Disabled repeat guards. They keep causing issues, and it's just simpler to rely on timeouts.
Version: 2022.10.31
Updated text for supported Unicode and Python versions.
Version: 2022.9.13
Updated to Unicode 15.0.0.
Version: 2022.9.11
Updated version.
Version: 2022.8.17
Git issue 477: \v for vertical spacing
Added \p{HorizSpace} (\p{H}) and \p{VertSpace} (\p{V}).
Version: 2022.7.25
Git issue 475: 2022.7.24 improperly released
The file https://pypi.org/pypi/regex/2022.7.24/json was missing references to most of the wheels, so this is a new release in the hope that it was just a glitch in GitHub Actions.
Version: 2022.7.24
Git issue 474: regex has no equivalent to re.Match.groups() for captures
Added 'allcaptures' and 'allspans' methods to match objects.
Fixed bug where compiling a pattern didn't always check for unused arguments.
Version: 2022.7.9
Git issue 473: Emoji classified as letter
The values for GC:Assigned and GC:LC were flipped.
Version: 2022.6.2
Git issue 472: Revisit compilation flag to prevent adding a single explicitly compiled regex to the cache
Added 'cache_pattern' parameter to 'compile' function to improve use of the cache.
Version: 2022.4.24
Git issue 467: Scoped inline flags 'a', 'u' and 'L' affect global flags
Those flags scan now be scoped.
Version: 2022.3.15
Git issue 457: Difference with `re`, when repl returns None
Make regex consistent with re by treating a replacement template of None as ''.
Also, now rejects invalid ASCII escapes like re module does.
Version: 2022.3.2
Git issue 453: Document last supported python2 version
Fixed problems with use of fast searching tables in opposite direction.
Version: 2021.8.27
Git issue 420: segmentation fault in finditer (maybe others)
Fixed a bugs in fast searches in reverse direction.
Version: 2021.8.21
Updated version.
Version: 2021.8.3
Forgot to update version!
Version: 2021.7.6
Additional fix for Git issue 415.
Version: 2021.7.5
Git issue 415: Fuzzy character restrictions don't apply to insertions at "right edge"
Version: 2021.7.1
Git issue 407: API is not a drop-in replacement for python's re when it comes to typing
Now exports Match object as well as Pattern object.
Git issue 414: Memory optimization questions
sys.getsizeof returns a more accurate size of a pattern object. It includes the size of internal data, but, as is the norm, does not include the size of public objects.
Version: 2021.4.4
Git issue 408: regex fails with a quantified backreference but succeeds with repeated backref
Git issue 407: API is not a drop-in replacement for python's re when it comes to typing
Version: 2021.3.17
Git issue 403: Fuzzy matching with wrong distance (unnecessary substitutions)
Reworked the fuzzy matching code.
Version: 2020.11.13
Git issue 394: Unexpected behaviour in fuzzy matching with limited character set with IGNORECASE flag
Version: 2020.11.11
Update version.
Version: 2020.11.2
Updated list of supported Python versions.
Added .travis.yml file.
Version: 2020.10.28
Git issue 362: Any LICENSE work for this project?
Changed licence to Apache 2.0 and added licence file.
Version: 2020.10.23
Git issue 387: Compilaton flag to avoid storing compiled regexp in internal cache
Slight reversion/revision. You can prevent explicitly-compiled patterns from being cached by using "cache_all(False)".
Version: 2020.10.22
Git issue 387: Compilaton flag to avoid storing compiled regexp in internal cache
No longer caches patterns that are compiled explicitly.
Version: 2020.10.15
Git issue 386: GCC 10 warnings
Fixed bugs in fuzzy_match_string_fld and fuzzy_match_group_fld.
Added more braces around data in some Unicode tables.
Version: 2020.10.11
Git issue 385: Comments in expressions
Didn't parse regex comments property when in VERBOSE mode.
Version: 2020.9.27
Git issue 383: Memory Error - regex.findall
The problem was caused by a lazy repeat looping forever, growing the backtracking stack. Greedy repeats were OK.
Version: 2020.7.14
Git issue 377: request: \h for horizontal space
Added \h as an alias to [[:blank:]].
Version: 2020.6.7
Git issue 376: Is the \L option as efficient as it can be?
Improved performance of string sets.
Version: 2020.6.7
Git issue 376: Is the \L option as efficient as it can be?
Switched StringSet to use fallback method due to inefficiencies in the engine. Needs more investigation.
