9007199254740992", EXESS_DOUBLE, EXESS_OUT_OF_RANGE);
+  check_failure(
+    EXESS_DOUBLE, "9007199254740992", EXESS_LONG, EXESS_OUT_OF_RANGE);
+  check_one_way(EXESS_DOUBLE, "1.0", EXESS_LOSSLESS, EXESS_LONG, "1");
+  check_one_way(EXESS_DOUBLE, "1.5", EXESS_ROUND, EXESS_LONG, "2");
+  check_one_way(EXESS_DOUBLE, "2.5", EXESS_ROUND, EXESS_LONG, "2");
+  check_one_way(EXESS_DOUBLE, "3.5", EXESS_ROUND, EXESS_LONG, "4");
+}
+
+static void
+test_ulong(void)
+{
+  ExessVariant unknown = {EXESS_NOTHING, {EXESS_SUCCESS}};
+
+  assert(exess_coerce(unknown, EXESS_ULONG, EXESS_LOSSLESS).datatype ==
+         EXESS_NOTHING);
+
+  // Truncating boolean conversion
+  check_one_way(EXESS_ULONG, "42", EXESS_TRUNCATE, EXESS_BOOLEAN, "true");
+
+  // All integer types
+  check_from_to(EXESS_ULONG, max_long_str, EXESS_LONG, max_long_str);
+  check_failure(EXESS_ULONG, max_ulong_str, EXESS_LONG, EXESS_OUT_OF_RANGE);
+  check_from_to(EXESS_ULONG, "2147483647", EXESS_INT, "2147483647");
+  check_failure(EXESS_ULONG, "2147483648", EXESS_INT, EXESS_OUT_OF_RANGE);
+  check_failure(EXESS_INT, "-1", EXESS_ULONG, EXESS_OUT_OF_RANGE);
+  check_from_to(EXESS_ULONG, "32767", EXESS_SHORT, "32767");
+  check_failure(EXESS_ULONG, "32768", EXESS_SHORT, EXESS_OUT_OF_RANGE);
+  check_failure(EXESS_SHORT, "-1", EXESS_ULONG, EXESS_OUT_OF_RANGE);
+  check_from_to(EXESS_ULONG, "127", EXESS_BYTE, "127");
+  check_failure(EXESS_ULONG, "128", EXESS_BYTE, EXESS_OUT_OF_RANGE);
+  check_failure(EXESS_BYTE, "-1", EXESS_ULONG, EXESS_OUT_OF_RANGE);
+
+  // All unsigned types
+  check_from_to(EXESS_ULONG, "4294967295", EXESS_UINT, "4294967295");
+  check_failure(EXESS_ULONG, "4294967296", EXESS_UINT, EXESS_OUT_OF_RANGE);
+  check_from_to(EXESS_ULONG, "65535", EXESS_USHORT, "65535");
+  check_failure(EXESS_ULONG, "65536", EXESS_USHORT, EXESS_OUT_OF_RANGE);
+  check_from_to(EXESS_ULONG, "255", EXESS_UBYTE, "255");
+  check_failure(EXESS_ULONG, "256", EXESS_UBYTE, EXESS_OUT_OF_RANGE);
+
+  // Small enough value Any value to/from integer
+  check_from_to(EXESS_ULONG, "0", EXESS_INTEGER, "0");
+  check_failure(
+    EXESS_ULONG, "9223372036854775808", EXESS_INTEGER, EXESS_OUT_OF_RANGE);
+
+  // Only zero to/from nonPositiveInteger
+  check_from_to(EXESS_ULONG, "0", EXESS_NON_POSITIVE_INTEGER, "0");
+  check_failure(
+    EXESS_ULONG, "1", EXESS_NON_POSITIVE_INTEGER, EXESS_OUT_OF_RANGE);
+
+  // Not convertible to/from negativeInteger
+  check_failure(EXESS_ULONG, "0", EXESS_NEGATIVE_INTEGER, EXESS_OUT_OF_RANGE);
+  check_failure(EXESS_ULONG, "1", EXESS_NEGATIVE_INTEGER, EXESS_OUT_OF_RANGE);
+
+  // Any value to/from nonNegativeInteger
+  check_from_to(EXESS_ULONG, "0", EXESS_NON_NEGATIVE_INTEGER, "0");
+  check_from_to(
+    EXESS_ULONG, max_ulong_str, EXESS_NON_NEGATIVE_INTEGER, max_ulong_str);
+
+  // Positive values to/from positiveInteger
+  check_failure(EXESS_ULONG, "0", EXESS_POSITIVE_INTEGER, EXESS_OUT_OF_RANGE);
+  check_from_to(EXESS_ULONG, "1", EXESS_POSITIVE_INTEGER, "1");
+
+  // Float
+  check_failure(EXESS_FLOAT, "-1", EXESS_ULONG, EXESS_OUT_OF_RANGE);
+  check_failure(EXESS_FLOAT, "1.5", EXESS_ULONG, EXESS_WOULD_ROUND);
+  check_from_to(EXESS_ULONG, "0", EXESS_FLOAT, "0.0E0");
+  check_from_to(EXESS_ULONG, "16777215", EXESS_FLOAT, "1.6777215E7");
+  check_failure(EXESS_ULONG, "16777216", EXESS_FLOAT, EXESS_OUT_OF_RANGE);
+  check_failure(EXESS_FLOAT, "16777216", EXESS_ULONG, EXESS_OUT_OF_RANGE);
+  check_one_way(EXESS_FLOAT, "1.0", EXESS_LOSSLESS, EXESS_ULONG, "1");
+  check_one_way(EXESS_FLOAT, "1.5", EXESS_ROUND, EXESS_ULONG, "2");
+  check_one_way(EXESS_FLOAT, "2.5", EXESS_ROUND, EXESS_ULONG, "2");
+  check_one_way(EXESS_FLOAT, "3.5", EXESS_ROUND, EXESS_ULONG, "4");
+
+  // Double
+  check_failure(EXESS_DOUBLE, "-1", EXESS_ULONG, EXESS_OUT_OF_RANGE);
+  check_failure(EXESS_DOUBLE, "1.5", EXESS_ULONG, EXESS_WOULD_ROUND);
+  check_from_to(EXESS_ULONG, "0", EXESS_DOUBLE, "0.0E0");
+  check_from_to(
+    EXESS_ULONG, "9007199254740991", EXESS_DOUBLE, "9.007199254740991E15");
+  check_failure(
+    EXESS_ULONG, "9007199254740992", EXESS_DOUBLE, EXESS_OUT_OF_RANGE);
+  check_failure(
+    EXESS_DOUBLE, "9007199254740992", EXESS_ULONG, EXESS_OUT_OF_RANGE);
+  check_one_way(EXESS_DOUBLE, "1.0", EXESS_LOSSLESS, EXESS_ULONG, "1");
+  check_one_way(EXESS_DOUBLE, "1.5", EXESS_ROUND, EXESS_ULONG, "2");
+  check_one_way(EXESS_DOUBLE, "2.5", EXESS_ROUND, EXESS_ULONG, "2");
+  check_one_way(EXESS_DOUBLE, "3.5", EXESS_ROUND, EXESS_ULONG, "4");
+}
+
+static void
+test_large_integers(void)
+{
+  check_failure(EXESS_TIME, "00:00:00", EXESS_INTEGER, EXESS_UNSUPPORTED);
+  check_failure(
+    EXESS_TIME, "00:00:00", EXESS_NON_POSITIVE_INTEGER, EXESS_UNSUPPORTED);
+  check_failure(
+    EXESS_TIME, "00:00:00", EXESS_NEGATIVE_INTEGER, EXESS_UNSUPPORTED);
+  check_failure(
+    EXESS_TIME, "00:00:00", EXESS_NON_NEGATIVE_INTEGER, EXESS_UNSUPPORTED);
+  check_failure(
+    EXESS_TIME, "00:00:00", EXESS_POSITIVE_INTEGER, EXESS_UNSUPPORTED);
+}
+
+static void
+test_coerce(void)
+{
+  check_one_way(EXESS_DOUBLE,
+                "1.0000000000001",
+                EXESS_REDUCE_PRECISION,
+                EXESS_FLOAT,
+                "1.0E0");
+
+  check_failure(
+    EXESS_DOUBLE, "1.0000000000001", EXESS_FLOAT, EXESS_WOULD_REDUCE_PRECISION);
+
+  check_one_way(EXESS_FLOAT, "1.5", EXESS_LOSSLESS, EXESS_DOUBLE, "1.5E0");
+
+  /* check_failure( */
+  /*   EXESS_LONG, "9007199254740993", EXESS_DOUBLE, EXESS_OUT_OF_RANGE); */
+
+  /* check_failure(EXESS_FLOAT, "1.0", EXESS_LONG, EXESS_SUCCESS); */
+  /* check_from_to(EXESS_BYTE, "-128", EXESS_LONG, "-128"); */
+
+  // DateTime
+}
+
+static void
+test_date_time(void)
+{
+  check_failure(
+    EXESS_DATETIME, "2001-02-03T04:05:06", EXESS_TIME, EXESS_WOULD_TRUNCATE);
+
+  check_one_way(EXESS_DATETIME,
+                "2001-02-03T04:05:06",
+                EXESS_TRUNCATE,
+                EXESS_TIME,
+                "04:05:06");
+
+  check_one_way(EXESS_DATETIME,
+                "2001-02-03T04:05:06Z",
+                EXESS_TRUNCATE,
+                EXESS_TIME,
+                "04:05:06Z");
+
+  check_failure(
+    EXESS_DATETIME, "2001-02-03T04:05:06", EXESS_DATE, EXESS_WOULD_TRUNCATE);
+
+  check_one_way(EXESS_DATETIME,
+                "2001-02-03T04:05:06",
+                EXESS_TRUNCATE,
+                EXESS_DATE,
+                "2001-02-03");
+
+  check_one_way(EXESS_DATETIME,
+                "2001-02-03T04:05:06Z",
+                EXESS_TRUNCATE,
+                EXESS_DATE,
+                "2001-02-03Z");
+}
+
+static void
+test_number_to_time(void)
+{
+  check_failure(EXESS_BOOLEAN, "true", EXESS_DURATION, EXESS_UNSUPPORTED);
+  check_failure(EXESS_BOOLEAN, "true", EXESS_DATETIME, EXESS_UNSUPPORTED);
+  check_failure(EXESS_BOOLEAN, "true", EXESS_TIME, EXESS_UNSUPPORTED);
+  check_failure(EXESS_BOOLEAN, "true", EXESS_DATE, EXESS_UNSUPPORTED);
+
+  check_failure(EXESS_LONG, "1", EXESS_DURATION, EXESS_UNSUPPORTED);
+  check_failure(EXESS_LONG, "1", EXESS_DATETIME, EXESS_UNSUPPORTED);
+  check_failure(EXESS_LONG, "1", EXESS_TIME, EXESS_UNSUPPORTED);
+  check_failure(EXESS_LONG, "1", EXESS_DATE, EXESS_UNSUPPORTED);
+
+  check_failure(EXESS_INT, "1", EXESS_DURATION, EXESS_UNSUPPORTED);
+  check_failure(EXESS_INT, "1", EXESS_DATETIME, EXESS_UNSUPPORTED);
+  check_failure(EXESS_INT, "1", EXESS_TIME, EXESS_UNSUPPORTED);
+  check_failure(EXESS_INT, "1", EXESS_DATE, EXESS_UNSUPPORTED);
+
+  check_failure(EXESS_SHORT, "1", EXESS_DURATION, EXESS_UNSUPPORTED);
+  check_failure(EXESS_SHORT, "1", EXESS_DATETIME, EXESS_UNSUPPORTED);
+  check_failure(EXESS_SHORT, "1", EXESS_TIME, EXESS_UNSUPPORTED);
+  check_failure(EXESS_SHORT, "1", EXESS_DATE, EXESS_UNSUPPORTED);
+
+  check_failure(EXESS_BYTE, "1", EXESS_DURATION, EXESS_UNSUPPORTED);
+  check_failure(EXESS_BYTE, "1", EXESS_DATETIME, EXESS_UNSUPPORTED);
+  check_failure(EXESS_BYTE, "1", EXESS_TIME, EXESS_UNSUPPORTED);
+  check_failure(EXESS_BYTE, "1", EXESS_DATE, EXESS_UNSUPPORTED);
+
+  check_failure(EXESS_ULONG, "1", EXESS_DURATION, EXESS_UNSUPPORTED);
+  check_failure(EXESS_ULONG, "1", EXESS_DATETIME, EXESS_UNSUPPORTED);
+  check_failure(EXESS_ULONG, "1", EXESS_TIME, EXESS_UNSUPPORTED);
+  check_failure(EXESS_ULONG, "1", EXESS_DATE, EXESS_UNSUPPORTED);
+
+  check_failure(EXESS_UINT, "1", EXESS_DURATION, EXESS_UNSUPPORTED);
+  check_failure(EXESS_UINT, "1", EXESS_DATETIME, EXESS_UNSUPPORTED);
+  check_failure(EXESS_UINT, "1", EXESS_TIME, EXESS_UNSUPPORTED);
+  check_failure(EXESS_UINT, "1", EXESS_DATE, EXESS_UNSUPPORTED);
+
+  check_failure(EXESS_USHORT, "1", EXESS_DURATION, EXESS_UNSUPPORTED);
+  check_failure(EXESS_USHORT, "1", EXESS_DATETIME, EXESS_UNSUPPORTED);
+  check_failure(EXESS_USHORT, "1", EXESS_TIME, EXESS_UNSUPPORTED);
+  check_failure(EXESS_USHORT, "1", EXESS_DATE, EXESS_UNSUPPORTED);
+
+  check_failure(EXESS_UBYTE, "1", EXESS_DURATION, EXESS_UNSUPPORTED);
+  check_failure(EXESS_UBYTE, "1", EXESS_DATETIME, EXESS_UNSUPPORTED);
+  check_failure(EXESS_UBYTE, "1", EXESS_TIME, EXESS_UNSUPPORTED);
+  check_failure(EXESS_UBYTE, "1", EXESS_DATE, EXESS_UNSUPPORTED);
+}
+
+static void
+test_time_to_number(void)
+{
+  static const char* const duration_str = "P1Y";
+  static const char* const datetime_str = "2001-02-03T04:05:06";
+  static const char* const time_str     = "04:05:06";
+  static const char* const date_str     = "2001-02-03";
+
+  check_failure(EXESS_DURATION, duration_str, EXESS_BOOLEAN, EXESS_UNSUPPORTED);
+  check_failure(EXESS_DATETIME, datetime_str, EXESS_BOOLEAN, EXESS_UNSUPPORTED);
+  check_failure(EXESS_TIME, time_str, EXESS_BOOLEAN, EXESS_UNSUPPORTED);
+  check_failure(EXESS_DATE, date_str, EXESS_BOOLEAN, EXESS_UNSUPPORTED);
+
+  check_failure(EXESS_DURATION, duration_str, EXESS_INT, EXESS_UNSUPPORTED);
+  check_failure(EXESS_DATETIME, datetime_str, EXESS_INT, EXESS_UNSUPPORTED);
+  check_failure(EXESS_TIME, time_str, EXESS_INT, EXESS_UNSUPPORTED);
+  check_failure(EXESS_DATE, date_str, EXESS_INT, EXESS_UNSUPPORTED);
+
+  check_failure(EXESS_DURATION, duration_str, EXESS_SHORT, EXESS_UNSUPPORTED);
+  check_failure(EXESS_DATETIME, datetime_str, EXESS_SHORT, EXESS_UNSUPPORTED);
+  check_failure(EXESS_TIME, time_str, EXESS_SHORT, EXESS_UNSUPPORTED);
+  check_failure(EXESS_DATE, date_str, EXESS_SHORT, EXESS_UNSUPPORTED);
+
+  check_failure(EXESS_DURATION, duration_str, EXESS_BYTE, EXESS_UNSUPPORTED);
+  check_failure(EXESS_DATETIME, datetime_str, EXESS_BYTE, EXESS_UNSUPPORTED);
+  check_failure(EXESS_TIME, time_str, EXESS_BYTE, EXESS_UNSUPPORTED);
+  check_failure(EXESS_DATE, date_str, EXESS_BYTE, EXESS_UNSUPPORTED);
+
+  check_failure(EXESS_DURATION, duration_str, EXESS_ULONG, EXESS_UNSUPPORTED);
+  check_failure(EXESS_DATETIME, datetime_str, EXESS_ULONG, EXESS_UNSUPPORTED);
+  check_failure(EXESS_TIME, time_str, EXESS_ULONG, EXESS_UNSUPPORTED);
+  check_failure(EXESS_DATE, date_str, EXESS_ULONG, EXESS_UNSUPPORTED);
+
+  check_failure(EXESS_DURATION, duration_str, EXESS_UINT, EXESS_UNSUPPORTED);
+  check_failure(EXESS_DATETIME, datetime_str, EXESS_UINT, EXESS_UNSUPPORTED);
+  check_failure(EXESS_TIME, time_str, EXESS_UINT, EXESS_UNSUPPORTED);
+  check_failure(EXESS_DATE, date_str, EXESS_UINT, EXESS_UNSUPPORTED);
+
+  check_failure(EXESS_DURATION, duration_str, EXESS_USHORT, EXESS_UNSUPPORTED);
+  check_failure(EXESS_DATETIME, datetime_str, EXESS_USHORT, EXESS_UNSUPPORTED);
+  check_failure(EXESS_TIME, time_str, EXESS_USHORT, EXESS_UNSUPPORTED);
+  check_failure(EXESS_DATE, date_str, EXESS_USHORT, EXESS_UNSUPPORTED);
+
+  check_failure(EXESS_DURATION, duration_str, EXESS_UBYTE, EXESS_UNSUPPORTED);
+  check_failure(EXESS_DATETIME, datetime_str, EXESS_UBYTE, EXESS_UNSUPPORTED);
+  check_failure(EXESS_TIME, time_str, EXESS_UBYTE, EXESS_UNSUPPORTED);
+  check_failure(EXESS_DATE, date_str, EXESS_UBYTE, EXESS_UNSUPPORTED);
+}
+
+static void
+test_binary(void)
+{
+  check_from_to(EXESS_HEX, "666F6F", EXESS_BASE64, "Zm9v");
+
+  check_failure(EXESS_LONG, "-2147483649", EXESS_HEX, EXESS_UNSUPPORTED);
+  check_failure(EXESS_LONG, "-2147483649", EXESS_BASE64, EXESS_UNSUPPORTED);
+
+  /* check_from_to(EXESS_BASE64, "Zm9v", EXESS_HEX, "666F6F"); */
+
+  /*   ////////// */
+
+  /* check_one_way(EXESS_LONG, "-1", EXESS_TRUNCATE, EXESS_BOOLEAN, "true"); */
+
+  /* // All smaller integer types */
+  /* check_from_to(EXESS_LONG, "-2147483648", EXESS_INT, "-2147483648"); */
+  /* check_failure(EXESS_LONG, "-2147483649", EXESS_INT, EXESS_OUT_OF_RANGE); */
+}
+
+int
+main(void)
+{
+  test_unknown();
+  test_boolean();
+  test_long();
+  test_ulong();
+  test_large_integers();
+  test_coerce();
+  test_date_time();
+  test_number_to_time();
+  test_time_to_number();
+  test_binary();
+
+  return 0;
+}
diff --git a/subprojects/exess/test/test_datatype.c b/subprojects/exess/test/test_datatype.c
new file mode 100644
index 00000000..7ad3fb5e
--- /dev/null
+++ b/subprojects/exess/test/test_datatype.c
@@ -0,0 +1,81 @@
+/*
+  Copyright 2011-2021 David Robillard <d@drobilla.net>
+
+  Permission to use, copy, modify, and/or distribute this software for any
+  purpose with or without fee is hereby granted, provided that the above
+  copyright notice and this permission notice appear in all copies.
+
+  THIS SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+  WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+  MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+  ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+  WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+  ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+  OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+*/
+
+#undef NDEBUG
+
+#include "exess/exess.h"
+
+#include <assert.h>
+
+static void
+test_datatype_uris(void)
+{
+  assert(!exess_datatype_uri(EXESS_NOTHING));
+  assert(!exess_datatype_uri((ExessDatatype)(EXESS_BASE64 + 1)));
+
+  for (unsigned i = 1; i <= EXESS_BASE64; ++i) {
+    const char* const uri = exess_datatype_uri((ExessDatatype)i);
+
+    assert(uri);
+    assert(exess_datatype_from_uri(uri) == i);
+  }
+
+  assert(exess_datatype_from_uri(EXESS_XSD_URI) == EXESS_NOTHING);
+  assert(exess_datatype_from_uri(EXESS_XSD_URI "unknown") == EXESS_NOTHING);
+  assert(exess_datatype_from_uri("garbage") == EXESS_NOTHING);
+  assert(exess_datatype_from_uri("http://example.org/very/long/unknown/uri") ==
+         EXESS_NOTHING);
+}
+
+static void
+test_datatype_is_bounded(void)
+{
+  assert(!exess_datatype_is_bounded(EXESS_NOTHING));
+  assert(exess_datatype_is_bounded(EXESS_BOOLEAN));
+  assert(!exess_datatype_is_bounded(EXESS_DECIMAL));
+  assert(exess_datatype_is_bounded(EXESS_DOUBLE));
+  assert(exess_datatype_is_bounded(EXESS_FLOAT));
+  assert(!exess_datatype_is_bounded(EXESS_INTEGER));
+  assert(!exess_datatype_is_bounded(EXESS_NON_POSITIVE_INTEGER));
+  assert(!exess_datatype_is_bounded(EXESS_NEGATIVE_INTEGER));
+  assert(exess_datatype_is_bounded(EXESS_LONG));
+  assert(exess_datatype_is_bounded(EXESS_INT));
+  assert(exess_datatype_is_bounded(EXESS_SHORT));
+  assert(exess_datatype_is_bounded(EXESS_BYTE));
+  assert(!exess_datatype_is_bounded(EXESS_NON_NEGATIVE_INTEGER));
+  assert(exess_datatype_is_bounded(EXESS_ULONG));
+  assert(exess_datatype_is_bounded(EXESS_UINT));
+  assert(exess_datatype_is_bounded(EXESS_USHORT));
+  assert(exess_datatype_is_bounded(EXESS_UBYTE));
+  assert(!exess_datatype_is_bounded(EXESS_POSITIVE_INTEGER));
+  assert(exess_datatype_is_bounded(EXESS_DURATION));
+  assert(exess_datatype_is_bounded(EXESS_DATETIME));
+  assert(exess_datatype_is_bounded(EXESS_TIME));
+  assert(exess_datatype_is_bounded(EXESS_DATE));
+  assert(!exess_datatype_is_bounded(EXESS_HEX));
+  assert(!exess_datatype_is_bounded(EXESS_BASE64));
+}
+
+int
+main(int argc, char** argv)
+{
+  (void)argv;
+
+  test_datatype_uris();
+  test_datatype_is_bounded();
+
+  return argc == 1 ? 0 : 1;
+}
diff --git a/subprojects/exess/test/test_date.c b/subprojects/exess/test/test_date.c
new file mode 100644
index 00000000..0ac8129a
--- /dev/null
+++ b/subprojects/exess/test/test_date.c
@@ -0,0 +1,257 @@
+/*
+  Copyright 2011-2021 David Robillard <d@drobilla.net>
+
+  Permission to use, copy, modify, and/or distribute this software for any
+  purpose with or without fee is hereby granted, provided that the above
+  copyright notice and this permission notice appear in all copies.
