This necessitates passing the input around through all the sub-parsers so they
can compute their own sub-spans, which in turn necessitates making most of the
subparsers into nom-style 'function builders' by wrapping them in closures and
returning the closures.
The current impl hides this functionality behind an on-by-default "span" feature.
Some fallout of this change:
There is now a private parser::KdlParser type that must be constructed with the
full input for all span computation. The parser::parser method has been moved to
a method of this to make sure the full_input used for span computation is the
same one passed to parsing. In doing this a bug where the span of errors would
be incorrectly computed for non-ascii strings has been fixed.
Types now manually implement PartialEq and Hash to avoid incorporating the span
of the item into those computations.
All parser combinators must first have the KdlParser applied before using them,
making their signatures far more complex. Unfortunately KdlParser::parse can't
apply that for you because I don't think it's possible to express that complex
of a signature and not have the lifetimes get messed up.
This commit includes a whole bunch of fixes, some of which are significant changes to the parser and some related functionality. But I consider all changes to be bugfixes because they were compliance failures.
This new version of kdl-rs is a complete rewrite that introduces
a formatting-aware-and-preserving parser, much like toml_edit et al.
BREAKING CHANGE: Completely new API and bumped MSRV to 1.56.0.
* Add line and column numbers to KdlError
These numbers are calculated using our `newline` rule, as I'm not aware
of any standard for how to calculate line numbers in error messages.
The error retains the original `offset` field as well, but it's now
expressed in characters rather than bytes.
* Clean up error handling in parse_document()
We don't need to worry about `Incomplete` errors, the `complete` family
of parsers (which we're using) doesn't ever produce those. We can use
the `.finish()` function to unwrap the error instead, so we don't have
to worry about the underlying error types.
Using the `ignore` marker makes `cargo test` think this is rust code
that should be ignored, which means it prints an "ignored" line in the
test output. It also makes `cargo doc --document-private-items` attempt
to colorize it as rust code. Marking this as `text` instead fixes both
issues.
Aria Beingessner
Kat Marchán
jam1garner
hclarke
Lily Ballard