2006-04-09 Jarkko Hietaniemi <jhi@Chicxulub.local>

• Try to underline, highlight, and explain the fact that String::Approx does not do a good job for comparing strings "with fuzz", use the Levenshtein et al for that.
• aindex() might return "too early" indices if either the pattern or the text contain repetitive characters, this seems to be a tricky defect to fix and somewhat conflicting with our model (return "as early as possible" matches, just like regular expressions) (to get "as late as possible" matches one would basically have to keep retrying at later indices until one fails), so for now only document this known problem. The same goes for aslice().
• [INTERNAL] use Test::More (not 100% yet)
• [INTERNAL] apse.c: do not reset text_position_range in apse_reset (thanks to Helmut Jarausch)
• [INTERNAL] apse.c: add #include <assert.h>.
• [INTERNAL] apse.c: small signed/unsigned cleanups.
• Released as 3.26.

2005-05-24 Jarkko Hietaniemi <jhi@iki.fi>

• Pure documentation cleanup release to address http://rt.cpan.org/NoAuth/Bug.html?id=12196 "Small String::Approx Pod issue"
• Released as 3.25.

2005-01-02 Jarkko Hietaniemi <jhi@iki.fi>

• Pure documentation cleanup release to address http://rt.cpan.org/NoAuth/Bug.html?id=6668 "Unfortunate perldoc rendering of String::Approx"
• Released as 3.24.

Mon Nov 30 15:18:15 2003 Jarkko Hietaniemi <jhi@iki.fi>

• Safeguards against trying to use greater edit distance than the pattern is long, inspired by 'idallen'.
• Advise against using String::Approx for text comparisons, since String::Approx is meant for strings.
• Released as 3.23.

Sun Oct 19 12:17:20 2003 Jarkko Hietaniemi <jhi@iki.fi>

• adistr($pattern, @inputs) returned absolute, not relative, distances in list context, reported by 'idallen'.
• Released as 3.22.

Sat Oct 18 10:29:30 2003 Jarkko Hietaniemi <jhi@iki.fi>

• asubstitute() didn't substitute in $_ as promised, reported by Tim Maher.
• Released as 3.21.

Mon May 12 22:09:59 2003 Jarkko Hietaniemi <jhi@iki>

• Bug report, analysis, and patch from Rich Williams for a nasty segfault (no easy test case, sadly).
• Clarify the documentation about the weirdness of the 'size' from aslice().
• Released as 3.20.

Sun Aug 18 01:37:53 2002 Jarkko Hietaniemi <jhi@iki.fi>

• Fixed a frontend bug which caused 0% to be rounded up to 1, (instead of being equal to 0) found by Ji Y Park.
• Released as 3.19.

Tue Oct 16 20:06:53 2001 Jarkko Hietaniemi <jhi@iki.fi>

• Documented how one can use the adist() and adistr() to sort the inputs according on their approximateness, suggested by Arthur Bergman.
• Fixed yet another aindex() bug found by Juha Muilu.
• Pasha Sadri figured out what was the stupid bug breaking complex cache flushing: I was calling a wrong non-existent sub: cacheflush_complex was called instead of flushcache_complex. Argh.
• Released as 3.18.

Fri Aug 3 15:47:10 2001 Jarkko Hietaniemi <jhi@iki.fi>

• 3.17: Add COPYRIGHT and LICENSE to the pod.

Thu Jun 28 23:50:14 2001 Jarkko Hietaniemi <jhi@iki.fi>

• Released as 3.16.
• Ben Kennedy found yet another silly memory leak, in Approx.pm:complex() this time.

Sat Feb 10 04:18:09 2001 Jarkko Hietaniemi <jhi@iki.fi>

• Rewrote _simple() and _complex(), now the silly memory leak in the caching code should be gone, the problem was reported at least by Frank Kirsch, Dag Kristian Rognlien, Chris Rosin, and Ben Kennedy.
• There were actually two memory leaks, the other one was mild enough not to be noticed. While fixing the mild one found a long standing bug, fixing it made asubst() to matches to greedy again (they were always stingy because of the bug).
• Fixed a bug in adist() (actually, aslice(), which adist() internally uses) which made adist('MOM','XXMOXXMOXX') to return 0, not 1, reported by Chris Rosin.
• Fixed a bug in aindex() where aindex('----foobar----', '----aoobar----', 1) would report 10 instead of 0, reported by Trank Tobin.
• Fixed a bug in aindex() where reported by Damian Keefe.
• Added an API to control the size of the cache(s), as suggested by B. Elijah Griffin.
• Added Aldo Calpini's test failure in Win32 as an open bug. Don't know what's wrong yet, the test output looks garbled.
• Internal: use sv_len() instead of SvCUR() to be Unicode-safe.

Thu Nov 16 22:35:13 2000 Jarkko Hietaniemi <jhi@iki.fi>

• Fixed the apse.c:apsematch() for loop bug reported by Ross Smith and Eric Promislow.
• Add adist() and adistr() which tell the edit distance and the relative edit distance. For both zero is an exact match, and negative means that the pattern is longer than the input. For the relative edit distances 1 means that the strings are completely different, and the negative values never reach -1. This feature has been requested by zillions.
• Add arindex() (a backward aindex()).
• Add BUGS list. Two open bugs.

Wed Sep 6 19:23:55 2000 Jarkko Hietaniemi <jhi@iki.fi>

• Released as 3.13.
• apse.c: fixed an aindex() bug where a better match doesn't move the match_begin forward. Reported by David Curiel and Stefan Ram.
• Approx.pm: fix a cache flushing bug where the most current pattern was accidentally flushed, fix from Ben Kennedy.

