Added

• The string-lenient function as a companion to string but that accepts any
  cl:string as an argument.

Fixed

• The type of the string parser has been restored to reflect that it has only
  ever accepted (simple-array character (*)) as an argument.

The crash involving sliding-take appears to be gone, at least with SBCL 2.5.10. Other compilers have no issue.

The theme of this release is "type information". Overall, function signatures
have been greatly elaborated across the various systems.

Added

• ->, fn, always, and maybe for concise definition of function
  signatures that involve parsers.
• Parser: sliding-take and sliding-take1 for parsing escaped characters
  efficiently.

Changed

• <$ is now a macro.
• between is now a macro.
• :fail can be returned from the lambda given to ap to fail the whole parse.
  This is useful for manually validating the results of parse.
• toml: Improved performance.
• xml: Improved performance.

Removed

• Certain parsers no longer have an internal, automatically managed "lambda
  cache". In any case, it is considered best practice to instead define your own
  top-level, preallocated parsers for known cost-centers. These often involve
  char, string, between, and consume. For instance, you should
  essentially always define:

(defparameter +colon+ (p:char #\:))

And call +colon+ at usage sites, instead of using p:char directly within
your composite parsers. This avoids a significant amount of lambda allocation at
runtime. By doing this consistently you can expect a 3x-10x reduction in memory
allocation and a 2x-3x speed up.

For parsers/combinators that have a &key id in their arguments lists, the
argument have been left alone to avoid API breakage, although it now has no
effect. Even so, you should define your own defparameters for these as well.

Fixed

• Depending on the parser they were given, many and many1 were consuming too
  much offset. I only discovered this now as I mostly avoid many in favour of
  take-while or sep, but in some recent code I needed many for its
  original purpose (collecting a list), and the bug surfaced.

Added

• Combinator: not for succeeding when a parser fails. Useful when trying to
  test a complicated parse boundary, say in combination with sep-end.
• Combinator: consume-sep and consume-sep1 for advancing the parser like
  sep but without allocating.
• datetime: The simple-local-time parser for lenient parsing of local times.
• email: The new parcom/email system.

Changed

• A number of parsers/combinators are now more memory efficient by avoiding
  internal lambda allocations.

Added

• Combinator: maybe for conditionally applying a function to the result of a parser.

Changed

• toml: parse checks if you've reached the end of the file.
• xml: parse checks if you've reached the end of the file.
• json: Numbers now parse as fixnum if they can be.
• json: Number parsing is now 2x as fast and uses half as much memory.

Fixed

• xml: Allow a <!DOCTYPE ...> tag near the top of the file.

Added

• The parcom/xml system.

Changed

• Improved performance of parcom/json.
• Improved contextual output in parse failure condition.
• consume: Yields the offset as a value instead of t.
• A few more functions have been given lambda caches.

Added

• The parcom/toml system.
• Parser: pure for injecting values as parsers.
• Utility: pmap for transforming parser successes more directly than fmap.

Added

• The parcom/datetime system.
• Parser: any-if.
• Parser: sneak.
• Combinator: take-until.

Changed

• Vastly improved performance. Runtime is ~3x faster and uses 25x less memory.
• The signalling of parsing success and failure is done through multiple return
  values, not structs. The fundamental signature of parser functions is thus now
  offset -> (value, offset).
• Contextual information when parsing fails is more accurate.
• anybut -> any-but

Removed

• The explicit parser, failure, and input types. This is now managed by
  multiple return values, as described above.

Initial release, including parcom and parcom/json.