Version: 2020.5.14
Git issue 372: Regression from 2020.4.4 -> 2020.5.7 in non-fuzzy matching pattern
Changed the 'state' member that's tested in is_repeat_guarded for a fuzzy match. The previously-used member wasn't initialised in a non-fuzzy match. The new test is a better one to use anyway.
Version: 2020.5.13
Git issue 371: Specifying character set when fuzzy-matching allows characters not in the set
fuzzy_ext_match and fuzzy_ext_match_group_fld didn't support sets!
Version: 2020.5.7
Git issue 370: Confusions about Fuzzy matching behavior (prob a bug?)
Version: 2020.4.4
Updated to Unicode 13.0.0.
Version: 2020.2.20
Git issue 365: Memory leak occurs in fuzzy match at some substitution use cases
Version: 2020.2.18
Git issue #364: Contradictory values in fuzzy_counts and fuzzy_changes
Version: 2020.1.7
Issue 357: New exception "ValueError: unused keyword argument" breaks use case
Added ignore_unused keyword argument.
Issue 359: 2020.1.7 source distribution release contains \r\n line endings
Fixed line endings for source distribution.
Issue 360: Invalid modeline in `_regex.c`
Removed vim modeline.
Version: 2020.1.7
Fix to previous change.
Version: 2019.12.21
Hg issue 353: fuzzy changes negative indexes
Fuzzy change positions were off by 1 for deletions.
Version: 2019.12.18
Another complaint from Linux.
Version: 2019.12.17
New release and upload because of problem with source distribution.
Version: 2019.12.17
New release and upload because of previous issues.
Fixed a couple of bugs in unicode_at_grapheme_boundary.
Version: 2019.10.31
Made "Additional Features" linkable.
Updated setup.
Minor bug fix.
Version: 2019.8.19
Hg issue 338: specifying allowed characters when fuzzy-matching
Added character testing to a fuzzy constraint.
Version: 2019.6.8
Hg issue 333: error when installing regex on PyPy2.7 v7.1.1 on Windows
PyPy isn't officially supported, but this might fix it!
Version: 2019.6.5
Updated for Python 3.8.
Version: 2019.6.2
Updated to Unicode 12.1.0.
Version: 2019.5.25
Hg issue 329: Wrong group matches when question mark quantifier is used within a look behind
REPEAT_ONE was backtracking in the wrong direction, so it never hit the limit.
Version: 2019.4.14
Hg issue 327: .fullmatch() causes MemoryError
For fullmatch, added check for end/start of string for RE_OP_SUCCESS in try_match.
Version: 2019.4.12
Missing brace in 'state_fini'.
Version: 2019.4.10
Hg issue 325: module docstring not accessible
Additional fix for regex_3/regex.py.
Hg issue 326: Version is out of sync with PyPI
Version: 2019.3.12
Hg issue 319: Support for a timespan parameter
Added timeout parameter in Python 3. TimeoutError was added in Python 3.3, and as Python 2.7 will soon reach EOL, I'm not bothered about supporting timeouts for Python 2.
Version: 2019.3.9
Hg issue 320: Abnormal performance
Forgot about negative lookarounds! Previous change now applies only to positive lookarounds.
Version: 2019.3.8
Hg issue 320: Abnormal performance
Included firstset from lookaround, where appropriate.
Version: 2019.2.21
Hg issue 316: __version__ no longer accessible via regex.__version__
Version: 2019.2.20
Hg issue 314: Import error: "No module named regex._regex_core"
Was OK for wheels, but not for setup.py.
Version: 2019.2.19
Hg issue 313: test_regex.py ends up in site-packages/test_regex.py
Tidied files away into subfolder in site-packages.
Version: 2019.2.18
Fixed bug in unicode_at_grapheme_boundary.
Version: 2019.2.7
Moved some StateData declarations to make the code conform to the C89 standard.
Version: 2019.2.6
Lookarounds no longer save the repeat data. Lookarounds no longer saves captures if they don't contain any groups.
Version: 2019.2.5
Atomic groups no longer save the repeat data; that proved to be unnecessary.
Version: 2019.2.3
Further improvements to the new code.
Version: 2019.1.24
Hg issue 308: infinite search
Fixed a re-allocation bug.
Version: 2019.1.23
Major overhaul of code to use simpler stacks. The result is now much easier to understand and maintain!
