WIP: Make Reader always read from a ByteSource

14 files changed, 322 insertions, 280 deletions

diff --git a/include/serd/serd.h b/include/serd/serd.h
index 61ae6099..e36dd24b 100644
--- a/include/serd/serd.h
+++ b/include/serd/serd.h
@@ -105,9 +105,6 @@ typedef struct SerdWriterImpl SerdWriter;
 /// An interface that receives a stream of RDF data
 typedef struct SerdSinkImpl SerdSink;
 
-/// A sink for bytes that receives string output
-typedef struct SerdByteSinkImpl SerdByteSink;
-
 /// Return status code
 typedef enum {
   SERD_SUCCESS,        ///< No error
@@ -123,6 +120,7 @@ typedef enum {
   SERD_ERR_NO_DATA,    ///< Unexpected end of input
   SERD_ERR_BAD_TEXT,   ///< Invalid text encoding
   SERD_ERR_BAD_WRITE,  ///< Error writing to file/stream
+  SERD_ERR_BAD_CALL,   ///< Invalid call
 } SerdStatus;
 
 /// RDF syntax type
@@ -371,10 +369,13 @@ serd_base64_decode(const char* SERD_NONNULL str,
 
 /**
    @}
-   @defgroup serd_streams Byte Streams
+   @defgroup serd_byte_source Byte Source
    @{
 */
 
+/// A source for bytes that provides text input
+typedef struct SerdByteSourceImpl SerdByteSource;
+
 /**
    Function to detect I/O stream errors.
 
@@ -402,7 +403,69 @@ typedef size_t (*SerdReadFunc)(void* SERD_NONNULL buf,
                                void* SERD_NONNULL stream);
 
 /**
-   Sink function for raw string output.
+   Create a new byte source that reads from a string.
+
+   @param string Null-terminated UTF-8 string to read from.
+   @param name Optional name of stream for error messages (string or URI).
+*/
+SERD_API
+SerdByteSource* SERD_ALLOCATED
+serd_byte_source_new_string(const char* SERD_NONNULL      string,
+                            const SerdNode* SERD_NULLABLE name);
+
+/**
+   Create a new byte source that reads from a file.
+
+   An arbitrary `FILE*` can be used via serd_byte_source_new_function() as
+   well, this is just a convenience function that opens the file properly,
+   sets flags for optimized I/O if possible, and automatically sets the name of
+   the source to the file path.
+
+   @param path Path of file to open and read from.
+   @param block_size Number of bytes to read per call.
+*/
+SERD_API
+SerdByteSource* SERD_ALLOCATED
+serd_byte_source_new_filename(const char* SERD_NONNULL path, size_t block_size);
+
+/**
+   Create a new byte source that reads from a user-specified function
+
+   The `stream` will be passed to the `read_func`, which is compatible with
+   the standard C `fread` if `stream` is a `FILE*`.  Note that the serd Reader
+   only ever reads individual bytes at a time, that is, the `size` parameter
+   will always be 1 (but `nmemb` may be higher).
+
+   @param read_func Function called with bytes to consume.
+   @param error_func Stream error function with `ferror` semantics.
+   @param stream Context parameter passed to `read_func` and `error_func`.
+   @param name Optional name of stream for error messages (string or URI).
+   @param block_size Number of bytes to read per call.
+*/
+SERD_API
+SerdByteSource* SERD_ALLOCATED
+serd_byte_source_new_function(SerdReadFunc SERD_NONNULL        read_func,
+                              SerdStreamErrorFunc SERD_NONNULL error_func,
+                              void* SERD_NULLABLE              stream,
+                              const SerdNode* SERD_NULLABLE    name,
+                              size_t                           block_size);
+
+/// Free `source`
+SERD_API
+void
+serd_byte_source_free(SerdByteSource* SERD_NULLABLE source);
+
+/**
+   @}
+   @defgroup serd_byte_sink Byte Sink
+   @{
+*/
+
+/// A sink for bytes that receives text output
+typedef struct SerdByteSinkImpl SerdByteSink;
+
+/**
+   Sink function for raw string output
 
    Identical semantics to `fwrite`, but may set errno for more informative
    error reporting than supported by SerdStreamErrorFunc.
@@ -1241,37 +1304,14 @@ void
 serd_reader_add_blank_prefix(SerdReader* SERD_NONNULL  reader,
                              const char* SERD_NULLABLE prefix);
 
-/// Prepare to read from the file at a local file `uri`
-SERD_API
-SerdStatus
-serd_reader_start_file(SerdReader* SERD_NONNULL reader,
-                       const char* SERD_NONNULL uri,
-                       bool                     bulk);
-
-/**
-   Prepare to read from a stream.
-
-   The `read_func` is guaranteed to only be called for `page_size` elements
-   with size 1 (i.e. `page_size` bytes).
-*/
-SERD_API
-SerdStatus
-serd_reader_start_stream(SerdReader* SERD_NONNULL         reader,
-                         SerdReadFunc SERD_NONNULL        read_func,
-                         SerdStreamErrorFunc SERD_NONNULL error_func,
-                         void* SERD_NONNULL               stream,
-                         const SerdNode* SERD_NULLABLE    name,
-                         size_t                           page_size);
-
-/// Prepare to read from a string
+/// Prepare to read from a byte source
 SERD_API
 SerdStatus
-serd_reader_start_string(SerdReader* SERD_NONNULL      reader,
-                         const char* SERD_NONNULL      utf8,
-                         const SerdNode* SERD_NULLABLE name);
+serd_reader_start(SerdReader* SERD_NONNULL     reader,
+                  SerdByteSource* SERD_NONNULL byte_source);
 
 /**
-   Read a single "chunk" of data during an incremental read
+   Read a single "chunk" of data during an incremental read.
 