+
+  THIS SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+  WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+  MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+  ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+  WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+  ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+  OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+*/
+
+#undef NDEBUG
+
+#include "date_utils.h"
+#include "int_test_data.h"
+#include "macros.h"
+#include "num_test_utils.h"
+#include "time_test_utils.h"
+
+#include "exess/exess.h"
+
+#include <assert.h>
+#include <stdbool.h>
+#include <stdint.h>
+#include <stdio.h>
+#include <string.h>
+
+static const ExessDate nozone   = {2001, 1, 2, {EXESS_LOCAL}};
+static const ExessDate utc      = {2002, 2, 3, {0}};
+static const ExessDate zoned    = {2003, 3, 4, INIT_ZONE(11, 30)};
+static const ExessDate early    = {99, 3, 4, INIT_ZONE(11, 30)};
+static const ExessDate future   = {12345, 3, 4, INIT_ZONE(11, 30)};
+static const ExessDate lowest   = {INT16_MIN, 1, 1, INIT_ZONE(-14, 0)};
+static const ExessDate highest  = {INT16_MAX, 1, 1, INIT_ZONE(14, 0)};
+static const ExessDate garbage1 = {2004, 0, 1, INIT_ZONE(11, 30)};
+static const ExessDate garbage2 = {2005, 13, 1, INIT_ZONE(11, 30)};
+static const ExessDate garbage3 = {2006, 1, 0, INIT_ZONE(11, 30)};
+static const ExessDate garbage4 = {2006, 1, 32, INIT_ZONE(11, 30)};
+
+static void
+check_read(const char* const string,
+           const ExessStatus expected_status,
+           const int64_t     expected_year,
+           const uint8_t     expected_month,
+           const uint8_t     expected_day,
+           const int8_t      expected_tz_hour,
+           const int8_t      expected_tz_minute,
+           const bool        expected_tz_is_present,
+           const size_t      expected_count)
+{
+  ExessDate value = {0, 0, 0, {EXESS_LOCAL}};
+
+  const ExessResult r = exess_read_date(&value, string);
+  assert(r.status == expected_status);
+  assert(r.count == expected_count);
+  assert(value.year == expected_year);
+  assert(value.month == expected_month);
+  assert(value.day == expected_day);
+  assert((!expected_tz_is_present && value.zone.quarter_hours == EXESS_LOCAL) ||
+         value.zone.quarter_hours ==
+           4 * expected_tz_hour + expected_tz_minute / 15);
+}
+
+static void
+test_read_date(void)
+{
+  // No input
+  check_read("", EXESS_EXPECTED_DIGIT, 0, 0, 0, 0, 0, false, 0);
+  check_read(" \f\n\r\t\v", EXESS_EXPECTED_DIGIT, 0, 0, 0, 0, 0, false, 6);
+
+  // Good values
+  check_read("2004-04-12", EXESS_SUCCESS, 2004, 4, 12, 0, 0, false, 10);
+  check_read("-0045-01-01", EXESS_SUCCESS, -45, 1, 1, 0, 0, false, 11);
+  check_read("12004-04-12", EXESS_SUCCESS, 12004, 4, 12, 0, 0, false, 11);
+  check_read("2004-04-12-05:00", EXESS_SUCCESS, 2004, 4, 12, -5, 0, true, 16);
+  check_read("2004-04-12Z", EXESS_SUCCESS, 2004, 4, 12, 0, 0, true, 11);
+  check_read("2001-10-26", EXESS_SUCCESS, 2001, 10, 26, 0, 0, false, 10);
+  check_read("2001-10-26+02:00", EXESS_SUCCESS, 2001, 10, 26, 2, 0, true, 16);
+  check_read("2001-10-26Z", EXESS_SUCCESS, 2001, 10, 26, 0, 0, true, 11);
+  check_read("2001-10-26+00:00", EXESS_SUCCESS, 2001, 10, 26, 0, 0, true, 16);
+  check_read("-2001-10-26", EXESS_SUCCESS, -2001, 10, 26, 0, 0, false, 11);
+  check_read("-20000-04-01", EXESS_SUCCESS, -20000, 04, 01, 0, 0, false, 12);
+
+  // Non-canonical
+  check_read("02004-04-12", EXESS_SUCCESS, 2004, 4, 12, 0, 0, false, 11);
+  check_read(" 02004-04-12 ", EXESS_SUCCESS, 2004, 4, 12, 0, 0, false, 12);
+
+  // Good common year values
+  check_read("1900-02-28", EXESS_SUCCESS, 1900, 2, 28, 0, 0, false, 10);
+
+  // Good leap year values
+  check_read("2000-02-29", EXESS_SUCCESS, 2000, 2, 29, 0, 0, false, 10);
+  check_read("2004-02-29", EXESS_SUCCESS, 2004, 2, 29, 0, 0, false, 10);
+
+  // Longest possible string
+  check_read("-32768-01-01-14:00",
+             EXESS_SUCCESS,
+             -32768,
+             1,
+             1,
+             -14,
+             0,
+             true,
+             EXESS_MAX_DATE_LENGTH);
+
+  // Limits
+  check_read("-32768-01-01", EXESS_SUCCESS, -32768, 1, 1, 0, 0, false, 12);
+  check_read("32767-01-01", EXESS_SUCCESS, 32767, 1, 1, 0, 0, false, 11);
+
+  // Out of range years
+  check_read("-32769-01-01", EXESS_OUT_OF_RANGE, 0, 0, 0, 0, 0, false, 6);
+  check_read("32768-01-01", EXESS_OUT_OF_RANGE, 0, 0, 0, 0, 0, false, 5);
+
+  // Out of range months
+  check_read("2001-00-26", EXESS_OUT_OF_RANGE, 2001, 0, 0, 0, 0, false, 7);
+  check_read("2001-13-26", EXESS_OUT_OF_RANGE, 2001, 13, 0, 0, 0, false, 7);
+  check_read("2001-10-00", EXESS_OUT_OF_RANGE, 2001, 10, 0, 0, 0, false, 10);
+  check_read("2001-10-32", EXESS_OUT_OF_RANGE, 2001, 10, 32, 0, 0, false, 10);
+
+  // Out of range days
+  check_read("2001-01-32", EXESS_OUT_OF_RANGE, 2001, 1, 32, 0, 0, false, 10);
+  check_read("2001-02-29", EXESS_OUT_OF_RANGE, 2001, 2, 29, 0, 0, false, 10);
+  check_read("2001-03-32", EXESS_OUT_OF_RANGE, 2001, 3, 32, 0, 0, false, 10);
+  check_read("2001-04-31", EXESS_OUT_OF_RANGE, 2001, 4, 31, 0, 0, false, 10);
+  check_read("2001-05-32", EXESS_OUT_OF_RANGE, 2001, 5, 32, 0, 0, false, 10);
+  check_read("2001-06-31", EXESS_OUT_OF_RANGE, 2001, 6, 31, 0, 0, false, 10);
+  check_read("2001-07-32", EXESS_OUT_OF_RANGE, 2001, 7, 32, 0, 0, false, 10);
+  check_read("2001-08-32", EXESS_OUT_OF_RANGE, 2001, 8, 32, 0, 0, false, 10);
+  check_read("2001-09-31", EXESS_OUT_OF_RANGE, 2001, 9, 31, 0, 0, false, 10);
+  check_read("2001-10-32", EXESS_OUT_OF_RANGE, 2001, 10, 32, 0, 0, false, 10);
+  check_read("2001-11-31", EXESS_OUT_OF_RANGE, 2001, 11, 31, 0, 0, false, 10);
+  check_read("2001-12-32", EXESS_OUT_OF_RANGE, 2001, 12, 32, 0, 0, false, 10);
+
+  // Garbage
+  check_read("f", EXESS_EXPECTED_DIGIT, 0, 0, 0, 0, 0, false, 0);
+  check_read("99-04-12", EXESS_EXPECTED_DIGIT, 99, 0, 0, 0, 0, false, 2);
+  check_read("2004-4-2", EXESS_EXPECTED_DIGIT, 2004, 4, 0, 0, 0, false, 6);
+  check_read("2004/04/02", EXESS_EXPECTED_DASH, 2004, 0, 0, 0, 0, false, 4);
+  check_read("04-12-2004", EXESS_EXPECTED_DIGIT, 4, 0, 0, 0, 0, false, 2);
+  check_read("2001-10", EXESS_EXPECTED_DASH, 2001, 10, 0, 0, 0, false, 7);
+  check_read("01-10-26", EXESS_EXPECTED_DIGIT, 1, 0, 0, 0, 0, false, 2);
+  check_read("2004-04-12A", EXESS_EXPECTED_SIGN, 2004, 4, 12, 0, 0, false, 10);
+}
+
+static void
+check_write(const ExessDate   value,
+            const ExessStatus expected_status,
+            const size_t      buf_size,
+            const char* const expected_string)
+{
+  char buf[EXESS_MAX_DATE_LENGTH + 1] = {
+    1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19};
+
+  assert(buf_size <= sizeof(buf));
+
+  const ExessResult r = exess_write_date(value, buf_size, buf);
+  assert(r.status == expected_status);
+  assert(r.count == strlen(buf));
+  assert(!strcmp(buf, expected_string));
+  assert(r.status || exess_write_date(value, 0, NULL).count == r.count);
+}
+
+static void
+test_write_date(void)
+{
+  check_write(nozone, EXESS_SUCCESS, 11, "2001-01-02");
+  check_write(utc, EXESS_SUCCESS, 12, "2002-02-03Z");
+  check_write(zoned, EXESS_SUCCESS, 17, "2003-03-04+11:30");
+  check_write(early, EXESS_SUCCESS, 17, "0099-03-04+11:30");
+  check_write(future, EXESS_SUCCESS, 18, "12345-03-04+11:30");
+  check_write(lowest, EXESS_SUCCESS, 19, "-32768-01-01-14:00");
+  check_write(highest, EXESS_SUCCESS, 18, "32767-01-01+14:00");
+
+  check_write(garbage1, EXESS_BAD_VALUE, 14, "");
+  check_write(garbage2, EXESS_BAD_VALUE, 14, "");
+  check_write(garbage3, EXESS_BAD_VALUE, 14, "");
+  check_write(garbage4, EXESS_BAD_VALUE, 14, "");
+
+  check_write(nozone, EXESS_NO_SPACE, 10, "");
+  check_write(future, EXESS_NO_SPACE, 5, "");
+  check_write(lowest, EXESS_NO_SPACE, 5, "");
+  check_write(highest, EXESS_NO_SPACE, 5, "");
+  check_write(utc, EXESS_NO_SPACE, 11, "");
+  check_write(zoned, EXESS_NO_SPACE, 16, "");
+
+  // Check that nothing is written when there isn't enough space
+  char              c = 42;
+  const ExessResult r = exess_write_date(nozone, 0, &c);
+  assert(c == 42);
+  assert(r.status == EXESS_NO_SPACE);
+  assert(r.count == 0);
+}
+
+static void
+check_round_trip(const ExessDate value)
+{
+  ExessDate parsed_value                   = {0, 0, 0, {0}};
+  char      buf[EXESS_MAX_DATE_LENGTH + 1] = {0};
+
+  assert(!exess_write_date(value, sizeof(buf), buf).status);
+  assert(!exess_read_date(&parsed_value, buf).status);
+  assert(parsed_value.year == value.year);
+  assert(parsed_value.month == value.month);
+  assert(parsed_value.day == value.day);
+  assert(parsed_value.zone.quarter_hours == value.zone.quarter_hours);
+}
+
+static void
+test_round_trip(const ExessNumTestOptions opts)
+{
+  fprintf(stderr, "Testing xsd:gDate randomly with seed %u\n", opts.seed);
+
+  const uint64_t n_tests = MAX(256, opts.n_tests / 16);
+
+  uint32_t rng = opts.seed;
+  for (uint64_t i = 0; i < n_tests; ++i) {
+    rng = lcg32(rng);
+
+    const int16_t year = (int16_t)(rng % UINT16_MAX);
+    for (uint8_t month = 1; month < 13; ++month) {
+      for (uint8_t day = 1; day <= days_in_month(year, month); ++day) {
+        const ExessDate no_zone      = {year, month, day, {EXESS_LOCAL}};
+        const ExessDate lowest_zone  = {year, month, day, {4 * -14 + 0}};
+        const ExessDate highest_zone = {year, month, day, {4 * 14}};
+
+        check_round_trip(no_zone);
+        check_round_trip(lowest_zone);
+        check_round_trip(highest_zone);
+
+        const ExessDate value = {year, month, day, random_timezone(&rng)};
+        check_round_trip(value);
+      }
+    }
+
+    print_num_test_progress(i, n_tests);
+  }
+}
+
+int
+main(int argc, char** argv)
+{
+  const ExessNumTestOptions opts = parse_num_test_options(argc, argv);
+  if (opts.error) {
+    return 1;
+  }
+
+  test_read_date();
+  test_write_date();
+  test_round_trip(opts);
+
+  return 0;
+}
diff --git a/subprojects/exess/test/test_datetime.c b/subprojects/exess/test/test_datetime.c
new file mode 100644
index 00000000..d6d2b378
--- /dev/null
+++ b/subprojects/exess/test/test_datetime.c
@@ -0,0 +1,459 @@
+/*
+  Copyright 2011-2021 David Robillard <d@drobilla.net>
+
+  Permission to use, copy, modify, and/or distribute this software for any
+  purpose with or without fee is hereby granted, provided that the above
+  copyright notice and this permission notice appear in all copies.
+
+  THIS SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+  WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+  MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+  ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+  WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+  ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+  OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+*/
+
+#undef NDEBUG
+
+#include "exess/exess.h"
+
+#include <assert.h>
+#include <stdbool.h>
+#include <stdint.h>
+#include <string.h>
+
+static const ExessDateTime local     = {2001, 2, 3, false, 4, 5, 6, 0};
+static const ExessDateTime utc       = {2001, 2, 3, true, 4, 5, 6, 0};
+static const ExessDateTime lowest    = {INT16_MIN, 1, 1, false, 0, 0, 0, 0};
+static const ExessDateTime highest   = {INT16_MAX, 12, 31, false, 24, 0, 0, 0};
+static const ExessDateTime utc_min   = {INT16_MIN, 1, 1, true, 0, 0, 0, 0};
+static const ExessDateTime utc_max   = {INT16_MAX, 12, 31, true, 24, 0, 0, 0};
+static const ExessDateTime nano      = {2001, 1, 1, false, 0, 0, 0, 1};
+static const ExessDateTime garbage1  = {2004, 0, 1, false, 12, 0, 0, 0};
+static const ExessDateTime garbage2  = {2005, 13, 1, false, 12, 0, 0, 0};
+static const ExessDateTime garbage3  = {2006, 1, 0, false, 12, 0, 0, 0};
+static const ExessDateTime garbage4  = {2006, 1, 32, false, 12, 0, 0, 0};
+static const ExessDateTime garbage5  = {2001, 2, 3, false, 0, 0, 0, 1000000000};
+static const ExessDateTime garbage6  = {2001, 2, 3, false, 0, 0, 60, 0};
+static const ExessDateTime garbage7  = {2001, 2, 3, false, 0, 60, 0, 0};
+static const ExessDateTime garbage8  = {2001, 2, 3, false, 24, 0, 0, 1};
+static const ExessDateTime garbage9  = {2001, 2, 3, false, 24, 0, 1, 0};
+static const ExessDateTime garbage10 = {2001, 2, 3, false, 24, 1, 0, 0};
+static const ExessDateTime garbage11 = {2001, 2, 3, false, 25, 0, 0, 0};
+
+static void
+check_add(const char* const datetime_string,
+          const char* const duration_string,
+          const char* const result_string)
+{
+  ExessDateTime datetime = {0, 0u, 0u, false, 0u, 0u, 0u, 0u};
+  ExessDuration duration = {0u, 0u, 0u};
+
+  ExessResult r = exess_read_datetime(&datetime, datetime_string);
+  assert(!r.status);
+
+  r = exess_read_duration(&duration, duration_string);
+  assert(!r.status);
+
+  const ExessDateTime result  = exess_add_datetime_duration(datetime, duration);
+  char                buf[28] = {0};
+
+  r = exess_write_datetime(result, sizeof(buf), buf);
+  assert(!r.status);
+  assert(!strcmp(buf, result_string));
+}
+
+static void
+check_is_underflow(const ExessDateTime datetime, const bool is_utc)
+{
+  assert(datetime.year == INT16_MIN);
+  assert(datetime.month == 0);
+  assert(datetime.day == 0);
+  assert(datetime.is_utc == is_utc);
+  assert(datetime.hour == 0);
+  assert(datetime.minute == 0);
+  assert(datetime.second == 0);
+  assert(datetime.nanosecond == 0);
+}
+
+static void
+check_is_overflow(const ExessDateTime datetime, const bool is_utc)
+{
+  assert(datetime.year == INT16_MAX);
+  assert(datetime.month == UINT8_MAX);
+  assert(datetime.day == UINT8_MAX);
+  assert(datetime.is_utc == is_utc);
+  assert(datetime.hour == UINT8_MAX);
+  assert(datetime.minute == UINT8_MAX);
+  assert(datetime.second == UINT8_MAX);
+  assert(datetime.nanosecond == UINT32_MAX);
+}
+
+static void
+test_add(void)
+{
+  // Simple cases
+  check_add("2001-01-01T00:00:00", "PT1.5S", "2001-01-01T00:00:01.5");
+  check_add("2001-01-01T00:00:00", "PT1M", "2001-01-01T00:01:00");
+  check_add("2001-01-01T00:00:00", "PT1H", "2001-01-01T01:00:00");
+  check_add("2001-01-01T00:00:00", "P1D", "2001-01-02T00:00:00");
+  check_add("2001-01-01T00:00:00", "P1M", "2001-02-01T00:00:00");
+  check_add("2001-01-01T00:00:00", "P1Y", "2002-01-01T00:00:00");
+  check_add("2001-02-02T02:02:02", "-PT1.5S", "2001-02-02T02:02:00.5");
+  check_add("2001-02-02T02:02:02", "-PT1M", "2001-02-02T02:01:02");
+  check_add("2001-02-02T02:02:02", "-PT1H", "2001-02-02T01:02:02");
+  check_add("2001-02-02T02:02:02", "-P1D", "2001-02-01T02:02:02");
+  check_add("2001-02-02T02:02:02", "-P1M", "2001-01-02T02:02:02");
+  check_add("2001-02-02T02:02:02", "-P1Y", "2000-02-02T02:02:02");
+
+  // Positive carrying
+  check_add("2001-01-01T00:00:59", "PT1S", "2001-01-01T00:01:00");
+  check_add("2001-01-01T00:59:00", "PT1M", "2001-01-01T01:00:00");
+  check_add("2001-01-01T23:00:00", "PT1H", "2001-01-02T00:00:00");
+  check_add("2001-01-31T00:00:00", "P1D", "2001-02-01T00:00:00");
+  check_add("2001-12-01T00:00:00", "P1M", "2002-01-01T00:00:00");
+
+  // Negative carrying
+  check_add("2001-01-01T00:01:00", "-PT1S", "2001-01-01T00:00:59");
+  check_add("2001-02-01T01:00:00", "-PT1M", "2001-02-01T00:59:00");
+  check_add("2001-02-02T00:00:00", "-PT1H", "2001-02-01T23:00:00");
+  check_add("2001-02-01T00:00:00", "-P1D", "2001-01-31T00:00:00");
+  check_add("2001-01-01T00:00:00", "-P1M", "2000-12-01T00:00:00");
+
+  // Underflow and overflow
+
+  static const ExessDuration minus_month      = {-1, 0, 0};
+  static const ExessDuration minus_second     = {0, -1, 0};
+  static const ExessDuration minus_nanosecond = {0, 0, -1};
+  static const ExessDuration plus_month       = {1, 0, 0};
+  static const ExessDuration plus_second      = {0, 1, 0};
+  static const ExessDuration plus_nanosecond  = {0, 0, 1};
+
+  check_is_underflow(exess_add_datetime_duration(lowest, minus_month), false);
+  check_is_underflow(exess_add_datetime_duration(lowest, minus_second), false);
+  check_is_underflow(exess_add_datetime_duration(lowest, minus_nanosecond),
+                     false);
+
+  check_is_underflow(exess_add_datetime_duration(utc_min, minus_month), true);
+  check_is_underflow(exess_add_datetime_duration(utc_min, minus_second), true);
+  check_is_underflow(exess_add_datetime_duration(utc_min, minus_nanosecond),
+                     true);
+
+  check_is_overflow(exess_add_datetime_duration(highest, plus_month), false);
+  check_is_overflow(exess_add_datetime_duration(highest, plus_second), false);
+  check_is_overflow(exess_add_datetime_duration(highest, plus_nanosecond),
+                    false);
+
+  check_is_overflow(exess_add_datetime_duration(utc_max, plus_month), true);
+  check_is_overflow(exess_add_datetime_duration(utc_max, plus_second), true);
+  check_is_overflow(exess_add_datetime_duration(utc_max, plus_nanosecond),
+                    true);
+}
+
+static void
+check_read(const char* const string,