Version: 2018.11.22
Hg issue 304: Unreasonable edge case that used to work
Now moves the minimum number of repeats out of a repeat if it contains a repeat. This allows a repeat guard to be put back, which reduces the chance of catastrophic backtracking.
Version: 2018.11.7
Hg issue 301: TypeError: character mapping must return integer, None or unicode
Fixed bug introduced by broken workflow.
Version: 2018.11.6
Hg issue 300: segmentation fault
Fixed a problem with not recording all fuzzy changes.
Also fixed the check for prefix/suffix in branches: when fuzzy subpatterns were compared, their constraints weren't compared.
Version: 2018.11.3
Hg issue 299
Reworked the fix to perform a normal match and then fall back to a partial match if that was originally requested.
Further changes to match re module's behaviour on zero-width matching for Python 3.7.
Changes to the locations of the source files.
Version: 2017.12.12
Hg issue 268: Update the escape() documentation
Added documentation for escape.
Hg issue 269: Building a bdist using setuptools throws an error.
Moved source files from subfolders to main folders for Python versions.
Version: 2017.12.9
Further changes to match re module's behaviour on zero-width matching for Python 3.7.
Version: 2017.12.5
Hg issue 266: fuzzy match alignment recovery
Added 'fuzzy_changes' attribute to match object to indicate positions of changes in fuzzy match.
Stopped supporting Python 2.5 and Python 3.1-3.2.
Made changes to zero-width matching for Python 3.7.
Version: 2017.11.9
Hg issue 264: Failure to import regex in pypy3-5.8.0
Version: 2017.11.8
Hg issue 264: Failure to import regex in pypy3-5.8.0
Hg issue 265: Invalid pointer in munmap_chunk (core dump) for specific inputs
Version: 2017.9.23
Hg issue 253: Run into error under PyPy 5.8.0
Version: 2017.8.1
Hg issue #240: Unable to build the project from source on OSX with PyPy3.5-5.7.1-beta
Needed to compensate for the differences between CPython and PyPy.
Version: 2017.4.6
Added setup.py for building from the Hg working directory.
Version: 2017.4.6
Hg issue 236: Incorrect references to bugs.python.org issues
Version: 2017.2.8
Failed to build on AIX using xlc because FALSE and TRUE were already #define'd. Replaced enum {FALSE, TRUE} with #define's.
Version: 2017.1.17
Hg issue 230: Is it a bug of (?(DEFINE)...)
Capture groups in (?(DEFINE)...) shouldn't be treated as 'visible' capture groups by .findall.
Version: 2017.1.14
Hg issue 227: Performance trap of (?V1i) flags
Further improvements. It now tries to split full case-folded literals into simple and full case-folded literals where full case-folding isn't needed because simple case-folding is faster.
Version: 2017.1.13
Hg issue 227: Performance trap of (?V1i) flags
Version: 2016.12.27
Hg issue 227: Incorrect behavior for ? operator with UNICODE + IGNORECASE
'end_pos' wasn't always initialised in 'locate_required_string'.
Version: 2016.11.21
Hg issue 226: Error matching at start of string
Fuzzy matching of zero-width items wasn't quite right.
Version: 2016.11.18
Hg issue 225: BESTMATCH in fuzzy match not working
Version: 2016.10.22
Hg issue 221: Got an exception using PyPy
pypy2-v5.4.1 appears not to support the buffer protocol on bytestrings (str in Python 2), so added code to handle them analogously to how unicode strings are handled.
Version: 2016.10.20
Make setup use setuptools if it's available.
Version: 2016.10.12
Hg issue 221: Got an exception using PyPy
Added check for error in fold_case which, I hope, will reveal the exception that it's not reporting.
Version: 2016.9.22
Hg issue 220: Misbehavior of group capture with OR operand
Repeats should not be factored out of branches because a branch should be exhausted before trying the next one.
Version: 2016.9.13
Also supported on Python 3.6.
Version: 2016.8.27
Hg issue 219: Unicode word boundries
For a Unicode word boundary (UAX #29), apostrophe in rule WB5a should include both U+0027 (APOSTROPHE) and U+2019 (RIGHT SINGLE QUOTATION MARK /
* curly apostrophe).
Version: 2016.7.21
Hg issue 217: Core dump in conditional ahead match and matching \! character
Fixed bug where it incorrectly tried to restore saved groups when none saved.
