2.2b1
- add a new line item in the PatchSet output which indicates which 
  branches fork off from after this PatchSet.  This is actually
  the only way to correctly indicate the branching.  The 'Ancestor
  Branch' thing is actually broken by design, and is now deprecated.
- sometimes cvs log has a timezone in it (from Ludwig Nussel)
- better buffer size handling

2.1
- lots of fixes to get the thing working sufficienly for cvs2git
- main features
  - optionally detect and display ancestor branch information for
    new detected branches
  - never allow more than one revision of a file to be in a single
    patchset (this may disturb the patchset numbers for historical
    data :-(
  - use file revisions to help sort, so that initial imports
    won't be listed as patchset 1, with patchset 2 as the actual
    creation of the files.
  - reports log times in localtime (use TZ to override)	
- many downstream patches from Debian, special thanks to Marcus
  Crafter for maintaining.  From the Debian changelog:
   - patch from Kim Hansen to fix time calculation bug
   - documentation patch from J. Bruce Fields
   - directory name handling patch from David Kilzer
   - others I'm not sure of, I'd happily update the credit here

2.0rc1
- resolve -r symbols during global symbol resolution
- using two -r arguments implicitly sets -b with branch
  from second -r (if you can speak cvsps arguments, you'll
  understand this)
- fix using cvsps against sourceforge where the actual
  server repository path has many instances of the project
  name in it and the 'strstr' was finding the wrong one.
  instead use the equiv. of 'strrstr'

2.0b10
- fix bug where filenames with 'bad' characters (such as ' ')
  were being sent unescaped to a shell
- add --no-cvs-direct and -Z 0 to disable cvs-direct and compression
- extend the 'fuzz-factor' concept by keeping a min/max date
  encountered for each patchset and fuzz on this interval,
  rather than fuzzing on the date of the first encountered
  member only, helps for handling REALLY slow commits where a
  HUGE number of individual files were modified at a slow trickle.
- it turns out that old versions of cvs have two problems
  1) they don't support 'rlog' (handled)
  2) they don't support 'version', which caused the workaround
     for 1) to fail.
  Now cvsps detects cvs versions too old to support version and/or
  rlog.  Additionally, if server is too old, cvs-direct is
  simply disabled.
- parse the rc file first so that the command line will override
- add -q to shut-up the warnings

2.0b9
- remove Mbinary from valid-response list - can't handle it
- much more (pedantic?) clean-up/close path in cvs_direct
  to fix all of the lost data and hanging problems at close
- added rlog capability to cvs_direct, switched main code
  to use it.
- added version capability to cvs_direct, now cvsps is 
  completely independant of cvs binary for all operations!
- added --root and repository command-line arguments,
  now cvsps is completely indepentant of having working
  directory!

2.0b8
- fix the cvs_rupdate calls to use the real repository
  instead of the mangled use_rep_path.  This broke
  cvs_direct generation of add/remove diffs
- fix the sed expression to account for the difference
  between +++ and --- in the first and second lines.
  This broke remove diffs for direct and external.
- attempt to fix a hang-at-flush bug in cvs-direct
  compression handling

2.0b7
- implement compression in cvs_direct
- add command line -Z to specify compression level.
  applies to cvs-direct and external cvs
- implement 'cvs_diff' in cvs-direct
- use cvs_diff instead of cvs_rdiff because it allows
  us to specify all of the lovely diff options
- change system to my_system so ctrl-c will get
  through.  indirectly, this is necessary because
  cvs has loused up the exit codes with the 'cvs diff'
  command, making it impossible to determine whether
  the program exited abnormally
- add parsing of ~/.cvsps/cvspsrc file at startup
  where command line arguments can be listed, one
  per line (parameters to arguments on the same line
  as the argument). --norc prevents this

2.0b6
- fix bug where if -r was specified along with other
  filter options, the other filter options were
  applied first, and the start/end tags were
  never found.
- change -f from strstr to regex
- change regex to EXTENDED regex
- add the --cvs-direct option

2.0b5
- add a cvs_check_cap for capability checking. 
- use cvs_check_cap to check whether rlog is 
  supported.  it's not supported in 1.11 and  
  before. warn if non-supported version is 
  detected
- add --no-rcmds to manually disable rlog and rdiff
- add a fix for some bizarre inconsistencies found
  in some random repositories, such as tagged
  revisions that don't exist.
- remove the adaptive crap filter and bk_log_border
  logic now that Larry has cleaned up the exported
  BK->CVS trees.

2.0b4
- add the bk_log_border parsing logic to handle the
  case where 'cvs log' text was committed into the log
- add 'adaptive crap filter' logic to handle all
  of the different corruptions of the cvs log file
- switch to rdiff instead of diff to not require working 
  files to be checked out.  Note: when --diff-opts are
  specified, diff must be used (not rdiff) because
  rdiff doesn't support options(?!!)