+           const ExessStatus expected_status,
+           const size_t      expected_count,
+           const int64_t     expected_year,
+           const uint8_t     expected_month,
+           const uint8_t     expected_day,
+           const uint8_t     expected_hour,
+           const uint8_t     expected_minute,
+           const uint8_t     expected_second,
+           const uint32_t    expected_nanosecond,
+           const bool        expected_is_utc)
+{
+  ExessDateTime value = {0, 0, 0, false, 0, 0, 0, 0};
+
+  const ExessResult r = exess_read_datetime(&value, string);
+  assert(r.status == expected_status);
+  assert(r.count == expected_count);
+  assert(value.year == expected_year);
+  assert(value.month == expected_month);
+  assert(value.day == expected_day);
+  assert(value.hour == expected_hour);
+  assert(value.minute == expected_minute);
+  assert(value.second == expected_second);
+  assert(value.nanosecond == expected_nanosecond);
+  assert(value.is_utc == expected_is_utc);
+}
+
+static void
+test_read_datetime(void)
+{
+  // Simple values
+
+  check_read(
+    "2001-02-03T04:05:06", EXESS_SUCCESS, 19, 2001, 2, 3, 4, 5, 6, 0, false);
+
+  check_read(
+    "2001-02-03T04:05:06Z", EXESS_SUCCESS, 20, 2001, 2, 3, 4, 5, 6, 0, true);
+
+  check_read("2004-04-12T13:20:15.5",
+             EXESS_SUCCESS,
+             21,
+             2004,
+             4,
+             12,
+             13,
+             20,
+             15,
+             500000000,
+             false);
+
+  check_read("-32768-01-01T00:00:00.000000001Z",
+             EXESS_SUCCESS,
+             EXESS_MAX_DATETIME_LENGTH,
+             -32768,
+             1,
+             1,
+             0,
+             0,
+             0,
+             1,
+             true);
+
+  // Simple timezones
+
+  check_read("2001-02-03T04:05:06-00:30",
+             EXESS_SUCCESS,
+             25,
+             2001,
+             2,
+             3,
+             4,
+             35,
+             6,
+             0,
+             true);
+
+  check_read("2001-02-03T04:05:06-01:00",
+             EXESS_SUCCESS,
+             25,
+             2001,
+             2,
+             3,
+             5,
+             5,
+             6,
+             0,
+             true);
+
+  check_read("2001-02-03T04:05:06+00:30",
+             EXESS_SUCCESS,
+             25,
+             2001,
+             2,
+             3,
+             3,
+             35,
+             6,
+             0,
+             true);
+
+  check_read("2001-02-03T04:05:06+01:00",
+             EXESS_SUCCESS,
+             25,
+             2001,
+             2,
+             3,
+             3,
+             5,
+             6,
+             0,
+             true);
+
+  // Positive timezone carry
+
+  // Minute => hour
+  check_read("2001-02-03T04:46:00-00:15",
+             EXESS_SUCCESS,
+             25,
+             2001,
+             2,
+             3,
+             5,
+             1,
+             0,
+             0,
+             true);
+
+  // Minute => hour => day
+  check_read("2001-02-03T23:46:00-00:15",
+             EXESS_SUCCESS,
+             25,
+             2001,
+             2,
+             4,
+             0,
+             1,
+             0,
+             0,
+             true);
+
+  // Minute => hour => day => month
+  check_read("2001-02-28T23:46:00-00:15",
+             EXESS_SUCCESS,
+             25,
+             2001,
+             3,
+             1,
+             0,
+             1,
+             0,
+             0,
+             true);
+
+  // Minute => hour => day => month => year
+  check_read("2001-12-31T23:46:00-00:15",
+             EXESS_SUCCESS,
+             25,
+             2002,
+             1,
+             1,
+             0,
+             1,
+             0,
+             0,
+             true);
+
+  // Negative timezone carry
+
+  // Minute => hour
+  check_read("2001-02-03T04:14:00+00:15",
+             EXESS_SUCCESS,
+             25,
+             2001,
+             2,
+             3,
+             3,
+             59,
+             0,
+             0,
+             true);
+
+  // Minute => hour => day
+  check_read("2001-02-03T00:14:00+00:15",
+             EXESS_SUCCESS,
+             25,
+             2001,
+             2,
+             2,
+             23,
+             59,
+             0,
+             0,
+             true);
+
+  // Minute => hour => day => month
+  check_read("2001-02-01T00:14:00+00:15",
+             EXESS_SUCCESS,
+             25,
+             2001,
+             1,
+             31,
+             23,
+             59,
+             0,
+             0,
+             true);
+
+  // Garbage
+
+  check_read(
+    "2004-04-12T13:00", EXESS_EXPECTED_COLON, 16, 0, 0, 0, 0, 0, 0, 0, false);
+
+  check_read("2004-04-1213:20:00",
+             EXESS_EXPECTED_TIME_SEP,
+             10,
+             0,
+             0,
+             0,
+             0,
+             0,
+             0,
+             0,
+             false);
+
+  check_read(
+    "99-04-12T13:00", EXESS_EXPECTED_DIGIT, 2, 0, 0, 0, 0, 0, 0, 0, false);
+
+  check_read(
+    "2004-04-12", EXESS_EXPECTED_TIME_SEP, 10, 0, 0, 0, 0, 0, 0, 0, false);
+
+  check_read("2004-04-12-05:00",
+             EXESS_EXPECTED_TIME_SEP,
+             10,
+             0,
+             0,
+             0,
+             0,
+             0,
+             0,
+             0,
+             false);
+}
+
+static void
+check_write(const ExessDateTime value,
+            const ExessStatus   expected_status,
+            const size_t        buf_size,
+            const char* const   expected_string)
+{
+  char buf[EXESS_MAX_DATETIME_LENGTH + 1] = {
+    1,  2,  3,  4,  5,  6,  7,  8,  9,  10, 11, 12, 13, 14, 15, 16,
+    17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32};
+
+  assert(buf_size <= sizeof(buf));
+
+  const ExessResult r = exess_write_datetime(value, buf_size, buf);
+  assert(r.status == expected_status);
+  assert(r.count == strlen(buf));
+  assert(!strcmp(buf, expected_string));
+  assert(r.status || exess_write_datetime(value, 0, NULL).count == r.count);
+}
+
+static void
+test_write_datetime(void)
+{
+  check_write(local, EXESS_SUCCESS, 20, "2001-02-03T04:05:06");
+  check_write(utc, EXESS_SUCCESS, 21, "2001-02-03T04:05:06Z");
+  check_write(lowest, EXESS_SUCCESS, 22, "-32768-01-01T00:00:00");
+  check_write(highest, EXESS_SUCCESS, 21, "32767-12-31T24:00:00");
+  check_write(nano, EXESS_SUCCESS, 30, "2001-01-01T00:00:00.000000001");
+
+  check_write(garbage1, EXESS_BAD_VALUE, 20, "");
+  check_write(garbage2, EXESS_BAD_VALUE, 20, "");
+  check_write(garbage3, EXESS_BAD_VALUE, 20, "");
+  check_write(garbage4, EXESS_BAD_VALUE, 20, "");
+  check_write(garbage5, EXESS_BAD_VALUE, 20, "");
+  check_write(garbage6, EXESS_BAD_VALUE, 20, "");
+  check_write(garbage7, EXESS_BAD_VALUE, 20, "");
+  check_write(garbage8, EXESS_BAD_VALUE, 20, "");
+  check_write(garbage9, EXESS_BAD_VALUE, 20, "");
+  check_write(garbage10, EXESS_BAD_VALUE, 20, "");
+  check_write(garbage11, EXESS_BAD_VALUE, 20, "");
+
+  check_write(lowest, EXESS_NO_SPACE, 12, "");
+  check_write(lowest, EXESS_NO_SPACE, 17, "");
+  check_write(lowest, EXESS_NO_SPACE, 18, "");
+  check_write(lowest, EXESS_NO_SPACE, 21, "");
+
+  // Check that nothing is written when there isn't enough space
+  char              c = 42;
+  const ExessResult r = exess_write_datetime(highest, 0, &c);
+  assert(c == 42);
+  assert(r.status == EXESS_NO_SPACE);
+  assert(r.count == 0);
+}
+
+int
+main(void)
+{
+  test_add();
+  test_read_datetime();
+  test_write_datetime();
+
+  return 0;
+}
diff --git a/subprojects/exess/test/test_decimal.c b/subprojects/exess/test/test_decimal.c
new file mode 100644
index 00000000..2a49565d
--- /dev/null
+++ b/subprojects/exess/test/test_decimal.c
@@ -0,0 +1,260 @@
+/*
+  Copyright 2011-2021 David Robillard <d@drobilla.net>
+
+  Permission to use, copy, modify, and/or distribute this software for any
+  purpose with or without fee is hereby granted, provided that the above
+  copyright notice and this permission notice appear in all copies.
+
+  THIS SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+  WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+  MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+  ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+  WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+  ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+  OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+*/
+
+#undef NDEBUG
+
+#include "float_test_data.h"
+#include "int_test_data.h"
+#include "num_test_utils.h"
+#include "string_utils.h"
+
+#include "exess/exess.h"
+
+#include <assert.h>
+#include <float.h>
+#include <math.h>
+#include <stdint.h>
+#include <stdio.h>
+#include <string.h>
+
+static void
+check_read(const char* const string,
+           const ExessStatus expected_status,
+           const double      expected_value,
+           const size_t      expected_count)
+{
+  double            value = 0;
+  const ExessResult r     = exess_read_decimal(&value, string);
+
+  assert(r.status == expected_status);
+  assert(r.count == expected_count);
+  assert(double_matches(value, expected_value));
+}
+
+static void
+test_read_decimal(void)
+{
+  // No value
+  check_read("", EXESS_EXPECTED_DIGIT, (double)NAN, 0);
+  check_read(" \f\n\r\t\v", EXESS_EXPECTED_DIGIT, (double)NAN, 6);
+
+  // Basic values
+  check_read("1.2", EXESS_SUCCESS, 1.2, 3);
+  check_read("0.01", EXESS_SUCCESS, 0.01, 4);
+  check_read("10.0", EXESS_SUCCESS, 10.0, 4);
+
+  // Non-canonical form
+  check_read(" \f\n\r\t\v42.24 ", EXESS_SUCCESS, 42.24, 11);
+  check_read("12.", EXESS_SUCCESS, 12., 3);
+  check_read(".34", EXESS_SUCCESS, 0.34, 3);
+  check_read("+.56", EXESS_SUCCESS, 0.56, 4);
+  check_read("-.78", EXESS_SUCCESS, -0.78, 4);
+
+  // Limits
+  check_read(
+    "0."
+    "00000000000000000000000000000000000000000000000000000000000000000000000000"
+    "00000000000000000000000000000000000000000000000000000000000000000000000000"
+    "00000000000000000000000000000000000000000000000000000000000000000000000000"
+    "00000000000000000000000000000000000000000000000000000000000000000000000000"
+    "0000000000022250738585072014",
+    EXESS_SUCCESS,
+    DBL_MIN,
+    326);
+
+  check_read("1797693134862315700000000000000000000000000000000000000000000000"
+             "0000000000000000000000000000000000000000000000000000000000000000"
+             "0000000000000000000000000000000000000000000000000000000000000000"
+             "0000000000000000000000000000000000000000000000000000000000000000"
+             "00000000000000000000000000000000000000000000000000000.0",
+             EXESS_SUCCESS,
+             DBL_MAX,
+             311);
+
+  // Superfluous digits
+  check_read("12345678901234567890", EXESS_SUCCESS, 12345678901234568000.0, 20);
+  check_read("1.2345678901234567890", EXESS_SUCCESS, 1.2345678901234568, 21);
+
+  // Special values
+  check_read("-0.0E0", EXESS_EXPECTED_END, -0.0, 4);
+  check_read("0.0E0", EXESS_EXPECTED_END, 0.0, 3);
+  check_read("+0.0E0", EXESS_EXPECTED_END, 0.0, 4);
+
+  // No exponent
+  check_read("1", EXESS_SUCCESS, 1.0, 1);
+  check_read("2.3", EXESS_SUCCESS, 2.3, 3);
+  check_read("-4.5", EXESS_SUCCESS, -4.5, 4);
+
+  // Garbage
+  check_read("NaN", EXESS_EXPECTED_DIGIT, (double)NAN, 0);
+  check_read("INF", EXESS_EXPECTED_DIGIT, (double)NAN, 0);
+  check_read("-INF", EXESS_EXPECTED_DIGIT, (double)NAN, 1);
+  check_read("true", EXESS_EXPECTED_DIGIT, (double)NAN, 0);
+  check_read("+true", EXESS_EXPECTED_DIGIT, (double)NAN, 1);
+  check_read("-false", EXESS_EXPECTED_DIGIT, (double)NAN, 1);
+}
+
+static void
+test_decimal_string_length(void)
+{
+  // Basic values
+  assert(exess_write_decimal(-1.0, 0, NULL).count == 4);
+  assert(exess_write_decimal(-0.0, 0, NULL).count == 4);
+  assert(exess_write_decimal(0.0, 0, NULL).count == 3);
+  assert(exess_write_decimal(1.0, 0, NULL).count == 3);
+
+  // Limits
+  assert(exess_write_decimal(DBL_MIN, 0, NULL).count == 326);
+  assert(exess_write_decimal(DBL_MAX, 0, NULL).count == 311);
+
+  // Special values
+  assert(exess_write_decimal((double)NAN, 0, NULL).count == 0);
+  assert(exess_write_decimal(-0.0, 0, NULL).count == 4);
+  assert(exess_write_decimal(0.0, 0, NULL).count == 3);
+  assert(exess_write_decimal((double)INFINITY, 0, NULL).count == 0);
+  assert(exess_write_decimal((double)-INFINITY, 0, NULL).count == 0);
+}
+
+/// Check that `str` is a canonical xsd:double string
+static void
+check_canonical(const char* const str)
+{
+  assert(strlen(str) > 2); // Shortest possible is something like 1.2
+  assert(str[0] == '-' || is_digit(str[0]));
+
+  const int first_digit = str[0] == '-' ? 1 : 0;
+
+  for (const char* s = str + first_digit; *s; ++s) {
+    assert(*s == '.' || is_digit(*s));
+  }
+}
+
+static void
+check_write(const double      value,
+            const ExessStatus expected_status,
+            const size_t      buf_size,
+            const char* const expected_string)
+{
+  char buf[EXESS_MAX_DECIMAL_LENGTH + 1] = {42};
+
+  assert(buf_size <= sizeof(buf));
+
+  const ExessResult r = exess_write_decimal(value, buf_size, buf);
+  assert(r.status == expected_status);
+  if (expected_string) {
+    assert(r.count == strlen(buf));
+    assert(!strcmp(buf, expected_string));
+    assert(r.status || exess_write_decimal(value, 0, NULL).count == r.count);
+
+    if (expected_string[0]) {
+      check_canonical(buf);
+    }
+  }
+}
+
+static void
+test_write_decimal(void)
+{
+  check_write((double)NAN, EXESS_BAD_VALUE, 4, "");
+  check_write((double)-INFINITY, EXESS_BAD_VALUE, 5, "");
+
+  check_write(
+    DBL_MIN,
+    EXESS_SUCCESS,
+    327,
+    "0."
+    "00000000000000000000000000000000000000000000000000000000000000000000000000"
+    "00000000000000000000000000000000000000000000000000000000000000000000000000"
+    "00000000000000000000000000000000000000000000000000000000000000000000000000"
+    "00000000000000000000000000000000000000000000000000000000000000000000000000"
+    "0000000000022250738585072014");
+
+  check_write(-1.2, EXESS_SUCCESS, 5, "-1.2");
+  check_write(-0.0, EXESS_SUCCESS, 5, "-0.0");
+  check_write(0.0, EXESS_SUCCESS, 4, "0.0");
+  check_write(1.2, EXESS_SUCCESS, 4, "1.2");
+
+  check_write(DBL_MAX,
+              EXESS_SUCCESS,
+              312,
+              "1797693134862315700000000000000000000000000000000000000000000000"
+              "0000000000000000000000000000000000000000000000000000000000000000"
+              "0000000000000000000000000000000000000000000000000000000000000000"
+              "0000000000000000000000000000000000000000000000000000000000000000"
+              "00000000000000000000000000000000000000000000000000000.0");
+
+  check_write((double)INFINITY, EXESS_BAD_VALUE, 4, "");
+
+  check_write(DBL_MIN, EXESS_NO_SPACE, 326, "");
+  check_write(-1.2, EXESS_NO_SPACE, 4, "");
+  check_write(-0.0, EXESS_NO_SPACE, 4, "");
+  check_write(0.0, EXESS_NO_SPACE, 3, "");
+  check_write(1.2, EXESS_NO_SPACE, 3, "");
+  check_write(DBL_MAX, EXESS_NO_SPACE, 311, "");
+
+  check_write(-1.0, EXESS_NO_SPACE, 1, "");
+  check_write(-1.0, EXESS_NO_SPACE, 0, NULL);
+}
+
+static void
+check_round_trip(const double value)
+{
+  double parsed_value                      = 0.0;
+  char   buf[EXESS_MAX_DECIMAL_LENGTH + 1] = {42};
+
+  /* fprintf(stderr, "%f\n", value); */
+  assert(!exess_write_decimal(value, sizeof(buf), buf).status);
+  /* fprintf(stderr, "Buf: %s\n", buf); */
+  assert(!exess_read_decimal(&parsed_value, buf).status);
+  assert(double_matches(parsed_value, value));
+}
+
+static void
+test_round_trip(const ExessNumTestOptions opts)
+{
+  check_round_trip(DBL_MIN);
+  check_round_trip(-0.0);
+  check_round_trip(0.0);
+  check_round_trip(DBL_MAX);
+
+  fprintf(stderr, "Testing xsd:double randomly with seed %u\n", opts.seed);
+
+  uint32_t rep = opts.seed;
+  for (uint64_t i = 0; i < opts.n_tests; ++i) {
+    rep = lcg32(rep);
+
+    const double value = double_from_rep(rep);
+
+    check_round_trip(value);
+    print_num_test_progress(i, opts.n_tests);
+  }
+}
+
+int
+main(int argc, char** argv)
+{
+  const ExessNumTestOptions opts = parse_num_test_options(argc, argv);
+  if (opts.error) {
+    return 1;
+  }
+
+  test_read_decimal();
+  test_decimal_string_length();
+  test_write_decimal();
+  test_round_trip(opts);
+
+  return 0;
+}
diff --git a/subprojects/exess/test/test_double.c b/subprojects/exess/test/test_double.c
new file mode 100644
index 00000000..d4901c5a
--- /dev/null
+++ b/subprojects/exess/test/test_double.c
@@ -0,0 +1,246 @@
+/*
+  Copyright 2011-2021 David Robillard <d@drobilla.net>
+
+  Permission to use, copy, modify, and/or distribute this software for any
+  purpose with or without fee is hereby granted, provided that the above
+  copyright notice and this permission notice appear in all copies.
