Revision history for Perl extension JSON::XS

TODO: maybe detect and croak on more invalid inputs (e.g. +-inf/nan)
TODO: maybe avoid the reblessing and better support readonly objects.
TODO: how to cope with tagged values and standard json decoders
TODO: investigate magic (Eric Brine)
TODO: [PATCH] Types::Serialiser: Inline true(), false() and error() functions
TODO: replace bool_stash by BOOL_STASH seems to work with mod_perl, make a compile time option?

4.03 Tue Oct 27 19:05:01 CET 2020
	- when parsing comments in relaxed mode, JSON::XS would detect garbage
          after the JSON text if the comment is after the end and does not end in
          a newline (reported by Felipe Gasper).

4.02 Wed Mar  6 08:31:24 CET 2019
	- undo the fix from 4.01, it breaks more things than it fixes
          (another testcase by Wesley Schwengle).
        - try a proper fix this time.

4.01 Sun Feb 24 05:03:30 CET 2019
	- fix some stack corruption caused mostly when calling methods
          in list context (testcase by Wesley Schwengle).

4.0  Fri Nov 16 00:06:54 CET 2018
	- SECURITY IMPLICATION: this release enables allow_nonref by default
          for compatibility with RFC 7159 and newer. See "old" vs. "new"
          JSON under SECURITY CONSIDERATIONS.
        - reworked the "old" vs. "new" JSON section.
        - add ->boolean_values to provide the values to which booleans
          decode (requested by Aristotle Pagaltzis).
        - decode would wrongly accept ASCII NUL characters instead of
          reporting them as trailing garbage.
        - work around what smells like a perl bug w.r.t. exceptions
          thrown in callbacks.
        - incremental parser now more or less respects allow_nonref.
	- json_xs json-pretty now enables canonical mode.
        - add documentation section about I-JSON.
        - minor documentation fixes/updates.

3.04 Thu Aug 17 04:30:47 CEST 2017
	- change exponential realloc algorithm on encoding and string decoding to be
          really exponential (this helps slow/debugging allocators such as libumem)
          (reported by Matthew Horsfall).
        - string encoding would needlessly overallocate output space
          (testcase by Matthew Horsfall).
        - be very paranoid about extending buffer lengths and croak if buffers get too large,
          which might (or might not) improve security.
	- add cbor-packed type to json_xs.
        - switch from YAML to YAML::XS in json_xs, as YAML is way too buggy and outdated.

3.03 Wed Nov 16 20:20:59 CET 2016
	- fix a bug introduced by a perl bug workaround that would cause
          incremental parsing to fail with a sv_chop panic.
	- json_xs: toformat failure error message fixed.
        - json_xs: allow cyclic data structures in CBOR.

3.02 Fri Feb 26 22:45:20 CET 2016
	- allow_nonref now affects booleans (\1, $Types::Serialiser::Boolean)
          as well (reported by Alex Efros).
	- allow literal tabs in strings in relaxed mode (patch by
          lubo.rintel@gooddata.com).
	- support "cbor" format in json_xs tool.
	- support (and fix) calling encode and decode in list context
          (reported by Вадим Власов).
        - work around a bug in older perls crashing when presented
          with shared hash keys (Reini Urban).
        - use stability canary.

3.01 Tue Oct 29 16:55:15 CET 2013
	- backport to perls < 5.18 (reported by Paul Howarth).

3.0  Tue Oct 29 01:35:37 CET 2013
        - implemented an object tagging extension (using the
          Types::Serialiser serialisation protocol).
        - reworked the documentation regarding object serialisation,
          add a new OBJECT SERIALISATION section that explains th
          whole process.
        - new setting: allow_tags.
	- switch to Types::Serialiser booleans.
	- remove to_json/from_json.
        - other minor improvements to the documentation.

2.34 Thu May 23 11:30:34 CEST 2013
	- work around bugs in perl 5.18 breaking more than 100
          widely used modules, without a fix in sight because
          p5pers don't care about CPAN anymore.
	- when canonicalising, only allocate up to 64 hash key
          pointers on the stack. for larger hashes, use the heap,
          to avoid using too much stackspace.
        - discuss the problem with setlocale (reported by a few victims).