Version: 2016.7.14
Hg Issue 216: Invalid match when using negative lookbehind and pipe
The creation and position of a branch firstset wasn't always correct.
Version: 2016.6.25
Updated to support Unicode 9.0.0.
Version: 2016.6.19
Hg issue 214: tests failure when using python debug flavor
Further tweaks re error handling.
Version: 2016.6.14
Hg issue 213: Segmentation Fault
Info about atomic groups wasn't pushed properly.
Version: 2016.6.5
Hg issue 212: Unexpected matching difference with .*? between re and regex
In 'add_repeat_guards', it wasn't propagating the status from the tail back towards the head across default nodes, so the guards weren't always correct.
Version: 2016.6.2
Hg issue 211: Segmentation fault with recursive matches and atomic groups
It wasn't saving the call frame and then restoring it when backtracking out of atomic groups.
Version: 2016.5.23
Hg issue 206: Incompatible with re if single { in the pattern
Brought the regex module's handling more in line with the re module. It now more readily treats an invalid fuzzy constraint as a literal.
Version: 2016.5.15
Hg issue 208: Named list, (?ri) flags, Backreference
Fixed more issues with the alignment of text in a buffer when using named lists.
Also changed how a compiled regex is pickled to use a bytestring for the packed code list. (It was actually pickling an unpacked list of ints.)
Version: 2016.5.14
Hg issue 205: Named list and (?ri) flags
string_set_match_ign_fwdrev wasn't taking into account that it fills the folded buffer from the end when searching in reverse.
Version: 2016.5.13
Hg issue 204: confusion of (?aif) flags
The FULLCASE flag is now ignored if the ASCII flag is turned on.
Version: 2016.4.25
Hg issue 203: partial matching bug
The text position wasn't always set correctly before returning the status.
Version: 2016.4.16
Hg issue 201: ENHANCEMATCH crashes interpreter
Fixed an issue with restoring group captures.
Version: 2016.4.7
Hg issue 199: Segfault in re.compile
Removed copies of groups that weren't called.
Version: 2016.4.3
Hg issue 197: ValueError in regex.compile
Hg issue 198: ValueError in regex.compile
It wasn't catching ValueError and then raising regex.error.
Version: 2016.4.2
Hg issue 196: Fuzzy matching on repeated regex not working as expected
Also reduced memory usage of pickle data.
Version: 2016.3.31
# Hg issue #194: .FULLCASE and Backreference
Capture groups failed to match when using full case folding because of a bug in the handling of "I", which needs to be treated specially to cope with the Turkic I).
Version: 2016.3.26
Hg issue #193: Alternation and .REVERSE flag.
The firstset before a branch was at the wrong end for a reverse pattern.
Added back some tests that were accidentally omitted.
Version: 2016.3.24
Hg issue 192: Named lists reverse matching doesn't work with IGNORECASE and V1
string_set_match_fld_fwdrev wasn't taking into account that it fills the folded buffer from the end when searching in reverse.
Version: 2016.3.2
Hg issue 190: Regression? Neverending regexp when upgrading to latest version.
The fix for Hg issue 187 wasn't quite right. Managed to remove recursion entirely from 'add_repeat_guards'.
Version: 2016.2.25
Hg issue 188: Crash during search
Stopped calling 'try_match' recursively when the tail of a branch is a branch. Increasing alternatives to 40000 caused the stack to overflow.
Version: 2016.2.24
Hg issue 187: Crash on Anaconda Python if large number of pattern
Remove the recursion in 'use_nodes'. Increasing alternatives to 50000 caused the stack to overflow.
Version: 2016.2.23
Hg issue 187: Crash on Anaconda Python if large number of pattern
Reduced the amount of recursion in 'add_repeat_guards'. The large number of alternatives (25154) caused the stack to overflow.
Version: 2016.1.10
Hg issue 177: Build fails on pypy (OS X 10.11, clang)
Stripped out the #if...#endif that was added for Hg issue 135, which is no longer needed by more recent versions of PyPy.
Version: 2015.11.22
Hg issue 180: bug of POSIX matching
Fixed bug where the groups aren't always correct with POSIX matching.
Version: 2015.11.14
Hg issue 172: Performance of V1 mode
Fixed bug where RE_FLAG_FULLCASE not turned off when RE_FLAG_IGNORECASE turned off, leading to omission of first set.