+
+  THIS SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+  WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+  MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+  ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+  WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+  ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+  OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+*/
+
+#undef NDEBUG
+
+#include "float_test_data.h"
+#include "int_test_data.h"
+#include "num_test_utils.h"
+#include "string_utils.h"
+
+#include "exess/exess.h"
+
+#include <assert.h>
+#include <float.h>
+#include <math.h>
+#include <stdint.h>
+#include <stdio.h>
+#include <string.h>
+
+static void
+check_read(const char* const string,
+           const ExessStatus expected_status,
+           const double      expected_value,
+           const size_t      expected_count)
+{
+  double            value = (double)NAN;
+  const ExessResult r     = exess_read_double(&value, string);
+
+  assert(r.status == expected_status);
+  assert(r.count == expected_count);
+  assert(double_matches(value, expected_value));
+}
+
+static void
+test_read_double(void)
+{
+  // Limits
+  check_read("-1.7976931348623157E308", EXESS_SUCCESS, -DBL_MAX, 23);
+  check_read("-2.2250738585072014E-308", EXESS_SUCCESS, -DBL_MIN, 24);
+  check_read("2.2250738585072014E-308", EXESS_SUCCESS, DBL_MIN, 23);
+  check_read("1.7976931348623157E308", EXESS_SUCCESS, DBL_MAX, 22);
+
+  // Beyond limits
+  check_read("1e-326", EXESS_SUCCESS, 0.0, 6);
+  check_read("12345678901234567123", EXESS_SUCCESS, 12345678901234567000.0, 20);
+  check_read("1e309", EXESS_SUCCESS, (double)INFINITY, 5);
+
+  // Non-canonical form
+  check_read("+1E3", EXESS_SUCCESS, 1e3, 4);
+  check_read("1E+3", EXESS_SUCCESS, 1e3, 4);
+  check_read("+1.5E3", EXESS_SUCCESS, 1.5e3, 6);
+  check_read(".5E3", EXESS_SUCCESS, 0.5e3, 4);
+  check_read("+.5E3", EXESS_SUCCESS, 0.5e3, 5);
+  check_read("-.5E3", EXESS_SUCCESS, -0.5e3, 5);
+  check_read("1.E3", EXESS_SUCCESS, 1e3, 4);
+  check_read("+1.E3", EXESS_SUCCESS, 1e3, 5);
+  check_read("-1.E3", EXESS_SUCCESS, -1e3, 5);
+
+  // Special values
+  check_read("NaN", EXESS_SUCCESS, (double)NAN, 3);
+  check_read("-INF", EXESS_SUCCESS, (double)-INFINITY, 4);
+  check_read("-0.0E0", EXESS_SUCCESS, -0.0, 6);
+  check_read("0.0E0", EXESS_SUCCESS, 0.0, 5);
+  check_read("+0.0E0", EXESS_SUCCESS, 0.0, 6);
+  check_read("INF", EXESS_SUCCESS, (double)INFINITY, 3);
+  check_read("+INF", EXESS_SUCCESS, (double)INFINITY, 4);
+
+  // No exponent
+  check_read("1", EXESS_SUCCESS, 1.0, 1);
+  check_read("2.3", EXESS_SUCCESS, 2.3, 3);
+  check_read("-4.5", EXESS_SUCCESS, -4.5, 4);
+
+  // Leading whitespace
+  check_read(" \f\n\r\t\v1.2", EXESS_SUCCESS, 1.2, 9);
+
+  // Garbage
+  check_read("true", EXESS_EXPECTED_DIGIT, (double)NAN, 0);
+  check_read("+true", EXESS_EXPECTED_DIGIT, (double)NAN, 1);
+  check_read("-false", EXESS_EXPECTED_DIGIT, (double)NAN, 1);
+  check_read("1.0eX", EXESS_EXPECTED_DIGIT, (double)NAN, 4);
+  check_read("1.0EX", EXESS_EXPECTED_DIGIT, (double)NAN, 4);
+}
+
+/// Check that `str` is a canonical xsd:double string
+static void
+check_canonical(const char* const str)
+{
+  if (!strcmp(str, "NaN") || !strcmp(str, "-INF") || !strcmp(str, "INF")) {
+    return;
+  }
+
+  assert(strlen(str) > 4); // Shortest possible is something like 1.2E3
+  assert(str[0] == '-' || is_digit(str[0]));
+
+  const int first_digit = str[0] == '-' ? 1 : 0;
+  assert(is_digit(str[first_digit]));
+  assert(str[first_digit + 1] == '.');
+  assert(is_digit(str[first_digit + 2]));
+
+  const char* const e = strchr(str, 'E');
+  assert(e);
+  assert(*e == 'E');
+  assert(*(e + 1) == '-' || is_digit(*(e + 1)));
+}
+
+static void
+check_write(const double      value,
+            const ExessStatus expected_status,
+            const size_t      buf_size,
+            const char* const expected_string)
+{
+  char buf[EXESS_MAX_DOUBLE_LENGTH + 1] = {
+    1,  2,  3,  4,  5,  6,  7,  8,  9,  10, //
+    11, 12, 13, 14, 15, 16, 17, 18, 19, 20, //
+    21, 22, 23};
+
+  assert(buf_size <= sizeof(buf));
+
+  const ExessResult r = exess_write_double(value, buf_size, buf);
+  assert(r.status == expected_status);
+  assert(r.count == strlen(buf));
+  assert(!strcmp(buf, expected_string));
+  assert((r.status && r.status != EXESS_NO_SPACE) ||
+         exess_write_double(value, 0, NULL).count == r.count);
+
+  if (!r.status) {
+    check_canonical(buf);
+  }
+}
+
+static void
+test_write_double(void)
+{
+  check_write((double)NAN, EXESS_SUCCESS, 4, "NaN");
+  check_write(DBL_MIN, EXESS_SUCCESS, 24, "2.2250738585072014E-308");
+  check_write(-0.0, EXESS_SUCCESS, 7, "-0.0E0");
+  check_write(0.0, EXESS_SUCCESS, 6, "0.0E0");
+  check_write(DBL_MAX, EXESS_SUCCESS, 23, "1.7976931348623157E308");
+
+  /* check_write((double)NAN, EXESS_NO_SPACE, 3, ""); */
+  /* check_write(DBL_MIN, EXESS_SUCCESS, 24, "2.2250738585072014E-308"); */
+  /* check_write(-0.0, EXESS_SUCCESS, 7, "-0.0E0"); */
+  /* check_write(0.0, EXESS_SUCCESS, 6, "0.0E0"); */
+  /* check_write(DBL_MAX, EXESS_SUCCESS, 23, "1.7976931348623157E308"); */
+}
+
+static void
+check_round_trip(const double value)
+{
+  double parsed_value = 0.0;
+
+  char buf[EXESS_MAX_DOUBLE_LENGTH + 1] = {
+    1,  2,  3,  4,  5,  6,  7,  8,  9,  10, //
+    11, 12, 13, 14, 15, 16, 17, 18, 19, 20, //
+    21, 22, 23};
+
+  assert(!exess_write_double(value, sizeof(buf), buf).status);
+  assert(!exess_read_double(&parsed_value, buf).status);
+  assert(double_matches(parsed_value, value));
+}
+
+static void
+test_round_trip(const ExessNumTestOptions opts)
+{
+  check_round_trip((double)NAN);
+  check_round_trip(-(double)INFINITY);
+  check_round_trip(DBL_MIN);
+  check_round_trip(-0.0);
+  check_round_trip(0.0);
+  check_round_trip(DBL_MAX);
+  check_round_trip((double)INFINITY);
+
+  check_round_trip(5.0);
+  check_round_trip(50.0);
+  check_round_trip(500000000000000000000.0);
+  check_round_trip(-0.5);
+  check_round_trip(0.5);
+  check_round_trip(0.05);
+  check_round_trip(0.005);
+  check_round_trip(0.00000000000000000005);
+
+  // Normal limits
+  check_round_trip(nextafter(DBL_MIN, (double)INFINITY));
+  check_round_trip(nextafter(DBL_EPSILON, (double)INFINITY));
+  check_round_trip(nextafter(DBL_MAX, -(double)INFINITY));
+
+  // Subnormals
+  check_round_trip(nextafter(0.0, 1.0));
+  check_round_trip(nextafter(nextafter(0.0, 1.0), 2.0));
+  check_round_trip(nextafter(0.0, -1.0));
+  check_round_trip(nextafter(nextafter(0.0, -1.0), -2.0));
+
+  // Various tricky cases
+  check_round_trip(1e23);
+  check_round_trip(6.02951420360127e-309);
+  check_round_trip(9.17857104364115e+288);
+  check_round_trip(2.68248422823759e+22);
+
+  // Powers of two (where the lower boundary is closer)
+  for (int i = -1023; i <= 1023; ++i) {
+    check_round_trip(pow(2, i));
+  }
+
+  fprintf(stderr, "Testing xsd:double randomly with seed %u\n", opts.seed);
+
+  uint64_t rep = opts.seed;
+  for (uint64_t i = 0; i < opts.n_tests; ++i) {
+    rep = lcg64(rep);
+
+    const double value = double_from_rep(rep);
+
+    check_round_trip(nextafter(value, -(double)INFINITY));
+    check_round_trip(value);
+    check_round_trip(nextafter(value, (double)INFINITY));
+
+    print_num_test_progress(i, opts.n_tests);
+  }
+}
+
+int
+main(int argc, char** argv)
+{
+  const ExessNumTestOptions opts = parse_num_test_options(argc, argv);
+  if (opts.error) {
+    return 1;
+  }
+
+  test_read_double();
+  test_write_double();
+  test_round_trip(opts);
+
+  return 0;
+}
diff --git a/subprojects/exess/test/test_duration.c b/subprojects/exess/test/test_duration.c
new file mode 100644
index 00000000..3d964e80
--- /dev/null
+++ b/subprojects/exess/test/test_duration.c
@@ -0,0 +1,310 @@
+/*
+  Copyright 2011-2021 David Robillard <d@drobilla.net>
+
+  Permission to use, copy, modify, and/or distribute this software for any
+  purpose with or without fee is hereby granted, provided that the above
+  copyright notice and this permission notice appear in all copies.
+
+  THIS SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+  WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+  MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+  ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+  WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+  ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+  OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+*/
+
+#undef NDEBUG
+
+#include "int_test_data.h"
+#include "macros.h"
+#include "num_test_utils.h"
+
+#include "exess/exess.h"
+
+#include <assert.h>
+#include <stdbool.h>
+#include <stdint.h>
+#include <stdio.h>
+#include <string.h>
+
+static const ExessDuration zero       = {0, 0, 0};
+static const ExessDuration lowest     = {-INT32_MAX, -INT32_MAX, -999999999};
+static const ExessDuration highest    = {INT32_MAX, INT32_MAX, 999999999};
+static const ExessDuration year       = {12, 0, 0};
+static const ExessDuration month      = {1, 0, 0};
+static const ExessDuration day        = {0, 24 * 60 * 60, 0};
+static const ExessDuration hour       = {0, 60 * 60, 0};
+static const ExessDuration minute     = {0, 60, 0};
+static const ExessDuration second     = {0, 1, 0};
+static const ExessDuration nanosecond = {0, 0, 1};
+
+static const ExessDuration n_year       = {-12, 0, 0};
+static const ExessDuration n_month      = {-1, 0, 0};
+static const ExessDuration n_day        = {0, -24 * 60 * 60, 0};
+static const ExessDuration n_hour       = {0, -60 * 60, 0};
+static const ExessDuration n_minute     = {0, -60, 0};
+static const ExessDuration n_second     = {0, -1, 0};
+static const ExessDuration n_nanosecond = {0, 0, -1};
+
+static const ExessDuration garbage1 = {1, 1, -1};
+static const ExessDuration garbage2 = {1, -1, 1};
+static const ExessDuration garbage3 = {1, -1, -1};
+static const ExessDuration garbage4 = {-1, 1, 1};
+static const ExessDuration garbage5 = {-1, 1, -1};
+static const ExessDuration garbage6 = {-1, -1, 1};
+static const ExessDuration garbage7 = {INT32_MIN, 0, -999999999};
+static const ExessDuration garbage8 = {0, INT32_MIN, -999999999};
+static const ExessDuration garbage9 = {INT32_MIN, INT32_MIN, -999999999};
+
+static void
+check_read(const char* const string,
+           const ExessStatus expected_status,
+           const size_t      expected_count,
+           const int32_t     expected_years,
+           const int32_t     expected_months,
+           const int32_t     expected_days,
+           const int32_t     expected_hours,
+           const int32_t     expected_minutes,
+           const int32_t     expected_seconds,
+           const int32_t     expected_nanoseconds,
+           const bool        expected_is_negative)
+{
+  ExessDuration     value = {0, 0, 0.0f};
+  const ExessResult r     = exess_read_duration(&value, string);
+
+  assert(r.status == expected_status);
+  assert(r.count == expected_count);
+
+  assert(value.months == (expected_is_negative ? -1 : 1) * 12 * expected_years +
+                           expected_months);
+
+  assert(value.seconds ==
+         (expected_is_negative ? -1 : 1) *
+           ((expected_seconds + (60 * expected_minutes) +
+             (60 * 60 * expected_hours) + (24 * 60 * 60 * expected_days))));
+
+  assert(value.nanoseconds == expected_nanoseconds);
+}
+
+static void
+test_read_duration(void)
+{
+  // No input
+  check_read("", EXESS_EXPECTED_DURATION, 0, 0, 0, 0, 0, 0, 0, 0, false);
+  check_read(
+    " \f\n\r\t\v", EXESS_EXPECTED_DURATION, 6, 0, 0, 0, 0, 0, 0, 0, false);
+
+  // Good values
+
+  check_read(
+    "P2Y6M5DT12H35M30S", EXESS_SUCCESS, 17, 2, 6, 5, 12, 35, 30, 0, false);
+
+  check_read("P1DT2H", EXESS_SUCCESS, 6, 0, 0, 1, 2, 0, 0, 0, false);
+  check_read("P20M", EXESS_SUCCESS, 4, 0, 20, 0, 0, 0, 0, 0, false);
+  check_read("PT20M", EXESS_SUCCESS, 5, 0, 0, 0, 0, 20, 0, 0, false);
+  check_read("P0Y20M0D", EXESS_SUCCESS, 8, 0, 20, 0, 0, 0, 0, 0, false);
+  check_read("P0Y", EXESS_SUCCESS, 3, 0, 0, 0, 0, 0, 0, 0, false);
+  check_read("-P60D", EXESS_SUCCESS, 5, 0, 0, 60, 0, 0, 0, 0, true);
+  check_read(
+    "PT1M30.5S", EXESS_SUCCESS, 9, 0, 0, 0, 0, 1, 30, 500000000, false);
+
+  // Leading and trailing whitespace
+  check_read(" \f\n\r\t\vP1Y", EXESS_SUCCESS, 9, 1, 0, 0, 0, 0, 0, 0, false);
+  check_read("P1MT2H \f\n\r\t\v", EXESS_SUCCESS, 6, 0, 1, 0, 2, 0, 0, 0, false);
+  check_read(" \f\n\r\t\vP1Y", EXESS_SUCCESS, 9, 1, 0, 0, 0, 0, 0, 0, false);
+  check_read("P1YT2H \f\n\r\t\v", EXESS_SUCCESS, 6, 1, 0, 0, 2, 0, 0, 0, false);
+
+  // Non-canonical form
+  check_read("P06D", EXESS_SUCCESS, 4, 0, 0, 6, 0, 0, 0, 0, false);
+  check_read("PT7.0S", EXESS_SUCCESS, 6, 0, 0, 0, 0, 0, 7, 0, false);
+  check_read(
+    "P0Y0M01DT06H00M00S", EXESS_SUCCESS, 18, 0, 0, 1, 6, 0, 0, 0, false);
+
+  // Out of range fields
+  check_read(
+    "P2147483647Y", EXESS_OUT_OF_RANGE, 11, 0, 0, 0, 0, 0, 0, 0, false);
+  check_read(
+    "P2147483647M", EXESS_OUT_OF_RANGE, 11, 0, 0, 0, 0, 0, 0, 0, false);
+  check_read(
+    "P2147483647D", EXESS_OUT_OF_RANGE, 11, 0, 0, 0, 0, 0, 0, 0, false);
+  check_read(
+    "PT2147483647H", EXESS_OUT_OF_RANGE, 12, 0, 0, 0, 0, 0, 0, 0, false);
+  check_read(
+    "PT2147483647M", EXESS_OUT_OF_RANGE, 12, 0, 0, 0, 0, 0, 0, 0, false);
+  check_read(
+    "PT2147483647S", EXESS_OUT_OF_RANGE, 12, 0, 0, 0, 0, 0, 0, 0, false);
+
+  // Garbage
+  check_read("P-20M", EXESS_EXPECTED_DIGIT, 1, 0, 0, 0, 0, 0, 0, 0, false);
+  check_read("P20MT", EXESS_EXPECTED_DIGIT, 5, 0, 20, 0, 0, 0, 0, 0, false);
+  check_read("P1YM5D", EXESS_EXPECTED_DIGIT, 3, 1, 0, 0, 0, 0, 0, 0, false);
+  check_read("P15.5Y", EXESS_EXPECTED_DATE_TAG, 3, 0, 0, 0, 0, 0, 0, 0, false);
+  check_read("P1D2H", EXESS_EXPECTED_TIME_SEP, 3, 0, 0, 1, 0, 0, 0, 0, false);
+  check_read("1Y2M", EXESS_EXPECTED_DURATION, 0, 0, 0, 0, 0, 0, 0, 0, false);
+  check_read("P2M1Y", EXESS_BAD_ORDER, 4, 0, 2, 0, 0, 0, 0, 0, false);
+  check_read("P2D1Y", EXESS_EXPECTED_TIME_SEP, 3, 0, 0, 2, 0, 0, 0, 0, false);
+  check_read("P2D1M", EXESS_EXPECTED_TIME_SEP, 3, 0, 0, 2, 0, 0, 0, 0, false);
+  check_read("P", EXESS_EXPECTED_DIGIT, 1, 0, 0, 0, 0, 0, 0, 0, false);
+  check_read("PT15.5H", EXESS_EXPECTED_TIME_TAG, 6, 0, 0, 0, 0, 0, 0, 0, false);
+  check_read("PT2M1H", EXESS_BAD_ORDER, 5, 0, 0, 0, 0, 2, 0, 0, false);
+  check_read("PT2S1H", EXESS_EXPECTED_END, 4, 0, 0, 0, 0, 0, 2, 0, false);
+  check_read("PT2S1M", EXESS_EXPECTED_END, 4, 0, 0, 0, 0, 0, 2, 0, false);
+  check_read("PT15.S", EXESS_EXPECTED_DIGIT, 5, 0, 0, 0, 0, 0, 0, 0, false);
+  check_read("P1Q", EXESS_EXPECTED_DATE_TAG, 2, 0, 0, 0, 0, 0, 0, 0, false);
+  check_read("PT1Q", EXESS_EXPECTED_TIME_TAG, 3, 0, 0, 0, 0, 0, 0, 0, false);
+  check_read("P-1Y", EXESS_EXPECTED_DIGIT, 1, 0, 0, 0, 0, 0, 0, 0, false);
+  check_read("P-1M", EXESS_EXPECTED_DIGIT, 1, 0, 0, 0, 0, 0, 0, 0, false);
+  check_read("P-1D", EXESS_EXPECTED_DIGIT, 1, 0, 0, 0, 0, 0, 0, 0, false);
+  check_read("PT-1H", EXESS_EXPECTED_DIGIT, 2, 0, 0, 0, 0, 0, 0, 0, false);
+  check_read("PT-1M", EXESS_EXPECTED_DIGIT, 2, 0, 0, 0, 0, 0, 0, 0, false);
+  check_read("PT-1S", EXESS_EXPECTED_DIGIT, 2, 0, 0, 0, 0, 0, 0, 0, false);
+  check_read(
+    "P4294967296Y", EXESS_OUT_OF_RANGE, 11, 0, 0, 0, 0, 0, 0, 0, false);
+  check_read(
+    "P4294967296M", EXESS_OUT_OF_RANGE, 11, 0, 0, 0, 0, 0, 0, 0, false);
+  check_read(
+    "P4294967296D", EXESS_OUT_OF_RANGE, 11, 0, 0, 0, 0, 0, 0, 0, false);
+  check_read(
+    "PT4294967296H", EXESS_OUT_OF_RANGE, 12, 0, 0, 0, 0, 0, 0, 0, false);
+  check_read(
+    "PT4294967296M", EXESS_OUT_OF_RANGE, 12, 0, 0, 0, 0, 0, 0, 0, false);
+  check_read("", EXESS_EXPECTED_DURATION, 0, 0, 0, 0, 0, 0, 0, 0, false);
+}
+
+static void
+check_write(const ExessDuration value,
+            const ExessStatus   expected_status,
+            const size_t        buf_size,
+            const char* const   expected_string)
+{
+  char buf[EXESS_MAX_DURATION_LENGTH + 1] = {42};
+
+  assert(buf_size <= sizeof(buf));
+
+  const ExessResult r = exess_write_duration(value, buf_size, buf);
+  assert(r.status == expected_status);
+  assert(r.count == strlen(buf));
+  assert(!strcmp(buf, expected_string));
+  assert(r.status || exess_write_duration(value, 0, NULL).count == r.count);
+}
+
+static void
+test_write_duration(void)
+{
+  check_write(zero, EXESS_SUCCESS, 4, "P0Y");
+
+  check_write(
+    lowest, EXESS_SUCCESS, 39, "-P178956970Y7M24855DT3H14M7.999999999S");
+
+  check_write(
+    highest, EXESS_SUCCESS, 38, "P178956970Y7M24855DT3H14M7.999999999S");
+
+  check_write(year, EXESS_SUCCESS, 4, "P1Y");
+  check_write(month, EXESS_SUCCESS, 4, "P1M");
+  check_write(day, EXESS_SUCCESS, 4, "P1D");
+  check_write(hour, EXESS_SUCCESS, 5, "PT1H");
+  check_write(minute, EXESS_SUCCESS, 5, "PT1M");
+  check_write(second, EXESS_SUCCESS, 5, "PT1S");
+  check_write(nanosecond, EXESS_SUCCESS, 15, "PT0.000000001S");
+
+  check_write(n_year, EXESS_SUCCESS, 5, "-P1Y");
+  check_write(n_month, EXESS_SUCCESS, 5, "-P1M");
+  check_write(n_day, EXESS_SUCCESS, 5, "-P1D");
+  check_write(n_hour, EXESS_SUCCESS, 6, "-PT1H");
+  check_write(n_minute, EXESS_SUCCESS, 6, "-PT1M");
+  check_write(n_second, EXESS_SUCCESS, 6, "-PT1S");
+  check_write(n_nanosecond, EXESS_SUCCESS, 16, "-PT0.000000001S");
+
+  check_write(garbage1, EXESS_BAD_VALUE, 41, "");
+  check_write(garbage2, EXESS_BAD_VALUE, 41, "");
+  check_write(garbage3, EXESS_BAD_VALUE, 41, "");
+  check_write(garbage4, EXESS_BAD_VALUE, 41, "");
+  check_write(garbage5, EXESS_BAD_VALUE, 41, "");
+  check_write(garbage6, EXESS_BAD_VALUE, 41, "");
+  check_write(garbage7, EXESS_OUT_OF_RANGE, 41, "");
+  check_write(garbage8, EXESS_OUT_OF_RANGE, 41, "");
+  check_write(garbage9, EXESS_OUT_OF_RANGE, 41, "");
+
+  check_write(zero, EXESS_NO_SPACE, 3, "");
+  check_write(lowest, EXESS_NO_SPACE, 24, "");
+  check_write(highest, EXESS_NO_SPACE, 4, "");
+  check_write(highest, EXESS_NO_SPACE, 10, "");
+  check_write(highest, EXESS_NO_SPACE, 13, "");
+  check_write(highest, EXESS_NO_SPACE, 16, "");
+  check_write(highest, EXESS_NO_SPACE, 20, "");
+  check_write(highest, EXESS_NO_SPACE, 23, "");
+  check_write(year, EXESS_NO_SPACE, 3, "");
+  check_write(month, EXESS_NO_SPACE, 3, "");
+  check_write(day, EXESS_NO_SPACE, 3, "");
+  check_write(hour, EXESS_NO_SPACE, 4, "");
+  check_write(minute, EXESS_NO_SPACE, 4, "");
+  check_write(second, EXESS_NO_SPACE, 4, "");
+
+  // Check that nothing is written when there isn't enough space
+  char              c = 42;
+  const ExessResult r = exess_write_duration(zero, 0, &c);
+  assert(c == 42);
+  assert(r.status == EXESS_NO_SPACE);
+  assert(r.count == 0);
+}
+
+static void
+check_round_trip(const ExessDuration value)
+{
+  ExessDuration parsed_value                       = {0, 0, 0};
+  char          buf[EXESS_MAX_DURATION_LENGTH + 1] = {0};
+
+  assert(exess_write_duration(value, 0, NULL).count <=
+         EXESS_MAX_DURATION_LENGTH);
+
+  assert(!exess_write_duration(value, sizeof(buf), buf).status);
+  assert(!exess_read_duration(&parsed_value, buf).status);
+  assert(parsed_value.months == value.months);
+  assert(parsed_value.seconds == value.seconds);
+  assert(parsed_value.nanoseconds == value.nanoseconds);
+}
+
+static void
+test_round_trip(const ExessNumTestOptions opts)
+{
+  fprintf(stderr, "Testing xsd:duration randomly with seed %u\n", opts.seed);
+
+  const uint64_t n_tests = MAX(256, opts.n_tests / 16);
+
+  uint32_t rng = opts.seed;
+  for (size_t i = 0; i < n_tests; ++i) {
+    rng = lcg32(rng);
+
+    const int32_t months = (int32_t)rng;
+
+    rng = lcg32(rng);
+
+    const int32_t seconds = (months < 0 ? -1 : 1) * (int32_t)(rng % INT32_MAX);
+
+    rng = lcg32(rng);
+
+    const int32_t nanoseconds =
+      (months < 0 ? -1 : 1) * (int32_t)(rng % 1000000000);
+
+    const ExessDuration value = {months, seconds, nanoseconds};
+    check_round_trip(value);
+
+    print_num_test_progress(i, n_tests);
+  }
+}
+
+int
+main(int argc, char** argv)
+{
+  const ExessNumTestOptions opts = parse_num_test_options(argc, argv);
+  if (opts.error) {
+    return 1;
+  }
+
+  test_read_duration();
+  test_write_duration();
+  test_round_trip(opts);
+
+  return 0;
+}
diff --git a/subprojects/exess/test/test_float.c b/subprojects/exess/test/test_float.c
new file mode 100644
index 00000000..e8cf9141
--- /dev/null
+++ b/subprojects/exess/test/test_float.c
@@ -0,0 +1,240 @@
+/*
+  Copyright 2011-2021 David Robillard <d@drobilla.net>
+
+  Permission to use, copy, modify, and/or distribute this software for any
+  purpose with or without fee is hereby granted, provided that the above
+  copyright notice and this permission notice appear in all copies.