    This function will read a single top level description, and return.  This
    may be a directive, statement, or several statements; essentially it reads
diff --git a/src/byte_source.c b/src/byte_source.c
index ca02ff3e..1c86fcce 100644
--- a/src/byte_source.c
+++ b/src/byte_source.c
@@ -20,118 +20,167 @@
 
 #include "serd/serd.h"
 
+#if defined(USE_POSIX_FADVISE) && defined(USE_FILENO)
+#  include <fcntl.h>
+#endif
+
 #include <stdbool.h>
 #include <stdint.h>
+#include <stdio.h>
+#include <stdlib.h>
 #include <string.h>
 
-SerdStatus
-serd_byte_source_page(SerdByteSource* source)
+SerdByteSource*
+serd_byte_source_new_string(const char* string, const SerdNode* name)
 {
-  uint8_t* const buf =
-    (source->page_size > 1 ? source->file_buf : &source->read_byte);
+  SerdByteSource* source = (SerdByteSource*)calloc(1, sizeof(SerdByteSource));
 
-  const size_t n_read =
-    source->read_func(buf, 1, source->page_size, source->stream);
+  source->page_size = 1;
+  source->read_buf  = (const uint8_t*)string;
+  source->type      = FROM_STRING;
 
-  source->buf_size  = n_read;
-  source->read_head = 0;
-  source->eof       = false;
+  source->name =
+    name ? serd_node_copy(name) : serd_new_string(SERD_STATIC_STRING("string"));
 
-  if (n_read < source->page_size) {
-    buf[n_read] = '\0';
-    if (n_read == 0) {
-      source->eof = true;
-      return (source->error_func(source->stream) ? SERD_ERR_UNKNOWN
-                                                 : SERD_FAILURE);
-    }
-  }
+  const SerdCursor cur = {source->name, 1, 1};
+  source->cur          = cur;
 
-  return SERD_SUCCESS;
+  return source;
 }
 
-SerdStatus
-serd_byte_source_open_source(SerdByteSource*     source,
-                             SerdReadFunc        read_func,
-                             SerdStreamErrorFunc error_func,
-                             SerdStreamCloseFunc close_func,
-                             void*               stream,
-                             const SerdNode*     name,
-                             size_t              page_size)
+SerdByteSource*
+serd_byte_source_new_filename(const char* path, size_t block_size)
 {
-  assert(page_size > 0);
-  memset(source, '\0', sizeof(*source));
-  source->read_func   = read_func;
-  source->error_func  = error_func;
-  source->close_func  = close_func;
-  source->stream      = stream;
-  source->page_size   = page_size;
-  source->buf_size    = page_size;
-  source->name        = serd_node_copy(name);
-  source->from_stream = true;
+  if (!path || !block_size) {
+    return NULL;
+  }
+
+  FILE* fd = fopen(path, "rb");
+  if (!fd) {
+    return NULL;
+  }
+
+  SerdByteSource* source = (SerdByteSource*)calloc(1, sizeof(SerdByteSource));
+
+  source->read_func  = (SerdReadFunc)fread;
+  source->error_func = (SerdStreamErrorFunc)ferror;
+  source->close_func = (SerdStreamCloseFunc)fclose;
+  source->stream     = fd;
+  source->page_size  = block_size;
+  source->buf_size   = block_size;
+  source->name =
+    serd_new_file_uri(SERD_MEASURE_STRING(path), SERD_EMPTY_STRING());
+  source->type = FROM_FILENAME;
 
   const SerdCursor cur = {source->name, 1, 1};
   source->cur          = cur;
 
-  if (page_size > 1) {
-    source->file_buf = (uint8_t*)serd_allocate_buffer(page_size);
+  if (block_size > 1) {
+    source->file_buf = (uint8_t*)serd_allocate_buffer(block_size);
     source->read_buf = source->file_buf;
-    memset(source->file_buf, '\0', page_size);
+    memset(source->file_buf, '\0', block_size);
   } else {
     source->read_buf = &source->read_byte;
   }
 
-  return SERD_SUCCESS;
+#if defined(USE_POSIX_FADVISE) && defined(USE_FILENO)
+  posix_fadvise(fileno(fd), 0, 0, POSIX_FADV_SEQUENTIAL);
+#endif
+
+  return source;
 }
 
-SerdStatus
-serd_byte_source_prepare(SerdByteSource* source)
+SerdByteSource*
+serd_byte_source_new_function(SerdReadFunc        read_func,
+                              SerdStreamErrorFunc error_func,
+                              void*               stream,
+                              const SerdNode*     name,
+                              size_t              block_size)
 {
-  if (source->page_size == 0) {
-    return SERD_FAILURE;
+  if (!read_func || !block_size) {
+    return NULL;
   }
 
