> -----Original Message-----
> From: Hyrum K. Wright [mailto:hyrum_wright_at_mail.utexas.edu]
> Sent: woensdag 17 december 2008 16:14
> To: dev_at_subversion.tigris.org
> Subject: Notifications missing items with 'svn merge'
> Just a minute ago, I tried to bring the 1.5.x-build-fixes branch up-to-
> date with
> the 1.5.x branch (to pull in the fixes to merge test 106). Using a
> build from this morning (r34764), I observed the following:
> hwright_at_spock:~/dev/svn-1.5.x-2$ svnd merge ^/branches/1.5.x
> --- Merging r34739 through r34764 into '.':
> U STATUS
> Hmmm...that's funny, no changes to merge_tests.py? I wonder what
> status reports:
> hwright_at_spock:~/dev/svn-1.5.x-2$ svnd st
> M .
> ? msg
> ? p.patch
> M STATUS
> M www/images/subversion_logo-200x173.png
> M www/images/subversion-diagram.png
> M www/images/subversion_logo_hor-468x64.png
> M www/images/subversion_logo-384x332.png
> M COMMITTERS
> M notes/tree-conflicts/scratch-pad.txt
> M subversion/include/private
> M subversion/libsvn_wc/adm_ops.c
> M subversion/libsvn_subr
> M subversion/libsvn_client/merge.c
> M subversion/bindings/swig
> M subversion/tests/cmdline/svntest/actions.py
> M subversion/tests/cmdline/merge_tests.py
> M CHANGES
> Sure enough, there were text changes to merge_tests.py, and 'svn diff'
> that. Something appears to be awry with the merge notifications.
This looks very disturbing.
Changing working copy nodes without a notification on it (or in some cases
one of its parents) will not only break the AnkhSVN in-memory status cache,
but normal users will ignore changes as well.
Subversion should not change the state of nodes without notifying the users
via the notification handler. Whether our cli client sends a full
notification to the user is another issue, but changing the state without
notifying library users is just bad behavior.
Received on 2008-12-18 14:36:22 CET