+
+  THIS SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+  WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+  MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+  ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+  WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+  ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+  OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+*/
+
+#undef NDEBUG
+
+#include "float_test_data.h"
+#include "int_test_data.h"
+#include "num_test_utils.h"
+#include "string_utils.h"
+
+#include "exess/exess.h"
+
+#include <assert.h>
+#include <float.h>
+#include <math.h>
+#include <stdint.h>
+#include <stdio.h>
+#include <string.h>
+
+static void
+check_read(const char* const string,
+           const ExessStatus expected_status,
+           const float       expected_value,
+           const size_t      expected_count)
+{
+  float             value = NAN;
+  const ExessResult r     = exess_read_float(&value, string);
+
+  assert(r.status == expected_status);
+  assert(r.count == expected_count);
+  assert(float_matches(value, expected_value));
+}
+
+static void
+test_read_float(void)
+{
+  // Limits
+  check_read("-3.40282347E38", EXESS_SUCCESS, -FLT_MAX, 14);
+  check_read("-1.17549435E-38", EXESS_SUCCESS, -FLT_MIN, 15);
+  check_read("1.17549435E-38", EXESS_SUCCESS, FLT_MIN, 14);
+  check_read("3.40282347E38", EXESS_SUCCESS, FLT_MAX, 13);
+
+  // Special values
+  check_read("NaN", EXESS_SUCCESS, NAN, 3);
+  check_read("-INF", EXESS_SUCCESS, -INFINITY, 4);
+  check_read("-0.0E0", EXESS_SUCCESS, -0.0f, 6);
+  check_read("0.0E0", EXESS_SUCCESS, 0.0f, 5);
+  check_read("+0.0E0", EXESS_SUCCESS, 0.0f, 6);
+  check_read("INF", EXESS_SUCCESS, INFINITY, 3);
+  check_read("+INF", EXESS_SUCCESS, INFINITY, 4);
+
+  // Various normal cases
+  check_read("-1.0E0", EXESS_SUCCESS, -1.0f, 6);
+  check_read("1.0E0", EXESS_SUCCESS, +1.0f, 5);
+  check_read("5.0E0", EXESS_SUCCESS, 5.0f, 5);
+  check_read("5.0E1", EXESS_SUCCESS, 50.0f, 5);
+  check_read("5.0E9", EXESS_SUCCESS, 5000000000.0f, 5);
+  check_read("-5.0E-1", EXESS_SUCCESS, -0.5f, 7);
+  check_read("5.0E-1", EXESS_SUCCESS, 0.5f, 6);
+  check_read("6.25E-2", EXESS_SUCCESS, 0.0625f, 7);
+  check_read("7.8125E-3", EXESS_SUCCESS, 0.0078125f, 9);
+
+  // No exponent
+  check_read("1", EXESS_SUCCESS, 1.0f, 1);
+  check_read("2.3", EXESS_SUCCESS, 2.3f, 3);
+  check_read("-4.5", EXESS_SUCCESS, -4.5f, 4);
+
+  // Trailing garbage
+  check_read("1.2.", EXESS_SUCCESS, 1.2f, 3);
+
+  // Garbage
+  check_read("true", EXESS_EXPECTED_DIGIT, NAN, 0);
+  check_read("+true", EXESS_EXPECTED_DIGIT, NAN, 1);
+  check_read("-false", EXESS_EXPECTED_DIGIT, NAN, 1);
+  check_read("1.0eX", EXESS_EXPECTED_DIGIT, NAN, 4);
+  check_read("1.0EX", EXESS_EXPECTED_DIGIT, NAN, 4);
+}
+
+static void
+test_float_string_length(void)
+{
+  // Limits
+  assert(exess_write_float(FLT_MIN, 0, NULL).count == 14);
+  assert(exess_write_float(FLT_MAX, 0, NULL).count == 13);
+
+  // Special values
+  assert(exess_write_float((float)NAN, 0, NULL).count == 3);
+  assert(exess_write_float(-1.0f, 0, NULL).count == 6);
+  assert(exess_write_float(-0.0f, 0, NULL).count == 6);
+  assert(exess_write_float(0.0f, 0, NULL).count == 5);
+  assert(exess_write_float(1.0f, 0, NULL).count == 5);
+  assert(exess_write_float((float)INFINITY, 0, NULL).count == 3);
+  assert(exess_write_float((float)-INFINITY, 0, NULL).count == 4);
+}
+
+/// Check that `str` is a canonical xsd:float string
+static void
+check_canonical(const char* const str)
+{
+  if (!strcmp(str, "NaN") || !strcmp(str, "-INF") || !strcmp(str, "INF")) {
+    return;
+  }
+
+  assert(strlen(str) > 4); // Shortest possible is something like 1.2E3
+  assert(str[0] == '-' || is_digit(str[0]));
+
+  const int first_digit = str[0] == '-' ? 1 : 0;
+  assert(is_digit(str[first_digit]));
+  assert(str[first_digit + 1] == '.');
+  assert(is_digit(str[first_digit + 2]));
+
+  const char* const e = strchr(str, 'E');
+  assert(e);
+  assert(*e == 'E');
+  assert(*(e + 1) == '-' || is_digit(*(e + 1)));
+}
+
+static void
+check_write(const float       value,
+            const ExessStatus expected_status,
+            const size_t      buf_size,
+            const char* const expected_string)
+{
+  char buf[EXESS_MAX_FLOAT_LENGTH + 1] = {
+    1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16};
+
+  assert(buf_size <= sizeof(buf));
+
+  const ExessResult r = exess_write_float(value, buf_size, buf);
+  assert(r.status == expected_status);
+  if (expected_string) {
+    assert(r.count == strlen(buf));
+    assert(!expected_string || !strcmp(buf, expected_string));
+    assert(r.status || exess_write_float(value, 0, NULL).count == r.count);
+    check_canonical(buf);
+  } else {
+    assert(r.count == 0);
+  }
+}
+
+static void
+test_write_float(void)
+{
+  check_write(NAN, EXESS_SUCCESS, 4, "NaN");
+  check_write(-INFINITY, EXESS_SUCCESS, 5, "-INF");
+  check_write(FLT_MIN, EXESS_SUCCESS, 15, "1.17549435E-38");
+  check_write(-0.0f, EXESS_SUCCESS, 7, "-0.0E0");
+  check_write(0.0f, EXESS_SUCCESS, 6, "0.0E0");
+  check_write(100.25f, EXESS_SUCCESS, 9, "1.0025E2");
+  check_write(FLT_MAX, EXESS_SUCCESS, 14, "3.40282346E38");
+  check_write(INFINITY, EXESS_SUCCESS, 4, "INF");
+
+  check_write(NAN, EXESS_NO_SPACE, 3, NULL);
+  check_write(-INFINITY, EXESS_NO_SPACE, 4, NULL);
+  check_write(FLT_MIN, EXESS_NO_SPACE, 13, NULL);
+  check_write(-1.0f, EXESS_NO_SPACE, 2, NULL);
+  check_write(-0.0f, EXESS_NO_SPACE, 6, NULL);
+  check_write(0.0f, EXESS_NO_SPACE, 5, NULL);
+  check_write(100.25f, EXESS_NO_SPACE, 5, NULL);
+  check_write(100.25f, EXESS_NO_SPACE, 8, NULL);
+  check_write(FLT_MAX, EXESS_NO_SPACE, 13, NULL);
+  check_write(INFINITY, EXESS_NO_SPACE, 3, NULL);
+}
+
+static void
+check_round_trip(const float value)
+{
+  float parsed_value                    = 0.0f;
+  char  buf[EXESS_MAX_FLOAT_LENGTH + 1] = {
+    1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16};
+
+  assert(!exess_write_float(value, sizeof(buf), buf).status);
+  assert(!exess_read_float(&parsed_value, buf).status);
+  assert(float_matches(parsed_value, value));
+}
+
+static void
+test_round_trip(const ExessNumTestOptions opts)
+{
+  check_round_trip(NAN);
+  check_round_trip(FLT_MIN);
+  check_round_trip(-0.0f);
+  check_round_trip(0.0f);
+  check_round_trip(FLT_MAX);
+
+  if (opts.exhaustive) {
+    fprintf(stderr, "Testing xsd:float exhaustively\n");
+
+    for (int64_t i = 0; i <= UINT32_MAX; ++i) {
+      const float value = float_from_rep((uint32_t)i);
+
+      check_round_trip(value);
+      print_num_test_progress((uint64_t)(i - (int64_t)INT32_MIN), UINT32_MAX);
+    }
+  } else {
+    fprintf(stderr, "Testing xsd:float randomly with seed %u\n", opts.seed);
+
+    uint32_t rep = opts.seed;
+    for (uint64_t i = 0; i < opts.n_tests; ++i) {
+      rep = lcg32(rep);
+
+      const float value = float_from_rep(rep);
+
+      check_round_trip(nextafterf(value, -INFINITY));
+      check_round_trip(value);
+      check_round_trip(nextafterf(value, INFINITY));
+
+      print_num_test_progress(i, opts.n_tests);
+    }
+  }
+}
+
+int
+main(int argc, char** argv)
+{
+  const ExessNumTestOptions opts = parse_num_test_options(argc, argv);
+  if (opts.error) {
+    return 1;
+  }
+
+  test_read_float();
+  test_float_string_length();
+  test_write_float();
+  test_round_trip(opts);
+
+  return 0;
+}
diff --git a/subprojects/exess/test/test_hex.c b/subprojects/exess/test/test_hex.c
new file mode 100644
index 00000000..5ceec90b
--- /dev/null
+++ b/subprojects/exess/test/test_hex.c
@@ -0,0 +1,180 @@
+/*
+  Copyright 2011-2021 David Robillard <d@drobilla.net>
+
+  Permission to use, copy, modify, and/or distribute this software for any
+  purpose with or without fee is hereby granted, provided that the above
+  copyright notice and this permission notice appear in all copies.
+
+  THIS SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+  WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+  MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+  ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+  WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+  ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+  OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+*/
+
+#undef NDEBUG
+
+#include "exess/exess.h"
+
+#include <assert.h>
+#include <stddef.h>
+#include <stdint.h>
+#include <stdlib.h>
+#include <string.h>
+
+static void
+check_read(const char* const string,
+           const ExessStatus expected_status,
+           const size_t      expected_value_length,
+           const char* const expected_value,
+           const size_t      expected_value_size,
+           const size_t      expected_count)
+{
+  char      buf[9] = {0, 0, 0, 0, 0, 0, 0, 0, 0};
+  ExessBlob blob   = {sizeof(buf), buf};
+
+  ExessResult r = exess_read_hex(&blob, string);
+  assert(r.status == expected_status);
+  assert(r.count == expected_count);
+  assert(r.status || blob.size == expected_value_size);
+  if (expected_value_length > 0) {
+    assert(!strncmp(buf, expected_value, expected_value_length));
+    assert(blob.size <= exess_hex_decoded_size(strlen(string)));
+  }
+}
+
+static void
+test_lowercase(void)
+{
+  char      buf[6] = {0, 0, 0, 0, 0, 0};
+  ExessBlob blob   = {sizeof(buf), buf};
+
+  ExessResult r = exess_read_hex(&blob, "6A6B6C6D6E6F");
+  assert(r.status == EXESS_SUCCESS);
+  assert(r.count == 12);
+  assert(blob.size == 6);
+  assert(!strncmp((const char*)blob.data, "jklmno", 6));
+
+  r = exess_read_hex(&blob, "6a6b6c6d6e6f");
+  assert(r.status == EXESS_SUCCESS);
+  assert(r.count == 12);
+  assert(blob.size == 6);
+  assert(!strncmp((const char*)blob.data, "jklmno", 6));
+}
+
+static void
+test_whitespace(void)
+{
+  check_read("666F6F", EXESS_SUCCESS, 3, "foo", 3, 6);
+  check_read(" 666F6F", EXESS_SUCCESS, 3, "foo", 3, 7);
+  check_read("6\f66F6F", EXESS_SUCCESS, 3, "foo", 3, 7);
+  check_read("66\n6F6F", EXESS_SUCCESS, 3, "foo", 3, 7);
+  check_read("666\rF6F", EXESS_SUCCESS, 3, "foo", 3, 7);
+  check_read("666F\t6F", EXESS_SUCCESS, 3, "foo", 3, 7);
+  check_read(" \f\n\r\t\v666F6F", EXESS_SUCCESS, 3, "foo", 3, 12);
+  check_read("666F6F \f\n\r\t\v", EXESS_SUCCESS, 3, "foo", 3, 12);
+}
+
+static void
+test_syntax_errors(void)
+{
+  check_read("G6", EXESS_EXPECTED_HEX, 0, NULL, 0, 1);
+  check_read("g6", EXESS_EXPECTED_HEX, 0, NULL, 0, 1);
+  check_read("!6", EXESS_EXPECTED_HEX, 0, NULL, 0, 1);
+  check_read("^6", EXESS_EXPECTED_HEX, 0, NULL, 0, 1);
+  check_read("6G", EXESS_EXPECTED_HEX, 0, NULL, 0, 2);
+  check_read("6g", EXESS_EXPECTED_HEX, 0, NULL, 0, 2);
+  check_read("6!", EXESS_EXPECTED_HEX, 0, NULL, 0, 2);
+  check_read("6^", EXESS_EXPECTED_HEX, 0, NULL, 0, 2);
+  check_read("6", EXESS_EXPECTED_HEX, 0, NULL, 0, 1);
+  check_read("66G6", EXESS_EXPECTED_HEX, 0, NULL, 1, 3);
+  check_read("66 G6", EXESS_EXPECTED_HEX, 0, NULL, 1, 4);
+}
+
+static void
+test_read_overflow(void)
+{
+  char      buf[9] = {0, 0, 0, 0, 0, 0, 0, 0, 0};
+  ExessBlob blob0  = {0, buf};
+  ExessBlob blob1  = {1, buf};
+  ExessBlob blob2  = {2, buf};
+
+  ExessResult r = exess_read_hex(&blob0, "666F6F");
+  assert(r.status == EXESS_NO_SPACE);
+  assert(r.count == 2);
+  assert(blob0.size == 0);
+
+  r = exess_read_hex(&blob1, "666F6F");
+  assert(r.status == EXESS_NO_SPACE);
+  assert(r.count == 4);
+  assert(blob1.size == 1);
+
+  r = exess_read_hex(&blob2, "666F6F");
+  assert(r.status == EXESS_NO_SPACE);
+  assert(r.count == 6);
+  assert(blob2.size == 2);
+}
+
+static void
+test_write_overflow(void)
+{
+  char buf[7] = {1, 2, 3, 4, 5, 6, 7};
+
+  assert(exess_write_hex(3, "foo", 0, buf).status == EXESS_NO_SPACE);
+  assert(exess_write_hex(3, "foo", 1, buf).status == EXESS_NO_SPACE);
+  assert(exess_write_hex(3, "foo", 2, buf).status == EXESS_NO_SPACE);
+  assert(exess_write_hex(3, "foo", 3, buf).status == EXESS_NO_SPACE);
+  assert(exess_write_hex(3, "foo", 4, buf).status == EXESS_NO_SPACE);
+  assert(exess_write_hex(3, "foo", 5, buf).status == EXESS_NO_SPACE);
+  assert(exess_write_hex(3, "foo", 6, buf).status == EXESS_NO_SPACE);
+  assert(exess_write_hex(3, "foo", 7, buf).status == EXESS_SUCCESS);
+}
+
+static void
+test_round_trip(void)
+{
+  for (size_t size = 1; size < 256; ++size) {
+    // Allocate and generate data
+    uint8_t* const data = (uint8_t*)malloc(size);
+    for (size_t i = 0; i < size; ++i) {
+      data[i] = (uint8_t)((size + i) % 256);
+    }
+
+    // Allocate buffer for encoding with minimum required size
+    const size_t str_len = exess_write_hex(size, data, 0, NULL).count;
+    char* const  str     = (char*)malloc(str_len + 1);
+
+    // Encode data to string buffer
+    assert(!exess_write_hex(size, data, str_len + 1, str).status);
+    assert(strlen(str) == str_len);
+    assert(str_len % 2 == 0);
+
+    // Allocate buffer for decoded data with the same size as the input
+    uint8_t* const decoded      = (uint8_t*)malloc(size);
+    ExessBlob      decoded_blob = {size, decoded};
+
+    // Decode and check that data matches the original input
+    assert(!exess_read_hex(&decoded_blob, str).status);
+    assert(decoded_blob.size == size);
+    assert(!memcmp(decoded, data, size));
+
+    free(decoded);
+    free(str);
+    free(data);
+  }
+}
+
+int
+main(void)
+{
+  test_lowercase();
+  test_whitespace();
+  test_syntax_errors();
+  test_read_overflow();
+  test_write_overflow();
+  test_round_trip();
+
+  return 0;
+}
diff --git a/subprojects/exess/test/test_int.c b/subprojects/exess/test/test_int.c
new file mode 100644
index 00000000..d3e43f7c
--- /dev/null
+++ b/subprojects/exess/test/test_int.c
@@ -0,0 +1,129 @@
+/*
+  Copyright 2011-2021 David Robillard <d@drobilla.net>
+
+  Permission to use, copy, modify, and/or distribute this software for any
+  purpose with or without fee is hereby granted, provided that the above
+  copyright notice and this permission notice appear in all copies.
+
+  THIS SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+  WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+  MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+  ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+  WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+  ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+  OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+*/
+
+#undef NDEBUG
+
+#include "int_test_data.h"
+#include "num_test_utils.h"
+
+#include "exess/exess.h"
+
+#include <assert.h>
+#include <stdint.h>
+#include <stdio.h>
+#include <string.h>
+
+static void
+check_read(const char* const string,
+           const ExessStatus expected_status,
+           const int32_t     expected_value,
+           const size_t      expected_count)
+{
+  int32_t           value = 0;
+  const ExessResult r     = exess_read_int(&value, string);
+
+  assert(r.status == expected_status);
+  assert(r.count == expected_count);
+  assert(value == expected_value);
+}
+
+static void
+test_read_int(void)
+{
+  // Limits
+  check_read("-2147483648", EXESS_SUCCESS, INT32_MIN, EXESS_MAX_INT_LENGTH);
+  check_read("2147483647", EXESS_SUCCESS, INT32_MAX, 10);
+
+  // Out of range
+  check_read("-2147483649", EXESS_OUT_OF_RANGE, 0, 11);
+  check_read("2147483648", EXESS_OUT_OF_RANGE, 0, 10);
+  check_read("10000000000", EXESS_OUT_OF_RANGE, 0, 11);
+
+  // Garbage
+  check_read("+", EXESS_EXPECTED_DIGIT, 0, 1);
+}
+
+static void
+check_write(const int32_t     value,
+            const ExessStatus expected_status,
+            const size_t      buf_size,
+            const char* const expected_string)
+{
+  char buf[EXESS_MAX_INT_LENGTH + 1] = {1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12};
+
+  assert(buf_size <= sizeof(buf));
+
+  const ExessResult r = exess_write_int(value, buf_size, buf);
+  assert(!strcmp(buf, expected_string));
+  assert(r.status == expected_status);
+  assert(r.count == strlen(buf));
+  assert(r.status || exess_write_int(value, 0, NULL).count == r.count);
+}
+
+static void
+test_write_int(void)
+{
+  check_write(INT32_MIN, EXESS_SUCCESS, 12, "-2147483648");
+  check_write(INT32_MAX, EXESS_SUCCESS, 11, "2147483647");
+}
+
+static void
+test_round_trip(const ExessNumTestOptions opts)
+{
+  int32_t parsed_value               = 0;
+  char buf[EXESS_MAX_INT_LENGTH + 1] = {1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12};
+
+  if (opts.exhaustive) {
+    fprintf(stderr, "Testing xsd:int exhaustively\n");
+
+    for (int64_t i = INT32_MIN; i <= INT32_MAX; ++i) {
+      assert(!exess_write_int((int32_t)i, sizeof(buf), buf).status);
+      assert(!exess_read_int(&parsed_value, buf).status);
+      assert(parsed_value == i);
+
+      print_num_test_progress((uint64_t)(i - (int64_t)INT32_MIN), UINT32_MAX);
+    }
+  } else {
+    fprintf(stderr, "Testing xsd:int randomly with seed %u\n", opts.seed);
+    uint32_t rep = opts.seed;
+    for (uint64_t i = 0; i < opts.n_tests; ++i) {
+      rep = lcg32(rep);
+
+      const int32_t value = (int32_t)rep;
+
+      assert(!exess_write_int(value, sizeof(buf), buf).status);
+      assert(!exess_read_int(&parsed_value, buf).status);
+      assert(parsed_value == value);
+
+      print_num_test_progress(i, opts.n_tests);
+    }
+  }
+}
+
+int
+main(int argc, char** argv)
+{
+  const ExessNumTestOptions opts = parse_num_test_options(argc, argv);
+  if (opts.error) {
+    return 1;
+  }
+
+  test_read_int();
+  test_write_int();
+  test_round_trip(opts);
+
+  return 0;
+}
diff --git a/subprojects/exess/test/test_int_math.c b/subprojects/exess/test/test_int_math.c
new file mode 100644
index 00000000..8ab3f63f
--- /dev/null
+++ b/subprojects/exess/test/test_int_math.c
@@ -0,0 +1,97 @@
+/*
+  Copyright 2019-2021 David Robillard <d@drobilla.net>
+
+  Permission to use, copy, modify, and/or distribute this software for any
+  purpose with or without fee is hereby granted, provided that the above
+  copyright notice and this permission notice appear in all copies.