2.0b3
- add the --diff-opts for explicitly setting the diff 
  options.  important for setting increased context for 
  example. 
- use 'cvs rlog' instead of 'cvs log'.  important if 
  there are empty directory pruned with cvs update -P
  for getting consistent patchset numbering
- clean up patch_set_affects_branch for handling INVALID
  tags and the -r especially around a branch point
- add the bkcvs proof-of-concept hack


2.0b2
- change the way INVALID tags are reported - show the exact
  patchsets and revisions that conflict, and don't print 
  anything for 'funky' tags/revisions
- add psid to the patchset structure and pass through tree
  once just to assign the numbers.  this allows reporting
  psid when problems are found (above)
- more code restructuring - move stats to own module
- commenting cleanups
- include the very rudimentary merge_utils.sh functions
  I use to assist merging stuff
- change the way diffs are generated, handle individual
  members that are before/after the -r tag restrictions
- change the semantics of the -s argument.  it now only
  restricts the patch sets the same as the other filtration
  arguments.  new argument -g turns on diff generation
- add manpage section discussing tag handling semantics

2.0b1
- major restructuring of source code.  single cvsps.c split into
  a few modules
- major rework of the data structures in order to support the most
  frequently requested, and most desired feature: -r.  cvsps
  can now associated symbolic tags with specific patchsets,
  branches are now associated with the patchset instead of 
  the individual revision.  you can view patchsets committed
  after a given symbolic tag, or between two tags.
- changed the handling of same author, same log message commits
  on different branches. this now creates multiple patchsets
- changed the handling of 'file xyz initially added on branch xyz'
  log messages, they now create hidden patchsets.
- these last two items have made the patchset numbering scheme
  incompatible with the old scheme.  patchsets are going
  to change numbers.

1.99.1 (not released)
- redid the data structures a bit to 'normalize' the references to 
  a file revision.  this allows the program to detect the case
  when a file is added on a branch, and generate the right output
  for 'cvsps -s' in this case (frequent bug report)
- added the '-l' option to restrict patchsets based on log descr
  content. (patch from Geoff Soutter).  added man page for it.
- added the --summary-first option suggested by Andi Kleen to
  have the PatchSet summary information for all patchsets at
  the head of the output, when multiple patchsets are given
  to the -s option.
- added the all: target to Makefile.  (patch from Ben Elliston)
- use a single centralized cvsps.cache file per root/repository
  this obviates the need to 'cvsps -u' in multiple checked out
  trees of the same repository (patch adopted from Baruch Even)
- change the magic name TRUNK (for -b option) to HEAD to be 
  consistent with cvs.  (patch from Henrik Nordstrom)
- when parsing 'cvs log' output, look for the exact strings
  that separate the log entries and the files, instead of 
  just the first 8 characters.  this allows a wider variety
  of stuff to be in the log format (including actual patchsets).
  Inspired by Andy Isaacson.
- added the -p option allowing patchset output to go into
  individual files in a named directory.  Based on a patch
  by Henrik Nordstrom.  This is a great idea.  Thanks Henrik.

1.3.3 (why isn't this 1.4 again?)
- added the 'memory reduction' patch from Baruch Even.  This patch
  greatly reduces the memory footprint of CVSps against a large 
  repository by using dynamically allocated buffers, and by using
  a tree to hold "common" strings (to avoid having a thousand copies
  of the string "1.1").  This patch also adds a new option '-t' which
  displays some memory usage statistics at the top.

- added the 00-strip-revision.patch from Steven Tweedie which fixes
  a parsing problem when there are locked files in the repository

- added the 'Tweedie Tweenie' patch from Steven Tweedie (01-stable-tree) 
  which fixes a bug where interspersed commits could cause unstable 
  tree behavior.  This would happen especially when multiple users commit 
  at the same time to different parts of a large tree over a slow link.

- added the --norc option (based on a suggestion by Soren S. Jorvang) 
  to handle cases where people have bad stuff in their .cvsrc which makes 
  cvsps fail.  I could have added the '-f' unconditionally to the command
  line for cvs when run under cvsps, but somehow that seemed dangerous.

1.3.2 (small feature release)
- added the 'multi-patchset' feature to the -s option, provided in its
  entirety by Daiki Ueno <ueno@unixuser.org>

1.3.1 (bugfix release mostly)
- fix bug with updating cache (-u) having to do with matching new and old 
  revisions

- fix timestamp_fuzz_factor bug where the fuzz was applied to loading
  from cache by mistake.

- add a spec file (from Jan IVEN <Jan.Iven@cern.ch>)

- improve the parameterization of Makefile, and make things relocatable
  (from many people, esp. Amitai Schlair <schmonz@schmonz.com>)

- fix strip_path_len calculation (again - I had munged the prior fix)
  Jeffrey Ebert <ebert@sonicsinc.com>


1.3
- fix case where CVSROOT is initial substring of Repository when
  creating strip_path variable.

  Thanks to Jean-Michel Rouet <jean-michel.rouet@philips.com> for
  this fix.

- fix case where a subdirectory has a different CVSROOT or Repository
  path from the toplevel directory (happens when one project is checked
  out inside another projects subdirectory).

  Thanks to Philippe M. Chiasson <gozer@cpan.org> for the fix.


1.2

- Moved cache file location to CVS/ subdirectory.  This is where CVS keeps
  it's meta-data, and putting the cache there keeps it out of the way of
  other CVS operations. (thanks to Amitai Schlair for suggestion)

- Added manual page.  (thanks to Amitai Schlair for suggestion)

- Fixed the '-b' option to take a special branch name 'TRUNK' which
  will restrict the output to patchsets made on the main trunk.


1.1 Initial public release.
-------- end of changelog ---------