2.33 Wed Aug  1 21:03:52 CEST 2012
	- internal encode/decode XS wrappers did not expect stack
          moves caused by callbacks (analyzed and testcase by Jesse Luehrs).
	- add bencode as to/from option in bin/json_xs.
        - add -e option to json_xs, and none and string in/out formats.

2.32 Thu Aug 11 19:06:38 CEST 2011
	- fix a bug in the initial whitespace accumulation.

2.31 Wed Jul 27 17:53:05 CEST 2011
	- don't accumulate initial whitespace in the incremental buffer
          (this can be useful to allow whitespace-keepalive on a tcp
          connection without triggering the max_size limit).
	- properly croak on some invalid inputs that are not strings
          (e.g. undef) when trying to decode a json text (reported
          and analyzed by Goro Fuji).

2.3   Wed Aug 18 01:26:47 CEST 2010
	- make sure decoder doesn't change the decoding in the incremental
          parser (testcase provided by Hendrik Schumacher).
	- applied patch by DaTa for Data::Dumper support in json_xs.
        - added -t dump support to json_xs, using Data::Dump.
        - added -f eval support to json_xs.

2.29  Wed Mar 17 02:39:12 CET 2010
	- fix a memory leak when callbacks set using filter_json_object
          or filter_json_single_key_object were called (great testcase
          by Eric Wilhelm).

2.28  Thu Mar 11 20:30:46 CET 2010
	- implement our own atof function - perl's can be orders of
          magnitudes slower than even the system one. on the positive
          side, ours seems to be more exact in general than perl's.
          (testcase provided by Tim Meadowcroft).
        - clarify floating point conversion issues a bit.
	- update jpsykes csrf article url.
        - updated benchmark section - JSON::PP became much faster!

2.27  Thu Jan  7 07:35:08 CET 2010
	- support relaxed option inside the incremental parser
          (testcase provided by IKEGAMI via Makamaka).

2.26  Sat Oct 10 03:26:19 CEST 2009
	- big integers could become truncated (based on patch
          by Strobl Anton).
	- output format change: indent now adds a final newline, which is
          more expected and more true to the documentation.

2.25  Sat Aug  8 12:04:41 CEST 2009
	- the perl debugger completely breaks lvalue subs - try to work
          around the issue.
	- ignore RMAGICAL hashes w.r.t. CANONICAL.
	- try to work around a possible char signedness issue on aix.
        - require common sense.

2.24  Sat May 30 08:25:45 CEST 2009
	- the incremental parser did not update its parse offset
          pointer correctly when parsing utf8-strings (nicely
          debugged by Martin Evans).
	- appending a non-utf8-string to the incremental parser
          in utf8 mode failed to upgrade the string.
        - wording of parse error messages has been improved.

2.232 Sun Feb 22 11:12:25 CET 2009
	- use an exponential algorithm to extend strings, to
          help platforms with bad or abysmal==windows memory
          allocater performance, at the expense of some memory
          wastage (use shrink to recover this extra memory).
          (nicely analysed by Dmitry Karasik).

2.2311 Thu Feb 19 02:12:54 CET 2009
        - add a section "JSON and ECMAscript" to explain some
          incompatibilities between the two (problem was noted by
          various people).
	- add t/20_faihu.t.

2.231 Thu Nov 20 04:59:08 CET 2008
	- work around 5.10.0 magic bugs where manipulating magic values
          (such as $1) would permanently damage them as perl would
          ignore the magicalness, by making a full copy of the string,
          reported by Dmitry Karasik.
        - work around spurious warnings under older perl 5.8's.

2.23 Mon Sep 29 05:08:29 CEST 2008
	- fix a compilation problem when perl is not using char * as, well,
          char *.
	- use PL_hexdigit in favour of rolling our own.

2.2222 Sun Jul 20 18:49:00 CEST 2008
	- same game again, broken 5.10 finds yet another assertion
          failure, and the workaround causes additional runtime warnings.
          Work around the next assertion AND the warning. 5.10 seriously
          needs to adjust it's attitude against working code.