+
+  THIS SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+  WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+  MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+  ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+  WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+  ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+  OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+*/
+
+#undef NDEBUG
+
+#include "attributes.h"
+#include "int_math.h"
+
+#include <assert.h>
+#include <stdint.h>
+
+static void
+test_clz32(void)
+{
+  for (unsigned i = 0; i < 32; ++i) {
+    assert(exess_clz32(1u << i) == 32u - i - 1u);
+  }
+}
+
+static void
+test_clz64(void)
+{
+  for (unsigned i = 0; i < 64; ++i) {
+    assert(exess_clz64(1ull << i) == 64u - i - 1u);
+  }
+}
+
+static void
+test_ilog2(void)
+{
+  for (unsigned i = 0; i < 64; ++i) {
+    assert(exess_ilog2(1ull << i) == i);
+  }
+}
+
+static void
+test_ilog10(void)
+{
+  uint64_t power = 1;
+  for (unsigned i = 0; i < 20; ++i, power *= 10) {
+    assert(exess_ilog10(power) == i);
+  }
+}
+
+static void
+test_num_digits(void)
+{
+  assert(1 == exess_num_digits(0));
+  assert(1 == exess_num_digits(1));
+  assert(1 == exess_num_digits(9));
+  assert(2 == exess_num_digits(10));
+  assert(2 == exess_num_digits(99ull));
+  assert(3 == exess_num_digits(999ull));
+  assert(4 == exess_num_digits(9999ull));
+  assert(5 == exess_num_digits(99999ull));
+  assert(6 == exess_num_digits(999999ull));
+  assert(7 == exess_num_digits(9999999ull));
+  assert(8 == exess_num_digits(99999999ull));
+  assert(9 == exess_num_digits(999999999ull));
+  assert(10 == exess_num_digits(9999999999ull));
+  assert(11 == exess_num_digits(99999999999ull));
+  assert(12 == exess_num_digits(999999999999ull));
+  assert(13 == exess_num_digits(9999999999999ull));
+  assert(14 == exess_num_digits(99999999999999ull));
+  assert(15 == exess_num_digits(999999999999999ull));
+  assert(16 == exess_num_digits(9999999999999999ull));
+  assert(17 == exess_num_digits(99999999999999999ull));
+  assert(18 == exess_num_digits(999999999999999999ull));
+  assert(19 == exess_num_digits(9999999999999999999ull));
+  assert(20 == exess_num_digits(18446744073709551615ull));
+}
+
+EXESS_I_PURE_FUNC
+int
+main(void)
+{
+  test_clz32();
+  test_clz64();
+  test_ilog2();
+  test_ilog10();
+  test_num_digits();
+
+  return 0;
+}
diff --git a/subprojects/exess/test/test_long.c b/subprojects/exess/test/test_long.c
new file mode 100644
index 00000000..5631ca9f
--- /dev/null
+++ b/subprojects/exess/test/test_long.c
@@ -0,0 +1,144 @@
+/*
+  Copyright 2011-2021 David Robillard <d@drobilla.net>
+
+  Permission to use, copy, modify, and/or distribute this software for any
+  purpose with or without fee is hereby granted, provided that the above
+  copyright notice and this permission notice appear in all copies.
+
+  THIS SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+  WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+  MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+  ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+  WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+  ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+  OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+*/
+
+#undef NDEBUG
+
+#include "exess/exess.h"
+
+#include <assert.h>
+#include <stddef.h>
+#include <stdint.h>
+#include <string.h>
+
+static void
+check_read(const char* const string,
+           const ExessStatus expected_status,
+           const int64_t     expected_value,
+           const size_t      expected_count)
+{
+  int64_t value = 0;
+
+  const ExessResult r = exess_read_long(&value, string);
+  assert(r.status == expected_status);
+  assert(r.count == expected_count);
+  assert(value == expected_value);
+}
+
+static void
+test_read_long(void)
+{
+  // No input
+  check_read("", EXESS_EXPECTED_DIGIT, 0, 0);
+  check_read(" \f\n\r\t\v", EXESS_EXPECTED_DIGIT, 0, 6);
+
+  // Canonical form
+  check_read("-1", EXESS_SUCCESS, -1, 2);
+  check_read("0", EXESS_SUCCESS, 0, 1);
+  check_read("1", EXESS_SUCCESS, 1, 1);
+  check_read("1234", EXESS_SUCCESS, 1234, 4);
+  check_read("-1234", EXESS_SUCCESS, -1234, 5);
+
+  // Non-canonical form
+  check_read(" \f\n\r\t\v1234 ", EXESS_SUCCESS, 1234, 10);
+  check_read(" \f\n\r\t\v-1234 ", EXESS_SUCCESS, -1234, 11);
+  check_read(" \f\n\r\t\v+1234 ", EXESS_SUCCESS, 1234, 11);
+  check_read(" \f\n\r\t\v01234 ", EXESS_SUCCESS, 1234, 11);
+  check_read(" \f\n\r\t\v-01234 ", EXESS_SUCCESS, -1234, 12);
+  check_read("-01", EXESS_SUCCESS, -1, 3);
+  check_read("-0", EXESS_SUCCESS, 0, 2);
+  check_read("-00", EXESS_SUCCESS, 0, 3);
+  check_read("00", EXESS_SUCCESS, 0, 2);
+  check_read("+0", EXESS_SUCCESS, 0, 2);
+  check_read("+00", EXESS_SUCCESS, 0, 3);
+  check_read("+1", EXESS_SUCCESS, 1, 2);
+  check_read("+01", EXESS_SUCCESS, 1, 3);
+  check_read("+1234", EXESS_SUCCESS, 1234, 5);
+  check_read("01234", EXESS_SUCCESS, 1234, 5);
+  check_read("-01234", EXESS_SUCCESS, -1234, 6);
+
+  // Limits
+  check_read(
+    "-9223372036854775808", EXESS_SUCCESS, INT64_MIN, EXESS_MAX_LONG_LENGTH);
+  check_read("9223372036854775807", EXESS_SUCCESS, INT64_MAX, 19);
+
+  // Out of range
+  check_read("-9223372036854775809", EXESS_OUT_OF_RANGE, 0, 20);
+  check_read("9223372036854775808", EXESS_OUT_OF_RANGE, 0, 19);
+  check_read("12345678901234567890", EXESS_OUT_OF_RANGE, 0, 20);
+
+  // Trailing garbage
+  check_read("1234extra", EXESS_EXPECTED_END, 1234, 4);
+
+  // Garbage
+  check_read("+", EXESS_EXPECTED_DIGIT, 0, 1);
+  check_read("-", EXESS_EXPECTED_DIGIT, 0, 1);
+  check_read("true", EXESS_EXPECTED_DIGIT, 0, 0);
+  check_read("false", EXESS_EXPECTED_DIGIT, 0, 0);
+  check_read("zero", EXESS_EXPECTED_DIGIT, 0, 0);
+  check_read("NaN", EXESS_EXPECTED_DIGIT, 0, 0);
+  check_read("INF", EXESS_EXPECTED_DIGIT, 0, 0);
+  check_read("-INF", EXESS_EXPECTED_DIGIT, 0, 1);
+}
+
+static void
+check_write(const int64_t     value,
+            const ExessStatus expected_status,
+            const size_t      buf_size,
+            const char* const expected_string)
+{
+  char buf[EXESS_MAX_LONG_LENGTH + 1] = {
+    1,  2,  3,  4,  5,  6,  7,  8,  9,  10, //
+    11, 12, 13, 14, 15, 16, 17, 18, 19, 20, //
+    21};
+
+  assert(buf_size <= sizeof(buf));
+
+  const ExessResult r = exess_write_long(value, buf_size, buf);
+  assert(!strcmp(buf, expected_string));
+  assert(r.status == expected_status);
+  assert(r.count == strlen(buf));
+  assert(r.status || exess_write_long(value, 0, NULL).count == r.count);
+}
+
+static void
+test_write_long(void)
+{
+  check_write(-1, EXESS_SUCCESS, 3, "-1");
+  check_write(0, EXESS_SUCCESS, 2, "0");
+  check_write(1, EXESS_SUCCESS, 2, "1");
+  check_write(INT64_MIN, EXESS_SUCCESS, 21, "-9223372036854775808");
+  check_write(INT64_MAX, EXESS_SUCCESS, 20, "9223372036854775807");
+
+  check_write(INT64_MIN, EXESS_NO_SPACE, 20, "");
+  check_write(INT64_MAX, EXESS_NO_SPACE, 19, "");
+  check_write(1234, EXESS_NO_SPACE, 4, "");
+  check_write(-1234, EXESS_NO_SPACE, 5, "");
+
+  // Check that nothing is written when there isn't enough space
+  char              c = 42;
+  const ExessResult r = exess_write_long(1234, 0, &c);
+  assert(c == 42);
+  assert(r.status == EXESS_NO_SPACE);
+  assert(r.count == 0);
+}
+
+int
+main(void)
+{
+  test_read_long();
+  test_write_long();
+  return 0;
+}
diff --git a/subprojects/exess/test/test_short.c b/subprojects/exess/test/test_short.c
new file mode 100644
index 00000000..e0f3aa82
--- /dev/null
+++ b/subprojects/exess/test/test_short.c
@@ -0,0 +1,100 @@
+/*
+  Copyright 2011-2021 David Robillard <d@drobilla.net>
+
+  Permission to use, copy, modify, and/or distribute this software for any
+  purpose with or without fee is hereby granted, provided that the above
+  copyright notice and this permission notice appear in all copies.
+
+  THIS SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+  WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+  MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+  ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+  WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+  ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+  OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+*/
+
+#undef NDEBUG
+
+#include "exess/exess.h"
+
+#include <assert.h>
+#include <stddef.h>
+#include <stdint.h>
+#include <string.h>
+
+static void
+check_read(const char* const string,
+           const ExessStatus expected_status,
+           const int16_t     expected_value,
+           const size_t      expected_count)
+{
+  int16_t value = 0;
+
+  const ExessResult r = exess_read_short(&value, string);
+  assert(value == expected_value);
+  assert(r.status == expected_status);
+  assert(r.count == expected_count);
+}
+
+static void
+test_read_short(void)
+{
+  // Limits
+  check_read("-32768", EXESS_SUCCESS, INT16_MIN, EXESS_MAX_SHORT_LENGTH);
+  check_read("32767", EXESS_SUCCESS, INT16_MAX, 5);
+
+  // Out of range
+  check_read("-32769", EXESS_OUT_OF_RANGE, 0, 6);
+  check_read("32768", EXESS_OUT_OF_RANGE, 0, 5);
+
+  // Garbage
+  check_read("+", EXESS_EXPECTED_DIGIT, 0, 1);
+}
+
+static void
+check_write(const int16_t     value,
+            const ExessStatus expected_status,
+            const size_t      buf_size,
+            const char* const expected_string)
+{
+  char buf[EXESS_MAX_SHORT_LENGTH + 1] = {1, 2, 3, 4, 5, 6};
+
+  assert(buf_size <= sizeof(buf));
+
+  const ExessResult r = exess_write_short(value, buf_size, buf);
+  assert(!strcmp(buf, expected_string));
+  assert(r.status == expected_status);
+  assert(r.count == strlen(buf));
+  assert(r.status || exess_write_short(value, 0, NULL).count == r.count);
+}
+
+static void
+test_write_short(void)
+{
+  check_write(INT16_MIN, EXESS_SUCCESS, 7, "-32768");
+  check_write(INT16_MAX, EXESS_SUCCESS, 6, "32767");
+}
+
+static void
+test_round_trip(void)
+{
+  int16_t value                           = 0;
+  char    buf[EXESS_MAX_SHORT_LENGTH + 1] = {1, 2, 3, 4, 5, 6};
+
+  for (int32_t i = INT16_MIN; i <= INT16_MAX; ++i) {
+    assert(!exess_write_short((int16_t)i, sizeof(buf), buf).status);
+    assert(!exess_read_short(&value, buf).status);
+    assert(value == i);
+  }
+}
+
+int
+main(void)
+{
+  test_read_short();
+  test_write_short();
+  test_round_trip();
+
+  return 0;
+}
diff --git a/subprojects/exess/test/test_strerror.c b/subprojects/exess/test/test_strerror.c
new file mode 100644
index 00000000..1e5534ad
--- /dev/null
+++ b/subprojects/exess/test/test_strerror.c
@@ -0,0 +1,35 @@
+/*
+  Copyright 2011-2021 David Robillard <d@drobilla.net>
+
+  Permission to use, copy, modify, and/or distribute this software for any
+  purpose with or without fee is hereby granted, provided that the above
+  copyright notice and this permission notice appear in all copies.
+
+  THIS SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+  WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+  MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+  ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+  WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+  ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+  OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+*/
+
+#undef NDEBUG
+
+#include "exess/exess.h"
+
+#include <assert.h>
+#include <string.h>
+
+int
+main(void)
+{
+  for (ExessStatus i = EXESS_SUCCESS; i <= EXESS_UNSUPPORTED;
+       i             = (ExessStatus)(i + 1)) {
+    assert(strlen(exess_strerror(i)) > 0);
+  }
+
+  assert(!strcmp(exess_strerror((ExessStatus)9999), "Unknown error"));
+
+  return 0;
+}
diff --git a/subprojects/exess/test/test_time.c b/subprojects/exess/test/test_time.c
new file mode 100644
index 00000000..6cf20f6d
--- /dev/null
+++ b/subprojects/exess/test/test_time.c
@@ -0,0 +1,222 @@
+/*
+  Copyright 2011-2021 David Robillard <d@drobilla.net>
+
+  Permission to use, copy, modify, and/or distribute this software for any
+  purpose with or without fee is hereby granted, provided that the above
+  copyright notice and this permission notice appear in all copies.
+
+  THIS SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+  WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+  MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+  ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+  WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+  ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+  OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+*/
+
+#undef NDEBUG
+
+#include "float_test_data.h"
+#include "int_test_data.h"
+#include "time_test_utils.h"
+
+#include "exess/exess.h"
+
+#include <assert.h>
+#include <stdbool.h>
+#include <stdint.h>
+#include <string.h>
+
+static const ExessTime nozone   = {{EXESS_LOCAL}, 0, 0, 0, 0};
+static const ExessTime utc      = {{0}, 12, 15, 1, 250000000};
+static const ExessTime zoned    = {INIT_ZONE(11, 30), 23, 59, 59, 1000000};
+static const ExessTime high     = {INIT_ZONE(11, 30), 24, 0, 0, 0};
+static const ExessTime garbage1 = {INIT_ZONE(11, 30), 0, 0, 0, 1000000000};
+static const ExessTime garbage2 = {INIT_ZONE(11, 30), 0, 0, 60, 0};
+static const ExessTime garbage3 = {INIT_ZONE(11, 30), 0, 60, 0, 0};
+static const ExessTime garbage4 = {INIT_ZONE(11, 30), 24, 0, 0, 1};
+static const ExessTime garbage5 = {INIT_ZONE(11, 30), 24, 0, 1, 0};
+static const ExessTime garbage6 = {INIT_ZONE(11, 30), 24, 1, 0, 0};
+static const ExessTime garbage7 = {INIT_ZONE(11, 30), 25, 0, 0, 0};
+
+static void
+check_read(const char* const string,
+           const ExessStatus expected_status,
+           const uint8_t     expected_hour,
+           const uint8_t     expected_minute,
+           const uint8_t     expected_second,
+           const uint32_t    expected_nanosecond,
+           const int8_t      expected_tz_hour,
+           const int8_t      expected_tz_minute,
+           const bool        expected_tz_is_present,
+           const size_t      expected_count)
+{
+  ExessTime value = {{0}, 0, 0, 0, 0};
+
+  const ExessResult r = exess_read_time(&value, string);
+  assert(r.status == expected_status);
+  assert(r.count == expected_count);
+  assert(value.hour == expected_hour);
+  assert(value.minute == expected_minute);
+  assert(value.second == expected_second);
+  assert(value.nanosecond == expected_nanosecond);
+  assert((!expected_tz_is_present && value.zone.quarter_hours == EXESS_LOCAL) ||
+         value.zone.quarter_hours ==
+           4 * expected_tz_hour + expected_tz_minute / 15);
+}
+
+static void
+test_read_time(void)
+{
+  // No value
+  check_read("", EXESS_EXPECTED_DIGIT, 0, 0, 0, 0, 0, 0, false, 0);
+  check_read(" \f\n\r\t\v", EXESS_EXPECTED_DIGIT, 0, 0, 0, 0, 0, 0, false, 6);
+
+  // Good values
+  check_read("13:20:00", EXESS_SUCCESS, 13, 20, 0, 0, 0, 0, false, 8);
+  check_read(
+    "13:20:30.5555", EXESS_SUCCESS, 13, 20, 30, 555500000, 0, 0, false, 13);
+  check_read("13:20:00-05:00", EXESS_SUCCESS, 13, 20, 0, 0, -5, 0, true, 14);
+  check_read("13:20:00Z", EXESS_SUCCESS, 13, 20, 0, 0, 0, 0, true, 9);
+  check_read("00:00:00", EXESS_SUCCESS, 0, 0, 0, 0, 0, 0, false, 8);
+  check_read("24:00:00", EXESS_SUCCESS, 24, 0, 0, 0, 0, 0, false, 8);
+  check_read("21:32:52", EXESS_SUCCESS, 21, 32, 52, 0, 0, 0, false, 8);
+  check_read("21:32:52+02:00", EXESS_SUCCESS, 21, 32, 52, 0, 2, 0, true, 14);
+  check_read("19:32:52Z", EXESS_SUCCESS, 19, 32, 52, 0, 0, 0, true, 9);
+  check_read("19:32:52+00:00", EXESS_SUCCESS, 19, 32, 52, 0, 0, 0, true, 14);
+  check_read(
+    "21:32:52.12679", EXESS_SUCCESS, 21, 32, 52, 126790000, 0, 0, false, 14);
+
+  // Longest possible string
+  check_read("24:59:59.000000001-14:00",
+             EXESS_SUCCESS,
+             24,
+             59,
+             59,
+             1,
+             -14,
+             0,
+             true,
+             EXESS_MAX_TIME_LENGTH);
+
+  // Non-canonical form
+  check_read(
+    " \f\n\r\t\v13:20:00 ", EXESS_SUCCESS, 13, 20, 0, 0, 0, 0, false, 14);
+
+  // Trailing garbage
+  check_read("13:20:00junk", EXESS_EXPECTED_SIGN, 13, 20, 0, 0, 0, 0, false, 8);
+  check_read("13:20:00Zjunk", EXESS_EXPECTED_END, 13, 20, 0, 0, 0, 0, true, 9);
+
+  // Garbage
+  check_read("13.20.00", EXESS_EXPECTED_COLON, 13, 0, 0, 0, 0, 0, false, 2);
+  check_read("13:20:", EXESS_EXPECTED_DIGIT, 13, 20, 0, 0, 0, 0, false, 6);
+  check_read("5:20:00", EXESS_EXPECTED_DIGIT, 5, 0, 0, 0, 0, 0, false, 1);
+  check_read("13:20", EXESS_EXPECTED_COLON, 13, 20, 0, 0, 0, 0, false, 5);
+  check_read("13:20.5:00", EXESS_EXPECTED_COLON, 13, 20, 0, 0, 0, 0, false, 5);
+  check_read("13:65:00", EXESS_OUT_OF_RANGE, 13, 65, 0, 0, 0, 0, false, 5);
+  check_read("21:32", EXESS_EXPECTED_COLON, 21, 32, 0, 0, 0, 0, false, 5);
+  check_read("25:25:10", EXESS_OUT_OF_RANGE, 25, 0, 0, 0, 0, 0, false, 2);
+  check_read("-10:00:00", EXESS_EXPECTED_DIGIT, 0, 0, 0, 0, 0, 0, false, 0);
+  check_read("1:20:10", EXESS_EXPECTED_DIGIT, 1, 0, 0, 0, 0, 0, false, 1);
+  check_read("13:20:00A", EXESS_EXPECTED_SIGN, 13, 20, 0, 0, 0, 0, false, 8);
+}
+
+static void
+check_write(const ExessTime   value,
+            const ExessStatus expected_status,
+            const size_t      buf_size,
+            const char* const expected_string)
+{
+  char buf[EXESS_MAX_TIME_LENGTH + 1] = {1,  2,  3,  4,  5,  6,  7,  8,  9,
+                                         10, 11, 12, 13, 14, 15, 16, 17, 18,
+                                         19, 20, 21, 22, 23, 24, 25};
+
+  assert(buf_size <= sizeof(buf));
+
+  const ExessResult r = exess_write_time(value, buf_size, buf);
+  assert(r.status == expected_status);
+  assert(r.count == strlen(buf));
+  assert(!strcmp(buf, expected_string));
+  assert(r.status || exess_write_time(value, 0, NULL).count == r.count);
+}
+
+static void
+test_write_time(void)
+{
+  check_write(nozone, EXESS_SUCCESS, 9, "00:00:00");
+  check_write(utc, EXESS_SUCCESS, 13, "12:15:01.25Z");
+  check_write(zoned, EXESS_SUCCESS, 19, "23:59:59.001+11:30");
+  check_write(high, EXESS_SUCCESS, 15, "24:00:00+11:30");
+
+  check_write(garbage1, EXESS_BAD_VALUE, 19, "");
+  check_write(garbage2, EXESS_BAD_VALUE, 19, "");
+  check_write(garbage3, EXESS_BAD_VALUE, 19, "");
+  check_write(garbage4, EXESS_BAD_VALUE, 19, "");
+  check_write(garbage5, EXESS_BAD_VALUE, 19, "");
+  check_write(garbage6, EXESS_BAD_VALUE, 19, "");
+  check_write(garbage7, EXESS_BAD_VALUE, 19, "");
+
+  check_write(nozone, EXESS_NO_SPACE, 8, "");
+  check_write(utc, EXESS_NO_SPACE, 12, "");
+  check_write(zoned, EXESS_NO_SPACE, 18, "");
+  check_write(zoned, EXESS_NO_SPACE, 12, "");
+  check_write(high, EXESS_NO_SPACE, 14, "");
+
+  // Check that nothing is written when there isn't enough space
+  char              c = 42;
+  const ExessResult r = exess_write_time(nozone, 0, &c);
+  assert(c == 42);
+  assert(r.status == EXESS_NO_SPACE);
+  assert(r.count == 0);
+}
+
+static void
+check_round_trip(const ExessTime value)
+{
+  ExessTime parsed_value                   = {{0}, 0, 0, 0, 0};
+  char      buf[EXESS_MAX_TIME_LENGTH + 1] = {0};
+
+  assert(!exess_write_time(value, sizeof(buf), buf).status);
+  assert(!exess_read_time(&parsed_value, buf).status);
+  assert(parsed_value.hour == value.hour);
+  assert(parsed_value.minute == value.minute);
+  assert(double_matches(parsed_value.second, value.second));
+  assert(parsed_value.zone.quarter_hours == value.zone.quarter_hours);
+}
+
+static void
+test_round_trip(void)
+{
+  uint32_t rng = 0;
+  for (uint8_t h = 0; h < 24; ++h) {
+    for (uint8_t m = 0; m < 60; ++m) {
+      rng = lcg32(rng);
+
+      const uint32_t ns = rng % 1000000000u;
+
+      rng = lcg32(rng);
+
+      const uint8_t   s            = (uint8_t)(rng % 60u);
+      const ExessTime no_zone      = {{EXESS_LOCAL}, h, m, s, ns};
+      const ExessTime lowest_zone  = {INIT_ZONE(-14, 0), h, m, s, ns};
+      const ExessTime highest_zone = {INIT_ZONE(14, 0), h, m, s, ns};
+
+      check_round_trip(no_zone);
+      check_round_trip(lowest_zone);
+      check_round_trip(highest_zone);
+
+      const ExessTime value = {random_timezone(&rng), h, m, s, ns};
+      check_round_trip(value);
+    }
+  }
+}
+
+int
+main(void)
+{
+  test_read_time();
+  test_write_time();
+  test_round_trip();
+
+  return 0;
+}
diff --git a/subprojects/exess/test/test_timezone.c b/subprojects/exess/test/test_timezone.c
new file mode 100644
index 00000000..8ccfb12b
--- /dev/null
+++ b/subprojects/exess/test/test_timezone.c
@@ -0,0 +1,178 @@
+/*
+  Copyright 2011-2021 David Robillard <d@drobilla.net>
+
+  Permission to use, copy, modify, and/or distribute this software for any
+  purpose with or without fee is hereby granted, provided that the above
+  copyright notice and this permission notice appear in all copies.
