Age | Commit message (Collapse) | Author | Files | Lines |
|
|
|
|
|
This is mainly for developer or power-user cases, where one wants to look at
some data for investigation or debugging. In such cases, it's common for the
set of prefixes to be implicitly known (because they are baked in to the
application, for example), so printing them just produces a large amount of
redundant noise.
That said, it can also be useful programmatically, because it allows several
snippets to be written independently and ultimately concatenated (with a header
to define the prefixes) without redundancy.
|
|
More or less the same rationale as the previous commit, but for reading. This
makes for nice symmetry with writing, at the cost of a slightly more annoying
reader interface since the source doesn't know its block size or name.
|
|
This makes the paging mechanism an internal detail once again. While it's
conceptually elegant to simply have a single write interface and have the block
dumper just be another implementation of that, unfortunately it is not
practical. The inlining of serd_block_dumper_write() is a significant
performance boost, because it avoids a non-inlinable function call of overhead
per character.
Compared to the SerdByteSink approach, this removes the burden and overhead of
needing to dynamically allocate the structure itself.
|
|
Essentially replaces serd_buffer_sink_finish() with serd_buffer_close(), which
makes writing to a buffer consistent with writing to a file or anything else.
|
|
This seems like pointless complexity now, since it's easy to just write simply
ordered statements yourself.
|
|
This is a common convention in Turtle and TriG because the special "a" syntax
for rdf type as the first property looks nice, makes things easier to read, and
can be useful for streaming implementations because the type of the instance is
known before reading its properties.
Also significantly clean up the pretty-printing implementation in the process.
|
|
|
|
The earlier "test" was just hitting the code without actually checking the
output. This new suite is a set of pretty-printed documents which serd must
reproduce from a model exactly to pass. This should make it easy to add cases
in the future, since each case is just a document, as it should look.
|
|
|
|
Especially with the new functionality, the complexity of the command-line
interface alone was really becoming unmanageable. The serdi implementation
also had the highest cyclomatic complexity of the entire codebase by a huge
margin.
So, take a page from the Unix philosophy and split serdi into several more
finely-honed tools that can be freely composed. Though there is still
unfortunately quite a bit of option overlap between them due to the common
details of reading RDF, I think the resulting tools are a lot easier to
understand, both from a user and a developer perspective.
|
|
Though potentially useful, I don't think the complexity cost of the old
interface (both to the implementation and to the user) is worth it. A special
tool to transform blank node labels (for example with regular expressions)
would be a better approach to this if it's ever needed in the future.
|
|
While occasionally useful, I almost always end up reproducing the issue live to
investigate something anyway. Not keeping the many tests results around
results in less clutter, and hopefully makes the test suites faster in
environments with bad I/O like Docker.
|
|
|
|
|
|
This simplifies everything by replacing special cases with a more general close
function. A type is no longer stored in the structure, so the other
constructors are now essentially syntactic sugar for the universal
serd_byte_sink_new_function().
|
|
|
|
This separates the command-line tool code from the library implementation.
|
|
This is more powerful, and reduces the number of command line options that
almost nobody needs to care about.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
This is more or less a total rewrite of SerdNodes and the underlying ZixHash to
make efficient use of the new node construction API.
|
|
|
|
Things get confusing without a term for this concept (which is roughly "nodes
that are not annoying to construct"), so "token" it is.
|
|
|
|
|
|
|
|
|
|
This allows a precise location to be reported for errors within literals, by
adding the offset of the error in the literal to the caret. This will be used
to report nice errors for things like regular expressions and supported XSD
datatypes.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
This expands relative and prefixed URIs in the reader on the stack, rather than
passing them to the caller to be dealt with. This pushes these context-full
forms to the edge of the system as much as possible to minimise the headaches
they can cause.
Towards having stricter guarantees about nodes and eliminating the CURIE node
type altogether.
|
|
Writing having side-effects seems questionable in general, and this prepares
things for expanding URIs in the reader.
|
|
|
|
This adds a reader flag and serdi option for extending a syntax with support
for SPARQL-like variables, for storing things like patterns or simple queries.
|
|
|
|
|
|
|
|
|
|
|
|
This allows parsing documents like "(42) ."
|