On Thu, 2011-02-10 at 12:10 -0500, Paul Burba wrote: > Hi Neels (or any other tree conflict gurus), > > I know you don't have much time for Subversion these days, but if you > find some time in the near future could you take a look at the tests > you added in r879162 and combined in r879206 (this test currently > resides at merge_tree_conflict_tests.py 23 'replace vs. delete > tree-conflicts'). > > I made a few tweaks to the test in r1069145 and r1069158 and it now > reaches the end of the test and fails. > > Currently the test does the following: > > 1) Merge a directory replacement onto a locally deleted directory > > 2) Merge a directory-replaced-with-a-file onto a locally deleted directory > > 3) Repeat the merge of a file-replaced-with-a-directory onto a locally > deleted directory done in #2, but this time target a subtree of #2's > target.
This doesn't repeat anything of #2, it's completely separate. #2 directory-replaced-with-a-file onto A/D/H (target A/D depth immed) #3 file-replaced-with-a-directory onto A/D/G/pi (target A/D/G) I would expect to see a fourth case tested too: #4 file-replaced-with-a-file onto mu (The test starts by calling merge_replace_setup() which sets up for all of these four cases.) > The only effect this has is the change the resulting TC type > from 'local delete, incoming delete upon merge' to 'local delete, > incoming replace upon merge'. The ideal expectation would be for all three or four cases to report 'local delete, incoming replace upon merge'. I don't know how close we are in implementation to being able to report these as incoming replacements; it might not be easy. > I'm a bit foggy on what we should expect here and the test doesn't > have an associated issue (I was looking at this in hopes of adding > one). > > Here's what we are currently seeing on trunk: > > Case #1: > > >svn st -v A\B\E > RM + C - 4 ? A\B\E > > local delete, incoming delete upon merge That's quite good except it should say "incoming replace". > Case #2: > > >svn st -v A\D\H > D C 3 1 jrandom A\D\H > > local delete, incoming replace upon merge That's quite good except it should say "R" not "D". > Case #3: > > >svn st -v A\D\G > M 3 1 jrandom A\D\G > RM + C - 4 ? A\D\G\pi > > local delete, incoming delete upon merge That's quite good except it should say "incoming replace". Wierd that in all these cases the "D"/"R" is opposite to the message. (Also, I think status should never report "M" in the properties column of a "R"eplaced node. But that's a separate issue, probably unrelated to conflicts.) - Julian > I was half expecting that the incoming delete of the replacement be > handled first and all of these be of the 'local delete, incoming > delete upon merge' variety and have a local status of 'D ', but that > is neither what the test expects or (obviously) what the results are. > > Any insight you have into this is appreciated. If you don't have time > to look at this, are there any threads where this was discussed or > particular docs I should be looking at? The only relevant thing I > could find was the 'Renames and Replacements' section of > https://svn.apache.org/repos/asf/subversion/trunk/notes/tree-conflicts/resolution.txt, > but that is somewhat inconclusive. > > Thanks, > > Paul