+
+  THIS SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+  WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+  MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+  ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+  WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+  ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+  OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+*/
+
+#undef NDEBUG
+
+#include "time_test_utils.h"
+
+#include "exess/exess.h"
+
+#include <assert.h>
+#include <stdbool.h>
+#include <stdint.h>
+#include <string.h>
+
+#include <stdio.h>
+
+static const ExessTimezone missing  = {EXESS_LOCAL};
+static const ExessTimezone utc      = INIT_ZONE(0, 0);
+static const ExessTimezone plus     = INIT_ZONE(11, 30);
+static const ExessTimezone minus    = INIT_ZONE(-11, -30);
+static const ExessTimezone slight   = INIT_ZONE(0, 30);
+static const ExessTimezone lowest   = INIT_ZONE(-14, 0);
+static const ExessTimezone highest  = INIT_ZONE(14, 0);
+static const ExessTimezone garbage1 = INIT_ZONE(-14, -15);
+static const ExessTimezone garbage2 = INIT_ZONE(14, 15);
+static const ExessTimezone garbage3 = INIT_ZONE(-15, 0);
+static const ExessTimezone garbage4 = INIT_ZONE(15, 0);
+
+static void
+check_read(const char* const string,
+           const ExessStatus expected_status,
+           const int8_t      expected_hour,
+           const int8_t      expected_minute,
+           const bool        expected_is_present,
+           const size_t      expected_count)
+{
+  // The timezone interface is not public, so we test it via time
+  char time_string[] = "12:00:00XXXXXX";
+  strncpy(time_string + 8, string, sizeof(time_string) - 9);
+
+  ExessTime         value = {{0}, 0, 0, 0, 0};
+  const ExessResult r     = exess_read_time(&value, time_string);
+
+  assert(r.status == expected_status);
+  assert(r.count == 8 + expected_count);
+  assert((!expected_is_present && value.zone.quarter_hours == EXESS_LOCAL) ||
+         value.zone.quarter_hours == 4 * expected_hour + expected_minute / 15);
+}
+
+static void
+test_read_timezone(void)
+{
+  // Basic values
+  check_read("Z", EXESS_SUCCESS, 0, 0, true, 1);
+  check_read("-05:00", EXESS_SUCCESS, -5, 0, true, 6);
+  check_read("+02:00", EXESS_SUCCESS, 2, 0, true, 6);
+  check_read("+00:00", EXESS_SUCCESS, 0, 0, true, 6);
+  check_read("-00:00", EXESS_SUCCESS, 0, 0, true, 6);
+
+  // Limits
+  check_read("-14:00", EXESS_SUCCESS, -14, 0, true, 6);
+  check_read("+14:00", EXESS_SUCCESS, 14, 0, true, 6);
+  check_read("-13:45", EXESS_SUCCESS, -13, -45, true, 6);
+  check_read("+13:45", EXESS_SUCCESS, 13, 45, true, 6);
+
+  // Out of range
+  check_read("-14:15", EXESS_OUT_OF_RANGE, 0, 0, false, 6);
+  check_read("+14:15", EXESS_OUT_OF_RANGE, 0, 0, false, 6);
+  check_read("-15:00", EXESS_OUT_OF_RANGE, 0, 0, false, 3);
+  check_read("+15:00", EXESS_OUT_OF_RANGE, 0, 0, false, 3);
+  check_read("-13:60", EXESS_OUT_OF_RANGE, 0, 0, false, 6);
+  check_read("+13:60", EXESS_OUT_OF_RANGE, 0, 0, false, 6);
+
+  // Garbage
+  check_read("+05:01", EXESS_UNSUPPORTED, 0, 0, false, 6);
+  check_read("05:00", EXESS_EXPECTED_SIGN, 0, 0, false, 0);
+  check_read("+5:00", EXESS_EXPECTED_DIGIT, 0, 0, false, 2);
+  check_read("+5:0", EXESS_EXPECTED_DIGIT, 0, 0, false, 2);
+  check_read("+5:", EXESS_EXPECTED_DIGIT, 0, 0, false, 2);
+  check_read("+:0", EXESS_EXPECTED_DIGIT, 0, 0, false, 1);
+  check_read("+A5:00", EXESS_EXPECTED_DIGIT, 0, 0, false, 1);
+  check_read("+0A:00", EXESS_EXPECTED_DIGIT, 0, 0, false, 2);
+  check_read("+05A00", EXESS_EXPECTED_COLON, 0, 0, false, 3);
+  check_read("+05:A0", EXESS_EXPECTED_DIGIT, 0, 0, false, 4);
+  check_read("+05:0A", EXESS_EXPECTED_DIGIT, 0, 0, false, 5);
+}
+
+static void
+check_write(const ExessTimezone value,
+            const ExessStatus   expected_status,
+            const size_t        buf_size,
+            const char* const   expected_string)
+{
+  // The timezone interface is not public, so we test it via time
+  char buf[EXESS_MAX_TIME_LENGTH + 1] = {1,  2,  3,  4,  5,  6,  7,  8,  9,
+                                         10, 11, 12, 13, 14, 15, 16, 17, 18,
+                                         19, 20, 21, 22, 23, 24, 25};
+
+  assert(buf_size <= sizeof(buf) - 8);
+
+  const ExessTime   time = {value, 12, 0, 0, 0};
+  const ExessResult r    = exess_write_time(time, 8 + buf_size, buf);
+  assert(r.status == expected_status);
+  assert(r.count == strlen(buf));
+  assert((!buf[0] && !expected_string[0]) || !strcmp(buf + 8, expected_string));
+  assert(r.status || exess_write_time(time, 0, NULL).count == r.count);
+}
+
+static void
+test_write_timezone(void)
+{
+  check_write(missing, EXESS_SUCCESS, 1, "");
+  check_write(utc, EXESS_SUCCESS, 2, "Z");
+  check_write(plus, EXESS_SUCCESS, 7, "+11:30");
+  check_write(minus, EXESS_SUCCESS, 7, "-11:30");
+  check_write(slight, EXESS_SUCCESS, 7, "+00:30");
+  check_write(lowest, EXESS_SUCCESS, 7, "-14:00");
+  check_write(highest, EXESS_SUCCESS, 7, "+14:00");
+
+  check_write(garbage1, EXESS_BAD_VALUE, 7, "");
+  check_write(garbage2, EXESS_BAD_VALUE, 7, "");
+  check_write(garbage3, EXESS_BAD_VALUE, 7, "");
+  check_write(garbage4, EXESS_BAD_VALUE, 7, "");
+
+  check_write(utc, EXESS_NO_SPACE, 1, "");
+  check_write(plus, EXESS_NO_SPACE, 6, "");
+}
+
+static void
+check_round_trip(const ExessTimezone value)
+{
+  ExessTime parsed_time                    = {{0}, 0, 0, 0, 0};
+  char      buf[EXESS_MAX_TIME_LENGTH + 1] = {42};
+
+  const ExessTime time = {value, 12, 0, 0, 0};
+
+  assert(!exess_write_time(time, sizeof(buf), buf).status);
+  assert(!exess_read_time(&parsed_time, buf).status);
+  assert(!memcmp(&parsed_time.zone, &value, sizeof(ExessTimezone)));
+}
+
+static void
+test_round_trip(void)
+{
+  check_round_trip(lowest);
+  check_round_trip(highest);
+
+  for (int8_t h = -13; h < 13; ++h) {
+    for (int8_t q = 0; q < 4; ++q) {
+      const ExessTimezone value = {(int8_t)(4 * h + q)};
+
+      check_round_trip(value);
+    }
+  }
+}
+
+int
+main(void)
+{
+  test_read_timezone();
+  test_write_timezone();
+  test_round_trip();
+
+  return 0;
+}
diff --git a/subprojects/exess/test/test_ubyte.c b/subprojects/exess/test/test_ubyte.c
new file mode 100644
index 00000000..7129c85d
--- /dev/null
+++ b/subprojects/exess/test/test_ubyte.c
@@ -0,0 +1,100 @@
+/*
+  Copyright 2011-2021 David Robillard <d@drobilla.net>
+
+  Permission to use, copy, modify, and/or distribute this software for any
+  purpose with or without fee is hereby granted, provided that the above
+  copyright notice and this permission notice appear in all copies.
+
+  THIS SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+  WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+  MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+  ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+  WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+  ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+  OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+*/
+
+#undef NDEBUG
+
+#include "exess/exess.h"
+
+#include <assert.h>
+#include <stddef.h>
+#include <stdint.h>
+#include <string.h>
+
+static void
+check_read(const char* const string,
+           const ExessStatus expected_status,
+           const uint8_t     expected_value,
+           const size_t      expected_count)
+{
+  uint8_t value = 0;
+
+  const ExessResult r = exess_read_ubyte(&value, string);
+  assert(r.status == expected_status);
+  assert(r.count == expected_count);
+  assert(value == expected_value);
+}
+
+static void
+test_read_ubyte(void)
+{
+  // Limits
+  check_read("0", EXESS_SUCCESS, 0, 1);
+  check_read("255", EXESS_SUCCESS, UINT8_MAX, EXESS_MAX_UBYTE_LENGTH);
+
+  // Out of range
+  check_read("256", EXESS_OUT_OF_RANGE, 0, 3);
+
+  // Garbage
+  check_read("-1", EXESS_EXPECTED_DIGIT, 0, 0);
+  check_read("+", EXESS_EXPECTED_DIGIT, 0, 0);
+}
+
+static void
+check_write(const uint8_t     value,
+            const ExessStatus expected_status,
+            const size_t      buf_size,
+            const char* const expected_string)
+{
+  char buf[EXESS_MAX_UBYTE_LENGTH + 1] = {1, 2, 3, 4};
+
+  assert(buf_size <= sizeof(buf));
+
+  const ExessResult r = exess_write_ubyte(value, buf_size, buf);
+  assert(r.status == expected_status);
+  assert(r.count == strlen(buf));
+  assert(!strcmp(buf, expected_string));
+  assert(r.status || exess_write_ubyte(value, 0, NULL).count == r.count);
+}
+
+static void
+test_write_ubyte(void)
+{
+  check_write(0, EXESS_SUCCESS, 2, "0");
+  check_write(UINT8_MAX, EXESS_SUCCESS, 4, "255");
+}
+
+static void
+test_round_trip(void)
+{
+  uint8_t value                           = 0;
+  char    buf[EXESS_MAX_UBYTE_LENGTH + 1] = {1, 2, 3, 4};
+
+  for (uint16_t i = 0; i <= UINT8_MAX; ++i) {
+    assert(!exess_write_ubyte((uint8_t)i, sizeof(buf), buf).status);
+    assert(!exess_read_ubyte(&value, buf).status);
+    assert(value == i);
+  }
+}
+
+int
+main(void)
+{
+  test_read_ubyte();
+  test_write_ubyte();
+  test_round_trip();
+
+  return 0;
+}
diff --git a/subprojects/exess/test/test_uint.c b/subprojects/exess/test/test_uint.c
new file mode 100644
index 00000000..5889fc20
--- /dev/null
+++ b/subprojects/exess/test/test_uint.c
@@ -0,0 +1,128 @@
+/*
+  Copyright 2011-2021 David Robillard <d@drobilla.net>
+
+  Permission to use, copy, modify, and/or distribute this software for any
+  purpose with or without fee is hereby granted, provided that the above
+  copyright notice and this permission notice appear in all copies.
+
+  THIS SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+  WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+  MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+  ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+  WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+  ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+  OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+*/
+
+#undef NDEBUG
+
+#include "int_test_data.h"
+#include "num_test_utils.h"
+
+#include "exess/exess.h"
+
+#include <assert.h>
+#include <stdint.h>
+#include <stdio.h>
+#include <string.h>
+
+static void
+check_read(const char* const string,
+           const ExessStatus expected_status,
+           const uint32_t    expected_value,
+           const size_t      expected_count)
+{
+  uint32_t          value = 0;
+  const ExessResult r     = exess_read_uint(&value, string);
+
+  assert(r.status == expected_status);
+  assert(r.count == expected_count);
+  assert(value == expected_value);
+}
+
+static void
+test_read_uint(void)
+{
+  // Limits
+  check_read("0", EXESS_SUCCESS, 0, 1);
+  check_read("4294967295", EXESS_SUCCESS, UINT32_MAX, EXESS_MAX_UINT_LENGTH);
+
+  // Out of range
+  check_read("4294967296", EXESS_OUT_OF_RANGE, 0, 10);
+
+  // Garbage
+  check_read("-1", EXESS_EXPECTED_DIGIT, 0, 0);
+  check_read("+", EXESS_EXPECTED_DIGIT, 0, 0);
+}
+
+static void
+check_write(const uint32_t    value,
+            const ExessStatus expected_status,
+            const size_t      buf_size,
+            const char* const expected_string)
+{
+  char buf[EXESS_MAX_UINT_LENGTH + 1] = {1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11};
+
+  assert(buf_size <= sizeof(buf));
+
+  const ExessResult r = exess_write_uint(value, buf_size, buf);
+  assert(r.status == expected_status);
+  assert(r.count == strlen(buf));
+  assert(!strcmp(buf, expected_string));
+  assert(r.status || exess_write_uint(value, 0, NULL).count == r.count);
+}
+
+static void
+test_write_uint(void)
+{
+  check_write(0u, EXESS_SUCCESS, 2, "0");
+  check_write(UINT32_MAX, EXESS_SUCCESS, 11, "4294967295");
+}
+
+static void
+test_round_trip(const ExessNumTestOptions opts)
+{
+  uint32_t parsed_value                   = 0;
+  char     buf[EXESS_MAX_UINT_LENGTH + 1] = {1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11};
+
+  if (opts.exhaustive) {
+    fprintf(stderr, "Testing xsd:unsignedInt exhaustively\n");
+
+    for (uint64_t i = 0u; i <= UINT32_MAX; ++i) {
+      assert(!exess_write_uint((uint32_t)i, sizeof(buf), buf).status);
+      assert(!exess_read_uint(&parsed_value, buf).status);
+      assert(parsed_value == i);
+
+      print_num_test_progress(i, UINT32_MAX);
+    }
+  } else {
+    fprintf(
+      stderr, "Testing xsd:unsignedInt randomly with seed %u\n", opts.seed);
+
+    uint32_t value = opts.seed;
+    for (uint64_t i = 0; i < opts.n_tests; ++i) {
+      value = lcg32(value);
+
+      assert(!exess_write_uint(value, sizeof(buf), buf).status);
+      assert(!exess_read_uint(&parsed_value, buf).status);
+      assert(parsed_value == value);
+
+      print_num_test_progress(i, opts.n_tests);
+    }
+  }
+}
+
+int
+main(int argc, char** argv)
+{
+  const ExessNumTestOptions opts = parse_num_test_options(argc, argv);
+  if (opts.error) {
+    return 1;
+  }
+
+  test_read_uint();
+  test_write_uint();
+  test_round_trip(opts);
+
+  return 0;
+}
diff --git a/subprojects/exess/test/test_ulong.c b/subprojects/exess/test/test_ulong.c
new file mode 100644
index 00000000..ba8ba05f
--- /dev/null
+++ b/subprojects/exess/test/test_ulong.c
@@ -0,0 +1,126 @@
+/*
+  Copyright 2011-2021 David Robillard <d@drobilla.net>
+
+  Permission to use, copy, modify, and/or distribute this software for any
+  purpose with or without fee is hereby granted, provided that the above
+  copyright notice and this permission notice appear in all copies.