Version: 2015.11.12
Hg issue 171: Weird performance of V1 mode
Fixed bug where RE_FLAG_FULLCASE not turned off when RE_FLAG_IGNORECASE turned off for required string, leading to required string not being used.
Version: 2015.11.9
Hg issue 169: Performance
Further tweaks to fuzzy matching.
Version: 2015.11.8
Hg issue 167: Performance of Backreference
No longer saves/restores groups or repeats around a lookaround if it doesn't contain any.
Version: 2015.11.7
Hg issue 166: Performance
Improved the performance of fuzzy matching.
Version: 2015.11.5
Hg issue 165: Performance / hung search
Made changes to fuzzy matching code, including refactoring different kind of fuzzy matching (exact/simple/enhanced/best).
Version: 2015.10.28
Hg issue 163: allow lookarounds in conditionals
Added support for a lookaround in a conditional pattern, e.g. r'(?(?=\d)\d+\b|\w+)'.
Version: 2015.10.22
Hg issue 161: Unexpected fuzzy match results
Fixed the bug and did some related tidying up.
Version: 2015.10.5
Hg issue 158: Group issue with (?(DEFINE)...)
The groups and repeats weren't restored properly when a lookaround completed and it contained a group call.
Version: 2015.10.1
Hg issue 157: regression: segfault on complex lookaround
Nested lookarounds/atomic groups didn't restore state correctly.
Version: 2015.9.28
Hg issue 156: regression on atomic grouping
It didn't initialise min_width when building the atomic group.
Version: 2015.9.23
Hg issue 154: Segmentation fault 11 when working with an atomic group
Version: 2015.9.15
Hg issue 150: Have an option for POSIX-compatible longest match of alternates
Added POSIX matching (leftmost longest).
Version: 2015.7.20
Hg issue #147: Fuzzy match can return match points beyond buffer end
It wasn't checking for the edge of the text when case-folding and it was also advancing even when case-folding had failed.
Version: 2015.7.12
Hg issue #146: Forced-fail (?!) works improperly in conditional
The capture groups weren't being cleared before retrying after failure.
Version: 2015.6.24
Hg issue #144: Latest version problem with matching 'R|R'
The prefix of a set of branches was omitted.
Version: 2015.6.21
Hg issue #143: Partial matches have incorrect span if prefix is '.' wildcard
Didn't set state->match_pos if search_start returned a partial match status.
Version: 2015.6.19
Updated to Unicode 8.0.
Some performance tweaks.
Version: 2015.6.15
Removed a few lines that should've been removed in the previous fix!
Version: 2015.6.14
Fixed a bug where it could sometimes search for the same required string multiple times.
Version: 2015.6.10
Hg issue #141: Crash on a certain partial match
It didn't check the result of 'try_match' correctly in certain places (the status returned isn't limited to success and failure).
Version: 2015.6.9
Hg issue #140: Replace with REVERSE and groups has unexpected behavior
subx needed to add the template items to the list in reverse order when searching backwards because the list will be reversed after completion.
Version: 2015.6.4
Hg issue #98: regex module is not thread safe because of _cache
Now iterates over a snapshot of the cache keys in case the dict resizes.
Version: 2015.6.2
Hg issue #139: Regular expression with multiple wildcards where first should match empty string does not always work
The problem was caused by a negative-character in the firstset, eg "[^a]".
Fixed.
Version: 2015.5.28
Hg issue #137: Posix character class :punct: does not seem to be supported
It _is_ supported.
Corrected Posix-style properties for 'alnum', 'digit', 'punct' and 'xdigit' which are different from that of Unicode. Now also available as \p[posix_alnum}, etc.
Hg issue #138: grapheme anchored search not working properly
Fixed.
Version: 2015.5.10
Hg issue #136: Use a DFA when possible
Made a slight tweak so that it now treats an optional subpattern like a repeated subpattern (e.g. "(?:xyz)?" -> "(?:xyz){0,1}") to make use of the repeat guards.
The advantage over DFA is that it'll work even for those patterns that aren't compatible with DFA.
Version: 2015.5.7
Hg issue 135: PyPy Support (with patch)
It should now build on PyPy.
Version: 2015.3.18
Hg issue 133: support for captures() in expandf().
Now supported in expandf and subf.
Issue 23692: Undocumented feature prevents re module from finding certain matches
This also applied to regex which failed to take into account group references by group capture testa, e.g. "(?(1)...|...)", when guarding against excessive repeats.
