Hi Paul. Not sure about this... > URL: http://svn.apache.org/viewvc?rev=1438683&view=rev
> Log: > Fix issue #4306 'multiple editor drive file merges record wrong mergeinfo > during conflicts'. > > * subversion/libsvn_client/merge.c > (do_file_merge): If we can only perform a partial merge because a conflict > broke the merge, then only record mergeinfo for that partial merge. That > was subsequent merges can finish the job. > > * subversion/tests/cmdline/merge_tests.py > (conflict_aborted_mergeinfo_described_partial_merge): Remove XFail > decorator and tweak failure status comment. > > Modified: > subversion/trunk/subversion/libsvn_client/merge.c > subversion/trunk/subversion/tests/cmdline/merge_tests.py > > Modified: subversion/trunk/subversion/libsvn_client/merge.c > URL: > http://svn.apache.org/viewvc/subversion/trunk/subversion/libsvn_client/merge.c?rev=1438683&r1=1438682&r2=1438683&view=diff > ============================================================================== > --- subversion/trunk/subversion/libsvn_client/merge.c (original) > +++ subversion/trunk/subversion/libsvn_client/merge.c Fri Jan 25 20:19:15 2013 > @@ -6923,7 +6923,10 @@ do_file_merge(svn_mergeinfo_catalog_t re > &filtered_rangelist, > mergeinfo_path, > merge_target->implicit_mergeinfo, > - &range, iterpool)); > + /* Only record partial mergeinfo if only a partial merge was > + performed before a conflict was encountered. */ > + conflicted_range ? conflicted_range : &range, > + iterpool)); Shouldn't that be "the range starting at 'range.start' and ending at 'conflicted_range.end'? It looks to me like it will only record mergeinfo for the sub-range that raised the conflict, and not for the earlier sub-ranges that completed successfully. I'm currently writing an extended test that will cover when a conflict is raised in: { first sub-range of first specified range; last sub-range of first specified range; first sub-range of last specified range; last sub-range of last specified range } x { single-file merge; dir merge } That should be able to confirm my suspicion about the above code. I'm doing this because I'm working on refactoring so that the three main merge subroutines called from do_merge() (that is the cases: single-file, dir-with-mergeinfo, dir-no-mergeinfo), each return info about where they stopped if they stopped because of a conflict, instead of throwing an error containing that info in human-readable form. Then I will be able to put the merge in a "do { merge; resolve any conflicts } while (merge is not complete)" loop at the 'svn' level. And thereby fix interactive conflict resolution which currently aborts at this point even if you resolved the conflicts. - Julian > > /* Only record mergeinfo if there is something other than > self-referential mergeinfo, but don't record mergeinfo if > > Modified: subversion/trunk/subversion/tests/cmdline/merge_tests.py > URL: > http://svn.apache.org/viewvc/subversion/trunk/subversion/tests/cmdline/merge_tests.py?rev=1438683&r1=1438682&r2=1438683&view=diff > ============================================================================== > --- subversion/trunk/subversion/tests/cmdline/merge_tests.py (original) > +++ subversion/trunk/subversion/tests/cmdline/merge_tests.py Fri Jan 25 > 20:19:15 > 2013 > @@ -18176,8 +18176,7 @@ def merge_properties_on_adds(sbox): > > @SkipUnless(server_has_mergeinfo) > @Issue(4306) > -@XFail() > -# Test for issue #4306 'multiple editor drive merges record wrong > +# Test for issue #4306 'multiple editor drive file merges record wrong > # mergeinfo during conflicts' > def conflict_aborted_mergeinfo_described_partial_merge(sbox): > "conflicted split merge can be repeated" > @@ -18218,8 +18217,8 @@ def conflict_aborted_mergeinfo_described > svntest.actions.run_and_verify_svn(None, None, '.*', 'merge', > '^/iota', > iota_copy_path, '--accept', > 'postpone') > > - # Currently this test fails because the merge fails after merging > - # only r2 (as it should) but mergeinfo for r5-6 is recorded, preventing > + # Previously this test failed because the merge failed after merging > + # only r2 (as it should) but mergeinfo for r5-6 was recorded, preventing > # subsequent repeat merges from applying the operative r5. > svntest.actions.run_and_verify_svn( > "Incorrect mergeinfo set during conflict aborted merge", >