-  source->prepared = true;
+  SerdByteSource* source = (SerdByteSource*)calloc(1, sizeof(SerdByteSource));
+
+  source->read_func  = read_func;
+  source->error_func = error_func;
+  // source->close_func  = close_func; FIXME
+  source->stream    = stream;
+  source->page_size = block_size;
+  source->buf_size  = block_size;
+  source->type      = FROM_FUNCTION;
+
+  source->name =
+    name ? serd_node_copy(name) : serd_new_string(SERD_STATIC_STRING("func"));
+
+  const SerdCursor cur = {source->name, 1, 1};
+  source->cur          = cur;
 
-  if (source->from_stream) {
-    return (source->page_size > 1 ? serd_byte_source_page(source)
-                                  : serd_byte_source_advance(source));
+  if (block_size > 1) {
+    source->file_buf = (uint8_t*)serd_allocate_buffer(block_size);
+    source->read_buf = source->file_buf;
+    memset(source->file_buf, '\0', block_size);
+  } else {
+    source->read_buf = &source->read_byte;
   }
 
-  return SERD_SUCCESS;
+  return source;
+}
+
+void
+serd_byte_source_free(SerdByteSource* source)
+{
+  if (source) {
+    if (source->close_func) {
+      source->close_func(source->stream);
+    }
+
+    if (source->page_size > 1) {
+      free(source->file_buf);
+    }
+
+    serd_node_free(source->name);
+    free(source);
+  }
 }
 
 SerdStatus
-serd_byte_source_open_string(SerdByteSource* source,
-                             const char*     utf8,
-                             const SerdNode* name)
+serd_byte_source_page(SerdByteSource* source)
 {
-  memset(source, '\0', sizeof(*source));
+  uint8_t* const buf =
+    (source->page_size > 1 ? source->file_buf : &source->read_byte);
 
-  source->page_size = 1;
-  source->name =
-    name ? serd_node_copy(name) : serd_new_string(SERD_STATIC_STRING("string"));
+  const size_t n_read =
+    source->read_func(buf, 1, source->page_size, source->stream);
 
-  source->read_buf = (const uint8_t*)utf8;
+  source->buf_size  = n_read;
+  source->read_head = 0;
+  source->eof       = false;
 
-  const SerdCursor cur = {source->name, 1, 1};
-  source->cur          = cur;
+  if (n_read < source->page_size) {
+    buf[n_read] = '\0';
+    if (n_read == 0) {
+      source->eof = true;
+      return (source->error_func(source->stream) ? SERD_ERR_UNKNOWN
+                                                 : SERD_FAILURE);
+    }
+  }
 
   return SERD_SUCCESS;
 }
 
 SerdStatus
-serd_byte_source_close(SerdByteSource* source)
+serd_byte_source_prepare(SerdByteSource* source)
 {
-  SerdStatus st = SERD_SUCCESS;
-  if (source->close_func) {
-    st = source->close_func(source->stream) ? SERD_ERR_UNKNOWN : SERD_SUCCESS;
-  }
+  source->prepared = true;
+  if (source->type != FROM_STRING) {
+    if (source->page_size > 1) {
+      return serd_byte_source_page(source);
+    }
 
-  if (source->page_size > 1) {
-    serd_free_aligned(source->file_buf);
+    return serd_byte_source_advance(source);
   }
 
-  serd_node_free(source->name);
-  memset(source, '\0', sizeof(*source));
-  return st;
+  return SERD_SUCCESS;
 }
diff --git a/src/byte_source.h b/src/byte_source.h
index ff281435..a420808f 100644
--- a/src/byte_source.h
+++ b/src/byte_source.h
@@ -26,42 +26,31 @@
 #include <stddef.h>
 #include <stdint.h>
 
-typedef int (*SerdStreamCloseFunc)(void*);
-
-typedef struct {
-  SerdReadFunc        read_func;   ///< Read function (e.g. fread)
-  SerdStreamErrorFunc error_func;  ///< Error function (e.g. ferror)
-  SerdStreamCloseFunc close_func;  ///< Function for closing stream
-  void*               stream;      ///< Stream (e.g. FILE)
-  size_t              page_size;   ///< Number of bytes to read at a time
-  size_t              buf_size;    ///< Number of bytes in file_buf
-  SerdNode*           name;        ///< Name of stream (referenced by cur)
-  SerdCursor          cur;         ///< Cursor for error reporting
-  uint8_t*            file_buf;    ///< Buffer iff reading pages from a file
-  const uint8_t*      read_buf;    ///< Pointer to file_buf or read_byte
-  size_t              read_head;   ///< Offset into read_buf
-  uint8_t             read_byte;   ///< 1-byte 'buffer' used when not paging
-  bool                from_stream; ///< True iff reading from `stream`
-  bool                prepared;    ///< True iff prepared for reading
-  bool                eof;         ///< True iff end of file reached
-} SerdByteSource;
-
-SerdStatus
-serd_byte_source_open_string(SerdByteSource* source,
-                             const char*     utf8,
-                             const SerdNode* name);
+typedef enum {
+  FROM_STRING,   ///< Reading from a user-provided buffer
+  FROM_FILENAME, ///< Reading from a file we opened
+  FROM_FUNCTION, ///< Reading from a user-provided function
+} SerdByteSourceType;
 