Version: 2014.12.24
Hg issue 132: index out of range on null property \p{}
It's now reported as an unknown property.
Version: 2014.12.15
Hg issue 131: nested sets behaviour
The set difference operator '--' wasn't handled correctly after an implicit set union.
Version: 2014.11.14
Unreported issue: no such builtin as 'ascii' in Python 2. Fixed.
Version: 2014.11.13
Hg issue 127: Infinite loop is found
Not an infinite loop, but slow because of repeated backtracking on a very long chunk of text.
This fix reduces the amount of backtracking and re-matching.
Version: 2014.11.3
Hg issue 125: Reference to entire match (\g<0>) in Pattern.sub() doesn't work as of 2014.09.22 release.
Version: 2014.10.24
Reverted licence.
Version: 2014.10.23
Fixed bug in determining line number in regex.
Changed licence to Apache License 2.0 and included copy in release.
Version: 2014.10.9
Issue 22578: Add additional attributes to re.error
Added the attributes .msg, .pattern, .pos, .lineno and .colno to the regex error class.
Version: 2014.10.7
Fixed bug in partial matching when required string occurs after repeat.
Version: 2014.9.22
Issue #22437: Added support for referring to a group by number using (?P=...). This is in addition to the existing \g<...>.
Fixed bug in handling of cache for locale-sensitive patterns.
Version: 2014.9.18
Adjusted line-endings in PKG-INFO.
Version: 2014.8.15
Hg issue 115: Infinite loop when processing backreferences
Version: 2014.6.28
Updated to Unicode 7.0.
Version: 2014.5.23
Fixed fuzzy counts that were wrong when using BESTMATCH or ENHANCEMATCH flags.
Version: 2014.5.17
Hg issue 112 in mrab-regex-hg: re: OK, but regex: SystemError
Version: 2014.4.10
Hg issue 102: Partial matches
Hg issue 109: Edit distance of fuzzy match
Added partial matches.
Added .fuzzy_counts attribute to match objects.
Version: 2014.2.19
Unicode properties sometimes failed to match when the IGNORECASE flag was set.
Version: 2014.2.16
Hg issue 108: Fails to build from source on s390x
Version: 2014.1.30
Hg issue 106: * operator not working correctly with sub()
Made to conform more to re module in version 0 behaviour.
Version: 2014.1.20
Hg issue 105: FAIL: test_case_folding (__main__.RegexTests) with py2.7
data.fold_len wasn't initialised.
Also, deleted some unused functions.
Version: 2014.1.10
Issue #17087: Improve the repr for regular expression match objects
Version: 2013.12.31
Hg issue 101: findall() broken (seems like memory corruption)
state->req_pos wasn't initialised when the required string is an initial string.
Version: 2013.11.29
Hg issue 100: strange results from regex.search
Version: 2013.10.25
Hg issue 98: regex module is not thread safe because of _cache
Version: 2013.10.24
Further fixes for Hg issue 96.
It should now use correctly use sets for named lists, except when there's fuzzy matching.
Version: 2013.10.23
Hg issue 96: compile '\L<...>' with 'i' flag was very slow
Version: 2013.10.22
Hg issue 95: 'pos' for regex.error
Version: 2013.10.21
Python crashes when executing regex updates pattern.findall
Version: 2013.10.12
Updated to Unicode 6.3.
Version: 2013.10.4
Issue #18468: re.group() should never return a bytearray
Applies to Python 3.4 and later for compatibility with the re module.
Also, some performance improvements.
Version: 2013.8.4
Update for Python 3.4a1 release.
Version: 2013.6.26
Performance improvements.
Version: 2013.6.5
Hg issue 92: running the following regex causes a segfault
Version: 2013.5.21
Hg issue 91: match.expand is extremely slow
Also tidied up code.
Version: 2013.3.11
Hg issue 89: Certain regexes extremely slow compared to re module
Disabled one of the optimisations that appears to cause performance problems.
Version: 2013.2.22
Fixed issue with LOCALE flag not working properly.
Version: 2013.2.16
Fixed a locale-specific test. Whether b'\xE0' is a word character depends on the locale.
Version: 2013.1.26
Another fix for Hg issue 87: Allow duplicate names of groups
It didn't correctly handle a name group within a group of the same name.
Version: 2013.1.25
Second attempt to fix
Version: 2013.1.24
Hg issue 86: Enhance API of captures() to enable retrieval of ALL groups at once, as a dictionary