2.222 Sat Jul 19 06:15:34 CEST 2008
	- you work around one -DDEBUGGING assertion bug in perl 5.10
          just to hit the next one. work around this one, too.

2.22 Tue Jul 15 13:26:51 CEST 2008
	- allow higher nesting levels in incremental parser.
        - error out earlier in some cases in the incremental parser
          (as suggested by Yuval Kogman).
        - improve incr-parser test (Yuval Kogman).

2.21 Tue Jun  3 08:43:23 CEST 2008
	- (hopefully) work around a perl 5.10 bug with -DDEBUGGING.
	- remove the experimental status of the incremental parser interface.
	- move =encoding around again, to avoid bugs with search.cpan.org.
          when can we finally have utf-8 in pod???
        - add ->incr_reset method.

2.2  Wed Apr 16 20:37:25 CEST 2008
	- lifted the log2 rounding restriction of max_depth and max_size.
	- make booleans mutable by creating a copy instead of handing out
          the same scalar (reported by pasha sadri).
	- added support for incremental json parsing (still EXPERIMENTAL).
	- implemented and added a json_xs command line utility that can convert
          from/to a number of serialisation formats - tell me if you need more.
        - implement allow_unknown/get_allow_unknown methods.
        - fixed documentation of max_depth w.r.t. higher and equal.
	- moved down =encoding directive a bit, too much breaks if it's the first
          pod directive :/.
        - removed documentation section on other modules, it became somewhat
          outdated and is nowadays mostly of historical interest.

2.1  Wed Mar 19 23:23:18 CET 2008
        - update documentation here and there: add a large section
          about utf8/latin1/ascii flags, add a security consideration
          and extend and clarify the JSON and YAML section.
        - medium speed enhancements when encoding/decoding non-ascii chars.
        - minor speedup in number encoding case.
	- extend and clarify the section on incompatibilities
          between YAML and JSON.
        - switch to static inline from just inline when using gcc.
        - add =encoding utf-8 to the manpage, now that perl 5.10 supports it.
        - fix some issues with UV to JSON conversion of unknown impact.
        - published the yahoo locals search result used in benchmarks as the
          original url changes so comparison is impossible.

2.01 Wed Dec  5 11:40:28 CET 2007
	- INCOMPATIBLE API CHANGE: to_json and from_json have been
          renamed to encode_json/decode_json for JSON.pm compatibility.
          The old functions croak and might be replaced by JSON.pm
          comaptible versions in some later release.

2.0  Tue Dec  4 11:30:46 CET 2007
	- this is supposed to be the first version of JSON::XS
          compatible with version 2.0+ of the JSON module.
          Using the JSON module as frontend to JSON::XS should be
          as fast as using JSON::XS directly, so consider using it
          instead.
	- added get_* methods for all "simple" options.
        - make JSON::XS subclassable.

1.53 Tue Nov 13 23:58:33 CET 2007
	- minor doc clarifications.
        - fixed many doc typos (patch by Thomas L. Shinnick).

1.52 Mon Oct 15 03:22:06 CEST 2007
	- remove =encoding pod directive again, it confuses too many pod
          parsers :/.

1.51 Sat Oct 13 03:55:56 CEST 2007
	- encode empty arrays/hashes in a compact way when pretty is enabled.
	- apparently JSON::XS was used to find some bugs in the
          JSON_checker testsuite, so add (the corrected) JSON_checker tests to
          the testsuite.
        - quite a bit of doc updates/extension.
        - require 5.8.2, as this seems to be the first unicode-stable version.

1.5  Tue Aug 28 04:05:38 CEST 2007
	- add support for tied hashes, based on ideas and testcase by
          Marcus Holland-Moritz.
        - implemented relaxed parsing mode where some extensions are being
          accepted. generation is still JSON-only.

1.44 Wed Aug 22 01:02:44 CEST 2007
	- very experimental process-emulation support, slowing everything down.
          the horribly broken perl threads are still not supported - YMMV.