-SerdStatus
-serd_byte_source_open_source(SerdByteSource*     source,
-                             SerdReadFunc        read_func,
-                             SerdStreamErrorFunc error_func,
-                             SerdStreamCloseFunc close_func,
-                             void*               stream,
-                             const SerdNode*     name,
-                             size_t              page_size);
+typedef int (*SerdStreamCloseFunc)(void*);
 
-SerdStatus
-serd_byte_source_close(SerdByteSource* source);
+struct SerdByteSourceImpl {
+  SerdReadFunc        read_func;  ///< Read function (e.g. fread)
+  SerdStreamErrorFunc error_func; ///< Error function (e.g. ferror)
+  SerdStreamCloseFunc close_func; ///< Function for closing stream
+  void*               stream;     ///< Stream (e.g. FILE)
+  size_t              page_size;  ///< Number of bytes to read at a time
+  size_t              buf_size;   ///< Number of bytes in file_buf
+  SerdNode*           name;       ///< Name of stream (referenced by cur)
+  SerdCursor          cur;        ///< Cursor for error reporting
+  uint8_t*            file_buf;   ///< Buffer iff reading pages from a file
+  const uint8_t*      read_buf;   ///< Pointer to file_buf or read_byte
+  size_t              read_head;  ///< Offset into read_buf
+  SerdByteSourceType  type;       ///< Type of input
+  uint8_t             read_byte;  ///< 1-byte 'buffer' used when not paging
+  bool                prepared;   ///< True iff prepared for reading
+  bool                eof;        ///< True iff end of file reached
+};
 
 SerdStatus
 serd_byte_source_prepare(SerdByteSource* source);
@@ -91,7 +80,7 @@ serd_byte_source_advance(SerdByteSource* source)
     ++source->cur.col;
   }
 
-  if (source->from_stream) {
+  if (source->type != FROM_STRING) {
     if (++source->read_head >= source->buf_size) {
       st = serd_byte_source_page(source);
     }
diff --git a/src/n3.c b/src/n3.c
index a91a8367..c9f2dd6b 100644
--- a/src/n3.c
+++ b/src/n3.c
@@ -1097,7 +1097,7 @@ read_object(SerdReader* reader, ReadContext* ctx, bool emit, bool* ate_dot)
   static const size_t      XSD_BOOLEAN_LEN = 40;
 
   const size_t orig_stack_size = reader->stack.size;
-  SerdCursor   orig_cursor     = reader->source.cur;
+  SerdCursor   orig_cursor     = reader->source->cur;
 
   SerdStatus ret    = SERD_FAILURE;
   bool       simple = (ctx->subject != 0);
@@ -1634,11 +1634,11 @@ skip_until(SerdReader* reader, uint8_t byte)
 SerdStatus
 read_turtleTrigDoc(SerdReader* reader)
 {
-  while (!reader->source.eof) {
+  while (!reader->source->eof) {
     const size_t     orig_stack_size = reader->stack.size;
     const SerdStatus st              = read_n3_statement(reader);
     if (st > SERD_FAILURE) {
-      if (reader->strict || reader->source.eof || st == SERD_ERR_OVERFLOW) {
+      if (reader->strict || reader->source->eof || st == SERD_ERR_OVERFLOW) {
         serd_stack_pop_to(&reader->stack, orig_stack_size);
         return st;
       }
@@ -1653,7 +1653,7 @@ SerdStatus
 read_nquadsDoc(SerdReader* reader)
 {
   SerdStatus st = SERD_SUCCESS;
-  while (!st && !reader->source.eof) {
+  while (!st && !reader->source->eof) {
     const size_t orig_stack_size = reader->stack.size;
 
     SerdStatementFlags flags   = 0;
diff --git a/src/reader.c b/src/reader.c
index b0b35387..1c9469af 100644
--- a/src/reader.c
+++ b/src/reader.c
@@ -19,7 +19,6 @@
 #include "namespaces.h"
 #include "stack.h"
 #include "statement.h"
-#include "system.h"
 #include "world.h"
 
 #include <errno.h>
@@ -36,7 +35,7 @@ r_err(SerdReader* reader, SerdStatus st, const char* fmt, ...)
 {
   va_list args;
   va_start(args, fmt);
-  const SerdError e = {st, &reader->source.cur, fmt, &args};
+  const SerdError e = {st, &reader->source->cur, fmt, &args};
   serd_world_error(reader->world, &e);
   va_end(args);
   return st;
@@ -123,7 +122,7 @@ emit_statement(SerdReader* reader, ReadContext ctx, SerdNode* o)
   serd_node_zero_pad(o);
 
   const SerdStatement statement = {{ctx.subject, ctx.predicate, o, ctx.graph},
-                                   &reader->source.cur};
+                                   &reader->source->cur};
 
