Beast development history recovery

[Tim Janik]

hey All.

the upstream Beast SVN repository, used since the first CVS->SVN migration
of the GNOME project (2006-07-13) was lost, due to a combination of unfortunate
events and omissions which would have been hard to predict/anticipate.

the repository used to store development history from 2006-07-15 (apparently
as SVN revision 3831) to 2006-12-28 (SVN revision 4206), round about 376
commits.

since last friday, i've done my best at reconstructing the development
history of the last half a year, using:
- SVN development checkouts from Stefan Westerfeld and myself
   (19 tarballs, including .svn/ history of single revisions)
- release tarballs (0.7.1-rc1 and 0.7.1)
- buildbot checkouts (r4157 - r4201), complete revisions with .svn/ histories
- miscellaneous infrequent backups of my development machine
   (23 tarballs, random snapshots without .svn/ informatiion)
- the svn-commit-list mailing list archive from svn.gnome.org:
     http://beast.gtk.org/mirror/scm-history/svn-commit-list-2006-Q3Q4.gz
   some of the patches are partially or broken, some are missing, the
   vast majority of the history is contained herein though.
   in particular, the list contains:
   * commit messages and patch for allmost all of revisions 3831 - 4160.
     missing revisions are: 3844 3919 3940 4007 4056 4091 4092 4093 4139
   * commit messages without patch due to list-config changes for revisions
     4161 - 4206.
   some patches are partial or don't apply readily due to brief SVN messages
   about copied or deleted files. also it seems like some smaller initial
   commits are missing between the completion of the CVS->SVN migration and
   the first recorded commit (r3831).

we are actually increadibly lucky with this diff/snapshot combination,
it covers most of the development history, including files missing from
the mailing list patches and binary file additions.
in the end, using a combination of the various intermediate tree snapshots
and forward/reverse patch sequences approximating the missing commits, this
allowed an allmost complete restauration of the development history.
in less than a dozen places, the history may be mildly inaccurate, meaning
multiple commits had to be squashed (e.g. around the 4091/4092/4093 hole),
or occasionally a committed file had to be time-warped because the exact
introduction date couldn't be determined anymore (mostly for po/ updates).
that means, about 97% of the history was fully restored and in particular,
the last ~50 revisions are near perfect because of the buildbot revision
checkouts (these include the recent release with security related fixes).
compared to the original SVN db, those last commits only miss some svn:ignore
proeprty values and CRLF->LF changes introduced by the second
CVS->SVN migration (2006-12-29).
author credits (svn blame) should be accurate throughout all the history,
and the newly craeted beast-mingw branch also could be fully recovered.

i'll try to get the sysadmins to put the new beast SVN repo back into place
today, and then we can resume beast project work at full steam ;)

---
ciaoTJ
_______________________________________________
beast mailing list
beast@...
http://mail.gnome.org/mailman/listinfo/beast