1.43 Thu Jul 26 13:26:37 CEST 2007
	- convert big json numbers exclusively consisting of digits to NV
          only when there is no loss of precision, otherwise to string.

1.42 Tue Jul 24 00:51:18 CEST 2007
	- fix a crash caused by not handling missing array elements
          (report and testcase by Jay Kuri).

1.41 Tue Jul 10 18:21:44 CEST 2007
	- fix compilation with NDEBUG (assert side-effect),
          affects convert_blessed only.
	- fix a bug in decode filters calling ENTER; SAVETMPS;
          one time too often.
        - catch a typical error in TO_JSON methods.
	- antique-ised XS.xs again to work with outdated
          C compilers (windows...).

1.4  Mon Jul  2 10:06:30 CEST 2007
	- add convert_blessed setting.
        - encode did not catch all blessed objects, encoding their
          contents in most cases. This has been fixed by introducing
          the allow_blessed setting.
        - added filter_json_object and filter_json_single_key_object
          settings that specify a callback to be called when
          all/specific json objects are encountered.
        - assume that most object keys are simple ascii words and
          optimise this case, penalising the general case. This can
          speed up decoding by 30% in typical cases and gives
          a smaller and faster perl hash.
        - implemented simpleminded, optional resource size checking
          in decode_json.
        - remove objToJson/jsonToObj aliases, as the next version
          of JSON will not have them either.
        - bit the bullet and converted the very simple json object
          into a more complex one.
        - work around a bug where perl wrongly claims an integer
          is not an integer.
        - unbundle JSON::XS::Boolean into own pm file so Storable
          and similar modules can resolve the overloading when thawing.

1.3  Sun Jun 24 01:55:02 CEST 2007
        - make JSON::XS::true and false special overloaded objects
          and return those instead of 1 and 0 for those json atoms
          (JSON::PP compatibility is NOT achieved yet).
        - add JSON::XS::is_bool predicate to test for those special
          values.
	- add a reference to
          http://jpsykes.com/47/practical-csrf-and-json-security.
        - removed require 5.8.8 again, it is just not very expert-friendly.
          Also try to be more compatible with slightly older versions,
          which are not recommended (because they are buggy).

1.24 Mon Jun 11 05:40:49 CEST 2007
	- added informative section on JSON-as-YAML.
        - get rid of some c99-isms again.
        - localise dec->cur in decode_str, speeding up
          string decoding considerably (>15% on my amd64 + gcc).
        - increased SHORT_STRING_LEN to 16kb: stack space is
          usually plenty, and this actually saves memory
          when !shrinking as short strings will fit perfectly.

1.23 Wed Jun  6 20:13:06 CEST 2007
	- greatly improved small integer encoding and decoding speed.
	- implement a number of µ-optimisations.
        - updated benchmarks.

1.22 Thu May 24 00:07:25 CEST 2007
	- require 5.8.8 explicitly as older perls do not seem to offer
          the required macros.
        - possibly made it compile on so-called C compilers by microsoft.

1.21 Wed May  9 18:40:32 CEST 2007
	- character offset reported for trailing garbage was random.

1.2  Wed May  9 18:35:01 CEST 2007
        - decode did not work with magical scalars (doh!).
        - added latin1 flag to produce JSON texts in the latin1 subset
          of unicode.
        - flag trailing garbage as error.
        - new decode_prefix method that returns the number
          of characters consumed by a decode.
        - max octets/char in perls UTF-X is actually 13, not 11,
          as pointed out by Glenn Linderman.
	- fixed typoe reported by YAMASHINA Hio.

1.11 Mon Apr  9 07:05:49 CEST 2007
	- properly 0-terminate sv's returned by encode to help
          C libraries that expect that 0 to be there.
	- partially "port" JSON from C to microsofts fucking broken
          pseudo-C. They should be burned to the ground for pissing
          on standards. And I should be stoned for even trying to
          support this filthy excuse for a c compiler.