  Revert back to using real words: "approximateness" it is (drop the "approximity" coinage).

Sun Apr 30 22:39:09 2000 Jarkko Hietaniemi <jhi@iki.fi>

• Released as 3.12.
• ChangeLog: Explain the bug fixed in 3.10 a little bit more.

Mon Apr 24 23:19:20 2000 Jarkko Hietaniemi <jhi@iki.fi>

• Makefile.PL: The -DAPSE_DEBUGGING was accidentally left on, resulting in a deluge of debug messages being spewed to stderr. Oops. Reported by Rob Fugina.
• Approx.pm: Bump to 3.11.

Tue Apr 18 06:48:34 2000 Jarkko Hietaniemi <jhi@iki.fi>

• Released as 3.10.
• apse.c: Fix an insidious buffer overrun, found by heroic debugging by J. D. Laub (already back in November 1999, but I'm amazingly good at procrastinating). This is probably the bug that plagued String::Approx 3.0* in HP-UX and Solaris. I couldn't anymore repeat it in HP-UX, in Solaris it was still going strong.

  The bug was really well camouflaged: it didn't overstep the buffer area as a whole, in which case it would have been found by Purify-like tools. What was happening was that the code is maintaining a 2-dimensional bitmask (or a 3-dimensional bitbuffer, if you will), and under certain circumstances the bits from one dimension leaked into another. The only user-visible API that was tickling the bug was aslice(), so if you weren't using that, you were probably okay. In Solaris (and at some point HP-UX) the malloced buffers happened to be laid out so that bug was visible (e.g. in Linux, AIX, and Digital UNIX, it wasn't).

Mon Jul 19 10:42:13 1999 Jarkko Hietaniemi <jhi@iki.fi>

• Approx.pm: Released 3.09.
• Approx.pm: Do the right things if pattern is empty (the bug found by Anirvan Chatterjee).
• apse.c: Fixed the cases "this-has-so-many-edits-that-it-isgoing -to-match-no-matter-what" and "there-are-not-enough-editsto -ever-match-this" (the latter bug found by Mitch Helle).

Thu Jun 24 02:00:26 1999 Jarkko Hietaniemi <jhi@iki.fi>

• Approx.pm: Released 3.08.
• Approx.pm: Add aslice() as suggested by Mike Hanafey.
• apse.h: Introduce use_minimal_distance.
• apse.c: Implement use_minimal_distance.
• t/aslice.t: Test the new feature.
• MANIFEST: Add t/aslice.t.

Wed Jun 23 20:07:05 1999 Jarkko Hietaniemi <jhi@iki.fi>

• Approx.pm: Released 3.07.
• Approx.pm: Add aindex() as suggested by Mike Hanafey.
• apse.h: Introduce text_initial_position and text_final_position.
• apse.c: Implement text_initial_position and text_final_position.
• t/aindex.t: Test the new feature.
• MANIFEST: Add t/aindex.t.

Wed Jun 16 14:19:39 1999 Jarkko Hietaniemi <jhi@iki.fi>

• Approx.pm: Released 3.06.
• Approx.pm: Release 3.06: Fixed a bug in caching of parsed parameters (the absolute length of the pattern must be cached, too). (Reported by Chris Rosin and Mark Land)
• Approx.pm: Fixed a couple of typos and introduced the coinage 'approximity' (to replace the clunky 'approximateness').
• t/user.t: Added four new tests from Rosin and Land.

Fri Jan 8 16:09:52 1999 Jarkko Hietaniemi <jhi@iki.fi>

• Approx.pm: Release 3.05. No functional changes or bug fixes (no bugs reported), just a release that includes both the Artistic License and LGPL.

Wed Dec 30 11:07:28 1998 Jarkko Hietaniemi <jhi@iki.fi>

• MANIFEST: Added Artistic and LGPL, reworded COPYRIGHT slightly to comply.

Thu Dec 17 13:17:58 1998 Jarkko Hietaniemi <jhi@iki.fi>

• Approx.pm: Released 3.04.
• Approx.pm: Fixed a parameter parsing bug: "i 1" was not accepted. (Reported by Bob J.A. Schijvenaars)
• Approx.pm: Documented that matching is asymmetric: the inputs are matched against the pattern, not the other way round.
• Approx.pm: Added a few C<>s and I<>s to the pod.
• Approx.pm: Added an automatic flush to the pattern compilation caches (triggered by a high water mark).
• Approx.pm: Added "require 5.004_04;", Previously only Makefile.PL had this, but better be paranoid.

Wed Dec 16 10:28:07 1998 Jarkko Hietaniemi <jhi@iki.fi>

• Approx.pm: Released 3.03.
• Approx.pm: Added confirmation from Udi Manber to README: it's not a problem that I have looked at agrep code. My code is my code and can be used within the limits set by my copyright.
• Approx.xs: Removed a lot of the glue code because it is not yet reachable via Approx.pm. Later.

Wed Dec 16 01:23:58 1998 Jarkko Hietaniemi <jhi@iki.fi>

• Approx.pm: Released 3.02.
• README: Added a clarification about our relationship with agrep. There is no common code. None. Waiting for confirmation from Udi Manber. (The concern raised by Slaven Rezic).
• Approx.xs: fixed the PL_na mess for now, will have to figure out the correct way. Tested under 5.004-maint, 5.005-maint, and 5.005-devel.
• Approx.pm: Released 3.01 (actually already Dec 15).