   const SerdStatus st =
     serd_sink_write_statement(reader->sink, *ctx.flags, &statement);
@@ -141,7 +140,11 @@ read_statement(SerdReader* reader)
 SerdStatus
 serd_reader_read_document(SerdReader* reader)
 {
-  if (!reader->source.prepared) {
+  if (!reader->source) {
+    return SERD_ERR_BAD_CALL;
+  }
+
+  if (!reader->source->prepared) {
     SerdStatus st = serd_reader_prepare(reader);
     if (st) {
       return st;
@@ -221,12 +224,12 @@ serd_reader_add_blank_prefix(SerdReader* reader, const char* prefix)
 static SerdStatus
 skip_bom(SerdReader* me)
 {
-  if (serd_byte_source_peek(&me->source) == 0xEF) {
-    serd_byte_source_advance(&me->source);
-    if (serd_byte_source_peek(&me->source) != 0xBB ||
-        serd_byte_source_advance(&me->source) ||
-        serd_byte_source_peek(&me->source) != 0xBF ||
-        serd_byte_source_advance(&me->source)) {
+  if (serd_byte_source_peek(me->source) == 0xEF) {
+    serd_byte_source_advance(me->source);
+    if (serd_byte_source_peek(me->source) != 0xBB ||
+        serd_byte_source_advance(me->source) ||
+        serd_byte_source_peek(me->source) != 0xBF ||
+        serd_byte_source_advance(me->source)) {
       r_err(me, SERD_ERR_BAD_SYNTAX, "corrupt byte order mark\n");
       return SERD_ERR_BAD_SYNTAX;
     }
@@ -236,64 +239,23 @@ skip_bom(SerdReader* me)
 }
 
 SerdStatus
-serd_reader_start_stream(SerdReader*         reader,
-                         SerdReadFunc        read_func,
-                         SerdStreamErrorFunc error_func,
-                         void*               stream,
-                         const SerdNode*     name,
-                         size_t              page_size)
-{
-  serd_reader_finish(reader);
-  return serd_byte_source_open_source(
-    &reader->source, read_func, error_func, NULL, stream, name, page_size);
-}
-
-SerdStatus
-serd_reader_start_file(SerdReader* reader, const char* uri, bool bulk)
+serd_reader_start(SerdReader* reader, SerdByteSource* byte_source)
 {
   serd_reader_finish(reader);
 
-  char* const path = serd_parse_file_uri(uri, NULL);
-  if (!path) {
-    return SERD_ERR_BAD_ARG;
-  }
-
-  FILE* fd = serd_world_fopen(reader->world, path, "rb");
-  free(path);
-  if (!fd) {
-    return SERD_ERR_UNKNOWN;
-  }
+  reader->source = byte_source;
 
-  SerdNode* const  name = serd_new_uri(SERD_MEASURE_STRING(uri));
-  const SerdStatus st   = serd_byte_source_open_source(
-    &reader->source,
-    bulk ? (SerdReadFunc)fread : serd_file_read_byte,
-    (SerdStreamErrorFunc)ferror,
-    (SerdStreamCloseFunc)fclose,
-    fd,
-    name,
-    bulk ? SERD_PAGE_SIZE : 1u);
-  serd_node_free(name);
-  return st;
-}
-
-SerdStatus
-serd_reader_start_string(SerdReader*     reader,
-                         const char*     utf8,
-                         const SerdNode* name)
-{
-  serd_reader_finish(reader);
-  return serd_byte_source_open_string(&reader->source, utf8, name);
+  return reader->source ? SERD_SUCCESS : SERD_ERR_BAD_ARG;
 }
 
 static SerdStatus
 serd_reader_prepare(SerdReader* reader)
 {
-  SerdStatus st = serd_byte_source_prepare(&reader->source);
+  SerdStatus st = serd_byte_source_prepare(reader->source);
   if (st == SERD_SUCCESS) {
     st = skip_bom(reader);
   } else if (st == SERD_FAILURE) {
-    reader->source.eof = true;
+    reader->source->eof = true;
   } else {
     r_err(reader, st, "error preparing read: %s\n", strerror(errno));
   }
@@ -304,10 +266,14 @@ SerdStatus
 serd_reader_read_chunk(SerdReader* reader)
 {
   SerdStatus st = SERD_SUCCESS;
-  if (!reader->source.prepared) {
+  if (!reader->source) {
+    return SERD_ERR_BAD_CALL;
+  }
+
+  if (!reader->source->prepared) {
     st = serd_reader_prepare(reader);
-  } else if (reader->source.eof) {
-    st = serd_byte_source_advance(&reader->source);
+  } else if (reader->source->eof) {
+    st = serd_byte_source_advance(reader->source);
   }
 