1.1  Wed Apr  4 01:45:00 CEST 2007
	- clarify documentation (pointed out by Quinn Weaver).
        - decode_utf8 sometimes did not correctly flag errors,
          leading to segfaults.
        - further reduced default nesting depth to 512 due to the test
          failure by that anonymous "chris" whose e-mail address seems
          to be impossible to get. Tests on other freebsd systems indicate
          that this is likely a problem in his/her configuration and not this
          module.
        - renamed json => JSON in error messages.
        - corrected the character offset in some error messages.

1.01 Sat Mar 31 16:15:40 CEST 2007
	- do not segfault when from_json/decode gets passed
          a non-string object (reported by Florian Ragwitz).
          This has no effect on normal operation.

1.0  Thu Mar 29 04:43:34 CEST 2007
	- the long awaited (by me) 1.0 version.
        - add \0 (JSON::XS::false) and \1 (JSON::XS::true) mappings to JSON
          true and false.
	- add some more notes to shrink, as suggested by Alex Efros.
        - improve testsuite.
        - halve the default nesting depth limit, to hopefully make it
          work on Freebsd (unfortunately, the cpan tester did not
          send me his report, so I cannot ask about the stack limit on fbsd).

0.8  Mon Mar 26 00:10:48 CEST 2007
	- fix a memleak when decoding hashes.
	- export jsonToBj and objToJson as aliases
          to to_json and from_json, to reduce incompatibilities
          between JSON/JSON::PC and JSON::XS. (experimental).
        - implement a maximum nesting depth for both en- and de-coding.
        - added a security considerations sections.

0.7  Sun Mar 25 01:46:30 CET 2007
	- code cleanup.
	- fix a memory overflow bug when indenting.
        - pretty-printing now up to 15% faster.
        - improve decoding speed of strings by
          up to 50% by specialcasing short strings.
        - further decoding speedups for strings using
          lots of \u escapes.
        - improve utf8 decoding speed for U+80 .. U+7FF.

0.5  Sat Mar 24 20:41:51 CET 2007
	- added the UTF-16 encoding example hinted at in previous
          versions.
        - minor documentation fixes.
        - fix a bug in and optimise canonicalising fastpath
          (reported by Craig Manley).
        - remove a subtest that breaks with bleadperl (reported
          by Andreas König).

0.31 Sat Mar 24 02:14:34 CET 2007
	- documentation updates.
        - do some casting to hopefully fix Andreas' problem.
        - nuke bogus json rpc stuff.

0.3  Fri Mar 23 19:33:21 CET 2007
	- remove spurious PApp::Util reference (John McNamara).
	- adapted lots of tests from other json modules
          (idea by Chris Carline).
        - documented mapping from json to perl and vice versa.
        - improved the documentation by adding more examples.
        - added short escaping forms, reducing the created
          json texts a bit.
        - added shrink flag.
        - when flag methods are called without enable argument
          they will by default enable their flag.
        - considerably improved string encoding speed (at least
          with gcc 4).
        - added a test that covers lots of different characters.
        - clarified some error messages.
        - error messages now use correct character offset
          with F_UTF8.
        - improve the "no bytes" and "no warnings" hacks in
          case the called functions do... stuff.
        - croak when encoding to ascii and an out-of-range
          (non-unicode) codepoint is encountered.

0.2  Fri Mar 23 00:23:34 CET 2007
	- the "could not sleep without debugging release".
          it should basically work now, with many bugs as
          no production tests have been run yet.
	- added more testcases.
	- the expected shitload of bugfixes.
	- handle utf8 flag correctly in decode.
        - fix segfault in decoder.
        - utf8n_to_uvuni sets retlen to -1, but retlen is an
          unsigned types (argh).
        - fix decoding of utf-8 strings.
        - improved error diagnostics.
        - fix decoding of 'null'.
        - fix parsing of empty array/hashes
        - silence warnings when we prepare the croak message.

0.1   Thu Mar 22 22:13:43 CET 2007
	- first release, very untested, basically just to claim
          the namespace.
 
0.01  Thu Mar 22 06:08:12 CET 2007
	- original version; cloned from Convert-Scalar