On Wed, Oct 24, 2012 at 7:16 PM, Paul Burba <ptbu...@gmail.com> wrote: > On Wed, Oct 24, 2012 at 2:14 PM, Stefan Küng <tortoise...@gmail.com> wrote: >> Hi, >> >> There were quite a few crash reports sent for TSVN which indicate a problem >> when merging. >> >> Until now I never really had a good crash dump and also no >> user to contact. >> >> But now that's changed: >> User report is here: >> http://tortoisesvn.tigris.org/ds/viewMessage.do?dsForumId=4061&dsMessageId=3023140 >> >> Crash dump available here: >> https://www.crash-server.com/Problem.aspx?ClientID=tsvn&Login=Guest&ProblemID=6224 >> (you should know the login data already, if not please contact my >> privately). >> >> Here's what I found so far: >> libsvn_client\merge.c, calculate_left_hand_side() >> >> /* Convert the absolute path with mergeinfo on it to a path relative >> to the session root. */ >> SVN_ERR(svn_client__path_relative_to_root(&path_rel_to_root, >> ctx->wc_ctx, absolute_path, >> target_repos_root, FALSE, >> NULL, scratch_pool, >> iterpool)); >> path_rel_to_session = svn_relpath_skip_ancestor(target_repos_rel_path, >> path_rel_to_root); >> SVN_ERR(svn_client__repos_location_segments(&segments, >> target_ra_session, >> path_rel_to_session, >> target_rev, target_rev, >> SVN_INVALID_REVNUM, >> ctx, scratch_pool)); >> >> here, the 'path_rel_to_session' is set to NULL by the >> svn_relpath_skip_ancestor() call, which then leads to a NULL-pointer access >> later on in svn_client__repos_location_segments and further down in >> svn_ra_get_location_segments(): >> SVN_ERR_ASSERT(*path != '/'); >> >> where 'path' is the NULL pointer of path_rel_to_session. >> >> svn_relpath_skip_ancestor() returns NULL if the paths are not related or if >> the second argument string does not end with a '/'. > > FWIW the second arg does not need to end with a '/' for a non-NULL return > value. > >> Gathering from the crash dump I can see that the second argument points to a >> file so it does not end with a '/' char. >> >> The two strings passed to svn_relpath_skip_ancestor() are according to the >> crash dump: >> somename/branches/dev62 >> and >> shared/trunk/somefile.bas >> >> (somename and somefile used to protect the innocent). >> >> I hope this helps to figure out why the crash happens - I'm not familiar >> with the merge code so someone else needs to look at this. > > I think I found the problem. The user mentioned he has "several > Externals as files, no folders, which may be worth mentioning as a > potential source of problem." I can produce a segfault during a merge > if the file externals in the reintegrate target have svn:mergeinfo set > on them. We should ignore these file externals as far as merge > tracking is concerned, but we aren't. Fixed on trunk in r1401915.
I also nominated r1401915 for backport to 1.7.x. -- Paul T. Burba CollabNet, Inc. -- www.collab.net -- Enterprise Cloud Development Skype: ptburba
