Hi Daniel, On Tue, Jun 22, 2010 at 08:50:33PM +0300, Daniel Shahaf wrote: > http://subversion.tigris.org/issues/show_bug.cgi?id=3641 > > Briefly, the issue is that when svnsync encounters the following history: > ------------------------------------------------------------------------ > r5 | pm | 2010-05-18 17:53:46 +0100 (Tue, 18 May 2010) > Changed paths: > A /H (from /A:4) > R /H/B (from /X:4) > ------------------------------------------------------------------------ > r3 | pm | 2010-05-18 17:53:46 +0100 (Tue, 18 May 2010) > Changed paths: > A /A/B/C > ------------------------------------------------------------------------ > it looks for /H/B/C in the sync source when trying to replay r5. > > I've attached a patch (with log msg) that seems to have some positive > effects: it causes the sync to pass (and properly add children of /X as > children of /H/B). > > I'm asking for review for two reasons: > > * authz. The whole function is authz-sensitive --- its goal is to > represent a copy as an add. I think the patch is okay from this > angle, but a second pair of eyes wouldn't hurt. > > * assertions. The patch asserts that svn_fs_path_change2_t->copyfrom_known > is TRUE. However, the FS API used --- svn_fs_paths_changed2() --- > does not guarantee that copyfrom_known will in fact be TRUE, and > I haven't found a different API that does promise to provide the > copyfrom information. (I think the patch only needs this information > for directory replaces.)
While I cannot comment on the above... > Testing, analysis, reviews are welcome. ... I built a 1.6.11 Subversion with the patch and svnsync finished syncing the affected repository. A quick comparison shows no evident problems. [...] Thanks, Tino. -- "What we nourish flourishes." - "Was wir nähren erblüht." www.tisc.de