   if (peek_byte(reader) == 0) {
@@ -321,5 +287,6 @@ serd_reader_read_chunk(SerdReader* reader)
 SerdStatus
 serd_reader_finish(SerdReader* reader)
 {
-  return serd_byte_source_close(&reader->source);
+  reader->source = NULL;
+  return SERD_SUCCESS;
 }
diff --git a/src/reader.h b/src/reader.h
index 7865fb39..62feeee6 100644
--- a/src/reader.h
+++ b/src/reader.h
@@ -57,7 +57,7 @@ struct SerdReaderImpl {
   SerdNode*       rdf_first;
   SerdNode*       rdf_rest;
   SerdNode*       rdf_nil;
-  SerdByteSource  source;
+  SerdByteSource* source;
   SerdStack       stack;
   SerdSyntax      syntax;
   unsigned        next_id;
@@ -114,7 +114,7 @@ read_turtleTrigDoc(SerdReader* reader);
 static inline int
 peek_byte(SerdReader* reader)
 {
-  SerdByteSource* source = &reader->source;
+  SerdByteSource* source = reader->source;
 
   return source->eof ? EOF : (int)source->read_buf[source->read_head];
 }
@@ -127,7 +127,7 @@ eat_byte_safe(SerdReader* reader, const int byte)
   const int c = peek_byte(reader);
   assert(c == byte);
 
-  serd_byte_source_advance(&reader->source);
+  serd_byte_source_advance(reader->source);
   return c;
 }
 
diff --git a/src/serdi.c b/src/serdi.c
index e8ef9897..1925cd88 100644
--- a/src/serdi.c
+++ b/src/serdi.c
@@ -103,25 +103,34 @@ read_file(SerdWorld* const      world,
   syntax = syntax ? syntax : serd_guess_syntax(filename);
   syntax = syntax ? syntax : SERD_TRIG;
 
-  SerdStatus  st     = SERD_SUCCESS;
-  SerdReader* reader = serd_reader_new(world, syntax, flags, sink, stack_size);
-
-  serd_reader_add_blank_prefix(reader, add_prefix);
-
+  SerdByteSource* byte_source = NULL;
   if (!strcmp(filename, "-")) {
     SerdNode* name = serd_new_string(SERD_STATIC_STRING("stdin"));
 
-    st = serd_reader_start_stream(
-      reader, serd_file_read_byte, (SerdStreamErrorFunc)ferror, stdin, name, 1);
+    byte_source = serd_byte_source_new_function(
+      serd_file_read_byte, (SerdStreamErrorFunc)ferror, stdin, name, 1);
 
     serd_node_free(name);
   } else {
-    st = serd_reader_start_file(reader, filename, bulk_read);
+    byte_source =
+      serd_byte_source_new_filename(filename, bulk_read ? SERD_PAGE_SIZE : 1u);
   }
 
+  if (!byte_source) {
+    SERDI_ERRORF("failed to open input file `%s'\n", filename);
+    return SERD_ERR_UNKNOWN;
+  }
+
+  SerdReader* reader = serd_reader_new(world, syntax, flags, sink, stack_size);
+
+  serd_reader_add_blank_prefix(reader, add_prefix);
+
+  SerdStatus st = serd_reader_start(reader, byte_source);
+
   st = st ? st : serd_reader_read_document(reader);
 
   serd_reader_free(reader);
+  serd_byte_source_free(byte_source);
 
   return st;
 }
@@ -307,6 +316,9 @@ main(int argc, char** argv)
   SerdStatus st         = SERD_SUCCESS;
   SerdNode*  input_name = NULL;
   if (input_string) {
+    SerdByteSource* const byte_source =
+      serd_byte_source_new_string(input_string, NULL);
+
     SerdReader* const reader =
       serd_reader_new(world,
                       input_syntax ? input_syntax : SERD_TRIG,
@@ -316,13 +328,12 @@ main(int argc, char** argv)
 
     serd_reader_add_blank_prefix(reader, add_prefix);
 
-    SerdNode* name = serd_new_string(SERD_STATIC_STRING("string"));
-    if (!(st = serd_reader_start_string(reader, input_string, name))) {
+    if (!(st = serd_reader_start(reader, byte_source))) {
       st = serd_reader_read_document(reader);
     }
 
-    serd_node_free(name);
     serd_reader_free(reader);
+    serd_byte_source_free(byte_source);
   }
 
   size_t prefix_len = 0;
diff --git a/src/string.c b/src/string.c
index f0aca005..b5904c84 100644
--- a/src/string.c
+++ b/src/string.c
@@ -62,6 +62,8 @@ serd_strerror(SerdStatus status)
     return "Invalid text encoding";
   case SERD_ERR_BAD_WRITE:
     return "Error writing to file";
+  case SERD_ERR_BAD_CALL:
+    return "Invalid call";
   default:
     break;
   }
diff --git a/src/world.c b/src/world.c
index 2cbee1a6..0e8df944 100644
--- a/src/world.c
+++ b/src/world.c
@@ -21,11 +21,6 @@
 #include "cursor.h"
 #include "namespaces.h"
 #include "node.h"
-#include "serd_config.h"
-
-#if defined(USE_POSIX_FADVISE)
-#  include <fcntl.h>
-#endif
 