+
+  THIS SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+  WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+  MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+  ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+  WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+  ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+  OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+*/
+
+#undef NDEBUG
+
+#include "exess/exess.h"
+
+#include <assert.h>
+#include <stddef.h>
+#include <stdint.h>
+#include <string.h>
+
+static void
+check_read(const char* const string,
+           const ExessStatus expected_status,
+           const uint64_t    expected_value,
+           const size_t      expected_count)
+{
+  uint64_t          value = 0;
+  const ExessResult r     = exess_read_ulong(&value, string);
+
+  assert(r.status == expected_status);
+  assert(r.count == expected_count);
+  assert(value == expected_value);
+}
+
+static void
+test_read_ulong(void)
+{
+  // No input
+  check_read("", EXESS_EXPECTED_DIGIT, 0, 0);
+  check_read(" \f\n\r\t\v", EXESS_EXPECTED_DIGIT, 0, 6);
+
+  // Canonical form
+  check_read("0", EXESS_SUCCESS, 0, 1);
+  check_read("1234", EXESS_SUCCESS, 1234, 4);
+
+  // Non-canonical form
+  check_read(" \f\n\r\t\v1234 ", EXESS_SUCCESS, 1234, 10);
+  check_read(" \f\n\r\t\v01234 ", EXESS_SUCCESS, 1234, 11);
+  check_read("01234", EXESS_SUCCESS, 1234, 5);
+  check_read("00", EXESS_SUCCESS, 0, 2);
+
+  // Limits
+  check_read("0", EXESS_SUCCESS, 0, 1);
+  check_read(
+    "18446744073709551615", EXESS_SUCCESS, UINT64_MAX, EXESS_MAX_ULONG_LENGTH);
+
+  // Out of range
+  check_read("18446744073709551616", EXESS_OUT_OF_RANGE, 0, 19);
+
+  // Trailing garbage
+  check_read("1234extra", EXESS_EXPECTED_END, 1234, 4);
+
+  // Garbage
+  check_read(" \f\n\r\t\v+1234 ", EXESS_EXPECTED_DIGIT, 0, 6);
+  check_read("+1234", EXESS_EXPECTED_DIGIT, 0, 0);
+  check_read("+0", EXESS_EXPECTED_DIGIT, 0, 0);
+  check_read("+", EXESS_EXPECTED_DIGIT, 0, 0);
+  check_read("-", EXESS_EXPECTED_DIGIT, 0, 0);
+  check_read("true", EXESS_EXPECTED_DIGIT, 0, 0);
+  check_read("false", EXESS_EXPECTED_DIGIT, 0, 0);
+  check_read("zero", EXESS_EXPECTED_DIGIT, 0, 0);
+  check_read("NaN", EXESS_EXPECTED_DIGIT, 0, 0);
+  check_read("INF", EXESS_EXPECTED_DIGIT, 0, 0);
+  check_read("-INF", EXESS_EXPECTED_DIGIT, 0, 0);
+}
+
+static void
+check_write(const uint64_t    value,
+            const ExessStatus expected_status,
+            const size_t      buf_size,
+            const char* const expected_string)
+{
+  char buf[EXESS_MAX_ULONG_LENGTH + 1] = {
+    1,  2,  3,  4,  5,  6,  7,  8,  9,  10, //
+    11, 12, 13, 14, 15, 16, 17, 18, 19, 20, //
+    21};
+
+  assert(buf_size <= sizeof(buf));
+
+  const ExessResult r = exess_write_ulong(value, buf_size, buf);
+
+  assert(r.status == expected_status);
+  assert(r.count == strlen(buf));
+  assert(!strcmp(buf, expected_string));
+  assert(r.status || exess_write_ulong(value, 0, NULL).count == r.count);
+}
+
+static void
+test_write_ulong(void)
+{
+  check_write(0u, EXESS_SUCCESS, 2, "0");
+  check_write(1u, EXESS_SUCCESS, 2, "1");
+  check_write(UINT64_MAX, EXESS_SUCCESS, 21, "18446744073709551615");
+
+  check_write(1234u, EXESS_NO_SPACE, 4, "");
+
+  // Check that nothing is written when there isn't enough space
+  char              c = 42;
+  const ExessResult r = exess_write_ulong(1234u, 0, &c);
+  assert(c == 42);
+  assert(r.status == EXESS_NO_SPACE);
+  assert(r.count == 0);
+}
+
+int
+main(void)
+{
+  test_read_ulong();
+  test_write_ulong();
+  return 0;
+}
diff --git a/subprojects/exess/test/test_ushort.c b/subprojects/exess/test/test_ushort.c
new file mode 100644
index 00000000..056e182a
--- /dev/null
+++ b/subprojects/exess/test/test_ushort.c
@@ -0,0 +1,100 @@
+/*
+  Copyright 2011-2021 David Robillard <d@drobilla.net>
+
+  Permission to use, copy, modify, and/or distribute this software for any
+  purpose with or without fee is hereby granted, provided that the above
+  copyright notice and this permission notice appear in all copies.
+
+  THIS SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+  WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+  MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+  ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+  WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+  ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+  OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+*/
+
+#undef NDEBUG
+
+#include "exess/exess.h"
+
+#include <assert.h>
+#include <stddef.h>
+#include <stdint.h>
+#include <string.h>
+
+static void
+check_read(const char* const string,
+           const ExessStatus expected_status,
+           const uint16_t    expected_value,
+           const size_t      expected_count)
+{
+  uint16_t value = 0;
+
+  const ExessResult r = exess_read_ushort(&value, string);
+  assert(value == expected_value);
+  assert(r.status == expected_status);
+  assert(r.count == expected_count);
+}
+
+static void
+test_read_ushort(void)
+{
+  // Limits
+  check_read("0", EXESS_SUCCESS, 0, 1);
+  check_read("65535", EXESS_SUCCESS, UINT16_MAX, EXESS_MAX_USHORT_LENGTH);
+
+  // Out of range
+  check_read("65536", EXESS_OUT_OF_RANGE, 0, 5);
+
+  // Garbage
+  check_read("-1", EXESS_EXPECTED_DIGIT, 0, 0);
+  check_read("+", EXESS_EXPECTED_DIGIT, 0, 0);
+}
+
+static void
+check_write(const uint16_t    value,
+            const ExessStatus expected_status,
+            const size_t      buf_size,
+            const char* const expected_string)
+{
+  char buf[EXESS_MAX_USHORT_LENGTH + 1] = {1, 2, 3, 4, 5, 6};
+
+  assert(buf_size <= sizeof(buf));
+
+  const ExessResult r = exess_write_ushort(value, buf_size, buf);
+  assert(!strcmp(buf, expected_string));
+  assert(r.status == expected_status);
+  assert(r.count == strlen(buf));
+  assert(r.status || exess_write_ushort(value, 0, NULL).count == r.count);
+}
+
+static void
+test_write_ushort(void)
+{
+  check_write(0u, EXESS_SUCCESS, 2, "0");
+  check_write(UINT16_MAX, EXESS_SUCCESS, 6, "65535");
+}
+
+static void
+test_round_trip(void)
+{
+  uint16_t value                            = 0;
+  char     buf[EXESS_MAX_USHORT_LENGTH + 1] = {1, 2, 3, 4, 5, 6};
+
+  for (uint32_t i = 0; i <= UINT16_MAX; ++i) {
+    assert(!exess_write_ushort((uint16_t)i, sizeof(buf), buf).status);
+    assert(!exess_read_ushort(&value, buf).status);
+    assert(value == i);
+  }
+}
+
+int
+main(void)
+{
+  test_read_ushort();
+  test_write_ushort();
+  test_round_trip();
+
+  return 0;
+}
diff --git a/subprojects/exess/test/test_variant.c b/subprojects/exess/test/test_variant.c
new file mode 100644
index 00000000..7ef7803d
--- /dev/null
+++ b/subprojects/exess/test/test_variant.c
@@ -0,0 +1,301 @@
+/*
+  Copyright 2011-2021 David Robillard <d@drobilla.net>
+
+  Permission to use, copy, modify, and/or distribute this software for any
+  purpose with or without fee is hereby granted, provided that the above
+  copyright notice and this permission notice appear in all copies.
+
+  THIS SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+  WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+  MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+  ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+  WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+  ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+  OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+*/
+
+#undef NDEBUG
+
+#include "exess/exess.h"
+
+#include <assert.h>
+#include <stdbool.h>
+#include <stddef.h>
+#include <string.h>
+
+#define CHECK_POINTEE_EQUALS(p, v) assert((p) && (*(p) == (v)))
+
+static const ExessDuration duration = {14,
+                                       3 * 24 * 60 * 60 + 4 * 60 * 60 + 5 * 60 +
+                                         6,
+                                       0};
+
+static const ExessDateTime datetime = {2001, 2, 3, false, 4, 5, 6, 0};
+static const ExessTime     time     = {{0}, 1, 2, 3, 0};
+static const ExessDate     date     = {2001, 2, 3, {0}};
+
+static void
+check_read(ExessVariant* const variant,
+           const ExessDatatype datatype,
+           const char* const   string,
+           const ExessStatus   expected_status,
+           const size_t        expected_count)
+{
+  const ExessResult r = exess_read_variant(variant, datatype, string);
+
+  assert(r.status == expected_status);
+  assert(r.count == expected_count);
+  assert(variant->datatype == datatype);
+}
+
+static void
+test_read_variant(void)
+{
+  ExessVariant variant;
+
+  check_read(&variant, EXESS_NOTHING, "42", EXESS_UNSUPPORTED, 0);
+
+  check_read(&variant, EXESS_DECIMAL, "1.2", EXESS_SUCCESS, 3);
+  CHECK_POINTEE_EQUALS(exess_get_double(&variant), 1.2);
+
+  check_read(&variant, EXESS_DOUBLE, "3.4", EXESS_SUCCESS, 3);
+  assert(variant.value.as_double == 3.4);
+
+  check_read(&variant, EXESS_FLOAT, "5.6", EXESS_SUCCESS, 3);
+  assert(variant.value.as_float == 5.6f);
+
+  check_read(&variant, EXESS_BOOLEAN, "true", EXESS_SUCCESS, 4);
+  assert(variant.value.as_bool);
+
+  check_read(&variant, EXESS_INTEGER, "7", EXESS_SUCCESS, 1);
+  assert(variant.value.as_long == 7);
+
+  check_read(
+    &variant, EXESS_NON_POSITIVE_INTEGER, "f", EXESS_EXPECTED_DIGIT, 0);
+  check_read(&variant, EXESS_NON_POSITIVE_INTEGER, "1", EXESS_OUT_OF_RANGE, 1);
+  check_read(&variant, EXESS_NON_POSITIVE_INTEGER, "-8", EXESS_SUCCESS, 2);
+  assert(variant.value.as_long == -8);
+
+  check_read(&variant, EXESS_NEGATIVE_INTEGER, "f", EXESS_EXPECTED_DIGIT, 0);
+  check_read(&variant, EXESS_NEGATIVE_INTEGER, "1", EXESS_OUT_OF_RANGE, 1);
+  check_read(&variant, EXESS_NEGATIVE_INTEGER, "-9", EXESS_SUCCESS, 2);
+  assert(variant.value.as_long == -9);
+
+  check_read(&variant, EXESS_LONG, "10", EXESS_SUCCESS, 2);
+  assert(variant.value.as_long == 10);
+
+  check_read(&variant, EXESS_INT, "11", EXESS_SUCCESS, 2);
+  assert(variant.value.as_int == 11);
+
+  check_read(&variant, EXESS_SHORT, "12", EXESS_SUCCESS, 2);
+  assert(variant.value.as_short == 12);
+
+  check_read(&variant, EXESS_BYTE, "13", EXESS_SUCCESS, 2);
+  assert(variant.value.as_byte == 13);
+
+  check_read(&variant, EXESS_ULONG, "14", EXESS_SUCCESS, 2);
+  assert(variant.value.as_long == 14);
+
+  check_read(&variant, EXESS_UINT, "15", EXESS_SUCCESS, 2);
+  assert(variant.value.as_int == 15);
+
+  check_read(&variant, EXESS_USHORT, "16", EXESS_SUCCESS, 2);
+  assert(variant.value.as_short == 16);
+
+  check_read(&variant, EXESS_UBYTE, "17", EXESS_SUCCESS, 2);
+  assert(variant.value.as_byte == 17);
+
+  check_read(&variant, EXESS_POSITIVE_INTEGER, "-1", EXESS_EXPECTED_DIGIT, 0);
+  check_read(&variant, EXESS_POSITIVE_INTEGER, "0", EXESS_OUT_OF_RANGE, 1);
+  check_read(&variant, EXESS_POSITIVE_INTEGER, "18", EXESS_SUCCESS, 2);
+  assert(variant.value.as_long == 18);
+
+  check_read(&variant, EXESS_DATE, "2001-01-02", EXESS_SUCCESS, 10);
+  assert(variant.value.as_date.year == 2001);
+  assert(variant.value.as_date.month == 1);
+  assert(variant.value.as_date.day == 2);
+
+  check_read(&variant, EXESS_TIME, "12:15:01.25", EXESS_SUCCESS, 11);
+  assert(variant.value.as_time.hour == 12);
+  assert(variant.value.as_time.minute == 15);
+  assert(variant.value.as_time.second == 1);
+  assert(variant.value.as_time.nanosecond == 250000000);
+
+  char blob_data[] = {0, 0, 0};
+
+  variant.datatype           = EXESS_HEX;
+  variant.value.as_blob.size = sizeof(blob_data);
+  variant.value.as_blob.data = blob_data;
+  check_read(&variant, EXESS_HEX, "666F6F", EXESS_SUCCESS, 6);
+  assert(!strncmp(blob_data, "foo", sizeof(blob_data)));
+
+  variant.datatype           = EXESS_BASE64;
+  variant.value.as_blob.size = sizeof(blob_data);
+  variant.value.as_blob.data = blob_data;
+  check_read(&variant, EXESS_BASE64, "Zm9v", EXESS_SUCCESS, 4);
+  assert(!strncmp(blob_data, "foo", sizeof(blob_data)));
+}
+
+static void
+test_variant_string_length(void)
+{
+  const ExessVariant variant = {EXESS_DECIMAL, {.as_double = 12.3456}};
+
+  assert(exess_write_variant(variant, 0, NULL).count == 7);
+}
+
+static void
+check_write(const ExessVariant value,
+            const ExessStatus  expected_status,
+            const size_t       buf_size,
+            const char* const  expected_string)
+{
+  char buf[328] = {42};
+
+  assert(buf_size <= sizeof(buf));
+
+  const ExessResult r = exess_write_variant(value, buf_size, buf);
+  assert(r.status == expected_status);
+  if (buf_size > 0) {
+    assert(r.count == strlen(buf));
+    assert(!strcmp(buf, expected_string));
+  }
+}
+
+static void
+test_write_variant(void)
+{
+  char            blob_data[] = {'f', 'o', 'o'};
+  const ExessBlob blob        = {sizeof(blob_data), blob_data};
+
+  const ExessVariant a_nothing  = exess_make_nothing(EXESS_SUCCESS);
+  const ExessVariant a_bool     = exess_make_boolean(true);
+  const ExessVariant a_decimal  = exess_make_decimal(1.2);
+  const ExessVariant a_double   = exess_make_double(3.4);
+  const ExessVariant a_float    = exess_make_float(5.6f);
+  const ExessVariant a_long     = exess_make_long(7);
+  const ExessVariant a_int      = exess_make_int(8);
+  const ExessVariant a_short    = exess_make_short(9);
+  const ExessVariant a_byte     = exess_make_byte(10);
+  const ExessVariant a_ulong    = exess_make_ulong(11);
+  const ExessVariant a_uint     = exess_make_uint(12);
+  const ExessVariant a_ushort   = exess_make_ushort(13);
+  const ExessVariant a_ubyte    = exess_make_ubyte(14);
+  const ExessVariant a_duration = exess_make_duration(duration);
+  const ExessVariant a_datetime = exess_make_datetime(datetime);
+  const ExessVariant a_time     = exess_make_time(time);
+  const ExessVariant a_date     = exess_make_date(date);
+  const ExessVariant a_hex      = exess_make_hex(blob);
+  const ExessVariant a_base64   = exess_make_base64(blob);
+
+  check_write(a_nothing, EXESS_BAD_VALUE, 0, "");
+  check_write(a_nothing, EXESS_BAD_VALUE, 1, "");
+  check_write(a_decimal, EXESS_SUCCESS, 4, "1.2");
+  check_write(a_double, EXESS_SUCCESS, 6, "3.4E0");
+  check_write(a_float, EXESS_SUCCESS, 12, "5.5999999E0");
+  check_write(a_bool, EXESS_SUCCESS, 5, "true");
+  check_write(a_long, EXESS_SUCCESS, 2, "7");
+  check_write(a_int, EXESS_SUCCESS, 2, "8");
+  check_write(a_short, EXESS_SUCCESS, 2, "9");
+  check_write(a_byte, EXESS_SUCCESS, 3, "10");
+  check_write(a_ulong, EXESS_SUCCESS, 3, "11");
+  check_write(a_uint, EXESS_SUCCESS, 3, "12");
+  check_write(a_ushort, EXESS_SUCCESS, 3, "13");
+  check_write(a_ubyte, EXESS_SUCCESS, 3, "14");
+  check_write(a_duration, EXESS_SUCCESS, 15, "P1Y2M3DT4H5M6S");
+  check_write(a_datetime, EXESS_SUCCESS, 40, "2001-02-03T04:05:06");
+  check_write(a_time, EXESS_SUCCESS, 40, "01:02:03Z");
+  check_write(a_date, EXESS_SUCCESS, 40, "2001-02-03Z");
+  check_write(a_hex, EXESS_SUCCESS, 7, "666F6F");
+  check_write(a_base64, EXESS_SUCCESS, 5, "Zm9v");
+
+  const ExessBlob null_blob = {0, NULL};
+
+  const ExessVariant null_hex = exess_make_hex(null_blob);
+  check_write(null_hex, EXESS_BAD_VALUE, 99, "");
+
+  const ExessVariant null_base64 = exess_make_base64(null_blob);
+  check_write(null_base64, EXESS_BAD_VALUE, 99, "");
+}
+
+static void
+test_make_get(void)
+{
+  char            blob_data[] = {'f', 'o', 'o'};
+  const ExessBlob blob        = {sizeof(blob_data), blob_data};
+
+  const ExessVariant a_nothing  = exess_make_nothing(EXESS_NO_SPACE);
+  const ExessVariant a_bool     = exess_make_boolean(true);
+  const ExessVariant a_decimal  = exess_make_decimal(1.2);
+  const ExessVariant a_double   = exess_make_double(3.4);
+  const ExessVariant a_float    = exess_make_float(5.6f);
+  const ExessVariant a_long     = exess_make_long(7);
+  const ExessVariant a_int      = exess_make_int(8);
+  const ExessVariant a_short    = exess_make_short(9);
+  const ExessVariant a_byte     = exess_make_byte(10);
+  const ExessVariant a_ulong    = exess_make_ulong(11);
+  const ExessVariant a_uint     = exess_make_uint(12);
+  const ExessVariant a_ushort   = exess_make_ushort(13);
+  const ExessVariant a_ubyte    = exess_make_ubyte(14);
+  const ExessVariant a_duration = exess_make_duration(duration);
+  const ExessVariant a_datetime = exess_make_datetime(datetime);
+  const ExessVariant a_time     = exess_make_time(time);
+  const ExessVariant a_date     = exess_make_date(date);
+  const ExessVariant a_hex      = exess_make_hex(blob);
+  const ExessVariant a_base64   = exess_make_base64(blob);
+
+  // Different types as status
+  assert(exess_get_status(&a_nothing) == EXESS_NO_SPACE);
+  assert(exess_get_status(&a_bool) == EXESS_SUCCESS);
+
+  // Basic successful get
+  CHECK_POINTEE_EQUALS(exess_get_boolean(&a_bool), true);
+  CHECK_POINTEE_EQUALS(exess_get_double(&a_decimal), 1.2);
+  CHECK_POINTEE_EQUALS(exess_get_double(&a_double), 3.4);
+  CHECK_POINTEE_EQUALS(exess_get_float(&a_float), 5.6f);
+  CHECK_POINTEE_EQUALS(exess_get_long(&a_long), 7);
+  CHECK_POINTEE_EQUALS(exess_get_int(&a_int), 8);
+  CHECK_POINTEE_EQUALS(exess_get_short(&a_short), 9);
+  CHECK_POINTEE_EQUALS(exess_get_byte(&a_byte), 10);
+  CHECK_POINTEE_EQUALS(exess_get_ulong(&a_ulong), 11u);
+  CHECK_POINTEE_EQUALS(exess_get_uint(&a_uint), 12u);
+  CHECK_POINTEE_EQUALS(exess_get_ushort(&a_ushort), 13u);
+  CHECK_POINTEE_EQUALS(exess_get_ubyte(&a_ubyte), 14u);
+  assert(!memcmp(exess_get_duration(&a_duration), &duration, sizeof(duration)));
+  assert(!memcmp(exess_get_datetime(&a_datetime), &datetime, sizeof(datetime)));
+  assert(!memcmp(exess_get_time(&a_time), &time, sizeof(time)));
+  assert(!memcmp(exess_get_date(&a_date), &date, sizeof(date)));
+  assert(exess_get_blob(&a_hex)->size == sizeof(blob_data));
+  assert(exess_get_blob(&a_hex)->data == blob_data);
+  assert(exess_get_blob(&a_base64)->size == sizeof(blob_data));
+  assert(exess_get_blob(&a_base64)->data == blob_data);
+
+  // Unsuccessful get
+  assert(!exess_get_boolean(&a_int));
+  assert(!exess_get_double(&a_int));
+  assert(!exess_get_float(&a_int));
+  assert(!exess_get_long(&a_bool));
+  assert(!exess_get_int(&a_bool));
+  assert(!exess_get_short(&a_int));
+  assert(!exess_get_byte(&a_int));
+  assert(!exess_get_ulong(&a_int));
+  assert(!exess_get_uint(&a_int));
+  assert(!exess_get_ushort(&a_int));
+  assert(!exess_get_ubyte(&a_int));
+  assert(!exess_get_duration(&a_int));
+  assert(!exess_get_datetime(&a_int));
+  assert(!exess_get_time(&a_int));
+  assert(!exess_get_date(&a_int));
+  assert(!exess_get_blob(&a_int));
+}
+
+int
+main(void)
+{
+  test_read_variant();
+  test_variant_string_length();
+  test_write_variant();
+  test_make_get();
+
+  return 0;
+}
diff --git a/subprojects/exess/test/time_test_utils.h b/subprojects/exess/test/time_test_utils.h
new file mode 100644
index 00000000..00993207
--- /dev/null
+++ b/subprojects/exess/test/time_test_utils.h
@@ -0,0 +1,52 @@
+/*
+  Copyright 2011-2021 David Robillard <d@drobilla.net>
+
+  Permission to use, copy, modify, and/or distribute this software for any
+  purpose with or without fee is hereby granted, provided that the above
+  copyright notice and this permission notice appear in all copies.
+
+  THIS SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+  WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+  MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+  ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+  WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+  ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+  OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+*/
+
+#include "int_test_data.h"
+
+#include "exess/exess.h"
+
+#include <stdint.h>
+
+#define INIT_ZONE(hour, minute) \
+  {                             \
+    4 * (hour) + (minute) / 15  \
+  }
+
+static inline ExessTimezone
+random_timezone(uint32_t* rng)
+{
+  *rng = lcg32(*rng);
+
+  const int8_t hour = (int8_t)((*rng % 27) - 13);
+
+  *rng = lcg32(*rng);
+
+  const int8_t minute = (int8_t)((hour < 0 ? -1 : 1) * (int32_t)(*rng % 60));
+
+  const ExessTimezone zone = {(int8_t)(4 * hour + minute / 15)};
+  return zone;
+}
+
+static inline bool
+timezone_matches(const ExessTimezone zone,
+                 const int8_t        expected_hour,
+                 const int8_t        expected_minute,
+                 const bool          expected_is_present)
+
+{
+  return (!expected_is_present && zone.quarter_hours == EXESS_LOCAL) ||
+         zone.quarter_hours == 4 * expected_hour + expected_minute / 15;
+}
-- 
cgit v1.2.1