 #include <errno.h>
 #include <stdarg.h>
@@ -35,24 +30,6 @@
 
 #define BLANK_CHARS 12
 
-FILE*
-serd_world_fopen(SerdWorld* world, const char* path, const char* mode)
-{
-  FILE* fd = fopen(path, mode);
-  if (!fd) {
-    serd_world_errorf(world,
-                      SERD_ERR_INTERNAL,
-                      "failed to open file %s (%s)\n",
-                      path,
-                      strerror(errno));
-    return NULL;
-  }
-#if defined(USE_POSIX_FADVISE) && defined(USE_FILENO)
-  posix_fadvise(fileno(fd), 0, 0, POSIX_FADV_SEQUENTIAL);
-#endif
-  return fd;
-}
-
 SerdStatus
 serd_world_error(const SerdWorld* world, const SerdError* e)
 {
diff --git a/src/world.h b/src/world.h
index 44e21166..4f4101e4 100644
--- a/src/world.h
+++ b/src/world.h
@@ -37,10 +37,6 @@ struct SerdWorldImpl {
   uint32_t        next_blank_id;
 };
 
-/// Open a file configured for fast sequential reading
-FILE*
-serd_world_fopen(SerdWorld* world, const char* path, const char* mode);
-
 SerdStatus
 serd_world_error(const SerdWorld* world, const SerdError* e);
 
diff --git a/test/test_free_null.c b/test/test_free_null.c
index 62615a02..a101379a 100644
--- a/test/test_free_null.c
+++ b/test/test_free_null.c
@@ -24,6 +24,7 @@ int
 main(void)
 {
   serd_free(NULL);
+  serd_byte_source_free(NULL);
   serd_byte_sink_free(NULL);
   serd_node_free(NULL);
   serd_world_free(NULL);
diff --git a/test/test_overflow.c b/test/test_overflow.c
index 8d451f2a..5a65d887 100644
--- a/test/test_overflow.c
+++ b/test/test_overflow.c
@@ -24,11 +24,14 @@
 static SerdStatus
 test(SerdWorld* world, SerdSink* sink, const char* str, size_t stack_size)
 {
+  SerdByteSource* byte_source = serd_byte_source_new_string(str, NULL);
+
   SerdReader* reader = serd_reader_new(world, SERD_TURTLE, 0, sink, stack_size);
 
-  serd_reader_start_string(reader, str, NULL);
+  serd_reader_start(reader, byte_source);
   const SerdStatus st = serd_reader_read_document(reader);
   serd_reader_free(reader);
+  serd_byte_source_free(byte_source);
   return st;
 }
 
diff --git a/test/test_read_chunk.c b/test/test_read_chunk.c
index 1bbe2709..1cb19aaf 100644
--- a/test/test_read_chunk.c
+++ b/test/test_read_chunk.c
@@ -94,20 +94,22 @@ main(void)
   SerdWorld* world = serd_world_new();
   SerdSink*  sink  = serd_sink_new(NULL, on_event, NULL);
 
+  SerdByteSource* byte_source =
+    serd_byte_source_new_string("@prefix eg: <http://example.org/> .\n"
+                                "@base <http://example.org/base> .\n"
+                                "eg:s1 eg:p1 eg:o1 ;\n"
+                                "      eg:p2 eg:o2 ,\n"
+                                "            eg:o3 .\n"
+                                "eg:s2 eg:p1 eg:o1 ;\n"
+                                "      eg:p2 eg:o2 .\n"
+                                "eg:s3 eg:p1 eg:o1 .\n"
+                                "eg:s4 eg:p1 [ eg:p3 eg:o1 ] .\n",
+                                NULL);
+
   SerdReader* reader = serd_reader_new(world, SERD_TURTLE, 0, sink, 4096);
   assert(reader);
 
-  assert(!serd_reader_start_string(reader,
-                                   "@prefix eg: <http://example.org/> .\n"
-                                   "@base <http://example.org/base> .\n"
-                                   "eg:s1 eg:p1 eg:o1 ;\n"
-                                   "      eg:p2 eg:o2 ,\n"
-                                   "            eg:o3 .\n"
-                                   "eg:s2 eg:p1 eg:o1 ;\n"
-                                   "      eg:p2 eg:o2 .\n"
-                                   "eg:s3 eg:p1 eg:o1 .\n"
-                                   "eg:s4 eg:p1 [ eg:p3 eg:o1 ] .\n",
-                                   NULL));
+  assert(!serd_reader_start(reader, byte_source));
 
   assert(!serd_reader_read_chunk(reader) && n_prefix == 1);
   assert(!serd_reader_read_chunk(reader) && n_base == 1);
@@ -121,6 +123,7 @@ main(void)
   assert(!serd_reader_finish(reader));
 
   serd_reader_free(reader);
+  serd_byte_source_free(byte_source);
   serd_sink_free(sink);
   serd_world_free(world);
 
diff --git a/test/test_reader_writer.c b/test/test_reader_writer.c
index a4f57a0f..86c6b442 100644
--- a/test/test_reader_writer.c
+++ b/test/test_reader_writer.c
@@ -19,7 +19,6 @@
 #include "serd/serd.h"
 
 #include <assert.h>
-#include <stdbool.h>
 #include <stdint.h>
 #include <stdio.h>
 #include <stdlib.h>
@@ -80,8 +79,10 @@ test_read_chunks(void)
   SerdReader* const reader = serd_reader_new(world, SERD_TURTLE, 0, sink, 4096);
   assert(reader);
 
-  SerdStatus st = serd_reader_start_stream(
-    reader, (SerdReadFunc)fread, (SerdStreamErrorFunc)ferror, f, NULL, 1);
+  SerdByteSource* byte_source = serd_byte_source_new_function(
+    (SerdReadFunc)fread, (SerdStreamErrorFunc)ferror, f, NULL, 1);
+
+  SerdStatus st = serd_reader_start(reader, byte_source);
   assert(st == SERD_SUCCESS);
 
   // Write two statement separated by null characters
@@ -123,6 +124,7 @@ test_read_chunks(void)
   assert(n_statements == 2);
 
   serd_reader_free(reader);
+  serd_byte_source_free(byte_source);
   serd_sink_free(sink);
   fclose(f);
   serd_world_free(world);
@@ -197,18 +199,20 @@ test_read_string(void)
   SerdReader* const reader = serd_reader_new(world, SERD_TURTLE, 0, sink, 4096);
   assert(reader);
 
+  SerdByteSource* byte_source =
+    serd_byte_source_new_string("<http://example.org/s> <http://example.org/p> "
+                                "<http://example.org/o> .",
+                                NULL);
+
   // Test reading a string that ends exactly at the end of input (no newline)
-  assert(
-    !serd_reader_start_string(reader,
-                              "<http://example.org/s> <http://example.org/p> "
-                              "<http://example.org/o> .",
-                              NULL));
+  assert(!serd_reader_start(reader, byte_source));
 
   assert(!serd_reader_read_document(reader));
   assert(n_statements == 1);
   assert(!serd_reader_finish(reader));
 
   serd_reader_free(reader);
+  serd_byte_source_free(byte_source);
   serd_sink_free(sink);
   serd_world_free(world);
 }
@@ -334,8 +338,8 @@ test_reader(const char* path)
   SerdReader* reader = serd_reader_new(world, SERD_TURTLE, 0, sink, 4096);
   assert(reader);
 
-  assert(serd_reader_read_document(reader) == SERD_FAILURE);
-  assert(serd_reader_read_chunk(reader) == SERD_FAILURE);
+  assert(serd_reader_read_document(reader) == SERD_ERR_BAD_CALL);
+  assert(serd_reader_read_chunk(reader) == SERD_ERR_BAD_CALL);
 
   serd_reader_add_blank_prefix(reader, "tmp");
 
@@ -348,14 +352,12 @@ test_reader(const char* path)
 #  pragma GCC diagnostic pop
 #endif
 
-  assert(serd_reader_start_file(reader, "http://notafile", false));
-  assert(serd_reader_start_file(reader, "file://invalid", false));
-  assert(serd_reader_start_file(reader, "file:///nonexistant", false));
-
-  assert(!serd_reader_start_file(reader, path, true));
+  SerdByteSource* byte_source = serd_byte_source_new_filename(path, 4096);
+  assert(!serd_reader_start(reader, byte_source));
   assert(!serd_reader_read_document(reader));
   assert(n_statements == 6);
   serd_reader_finish(reader);
+  serd_byte_source_free(byte_source);
 
   // A read of a big page hits EOF then fails to read chunks immediately
   {
@@ -365,35 +367,37 @@ test_reader(const char* path)
     fflush(temp);
     fseek(temp, 0L, SEEK_SET);
 
-    serd_reader_start_stream(reader,
-                             (SerdReadFunc)fread,
-                             (SerdStreamErrorFunc)ferror,
-                             temp,
-                             NULL,
-                             4096);
+    byte_source = serd_byte_source_new_function(
+      (SerdReadFunc)fread, (SerdStreamErrorFunc)ferror, temp, NULL, 4096);
 
+    assert(serd_reader_start(reader, byte_source) == SERD_SUCCESS);
     assert(serd_reader_read_chunk(reader) == SERD_SUCCESS);
     assert(serd_reader_read_chunk(reader) == SERD_FAILURE);
     assert(serd_reader_read_chunk(reader) == SERD_FAILURE);
 
     serd_reader_finish(reader);
+    serd_byte_source_free(byte_source);
     fclose(temp);
   }
 
   // A byte-wise reader that hits EOF once then continues (like a socket)
   {
     size_t n_reads = 0;
-    serd_reader_start_stream(reader,
-                             (SerdReadFunc)eof_test_read,
-                             (SerdStreamErrorFunc)eof_test_error,
-                             &n_reads,
-                             NULL,
-                             1);
 
+    byte_source =
+      serd_byte_source_new_function((SerdReadFunc)eof_test_read,
+                                    (SerdStreamErrorFunc)eof_test_error,
+                                    &n_reads,
+                                    NULL,
+                                    1);
+
+    assert(serd_reader_start(reader, byte_source) == SERD_SUCCESS);
     assert(serd_reader_read_chunk(reader) == SERD_SUCCESS);
     assert(serd_reader_read_chunk(reader) == SERD_FAILURE);
     assert(serd_reader_read_chunk(reader) == SERD_SUCCESS);
     assert(serd_reader_read_chunk(reader) == SERD_FAILURE);
+    assert(!serd_reader_finish(reader));
+    serd_byte_source_free(byte_source);
   }
 
   serd_reader_free(reader);