You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@subversion.apache.org by Mark Phippard <ma...@gmail.com> on 2009/08/18 12:54:07 UTC
Tree conflict fix in 1.6.x STATUS
Julian,
Just curious, are you planning to review this backport? The 1.6.5
tarball should be rolling tomorrow and it would be good to get it in.
* r38000, r38006, r38099, r38100, r38181
Don't break the working copy when a tree-conflict features an incoming
replace.
Justification:
My colleagues and some clients are really annoyed by this problem.
Although it's been fixed, it is not in 1.6.x yet, which is met with
little understanding. Releasing 1.6.5 without this patch would
probably be some bad publicity...
Notes:
If a merge has a delete followed by an add on the same node, and
also that node is modified locally, svn tried to create two tree-
conflicts on the same node. That hit an assertion and bailed out,
breaking the working copy.
The cmdline test for this issue is merge_replace_causes_tree_conflict
in merge_tests.py, should become nr. 133.
(My `make check' with these revisions merged to 1.6.x is clean)
Votes:
+1: neels, stsp
--
Thanks
Mark Phippard
http://markphip.blogspot.com/
------------------------------------------------------
http://subversion.tigris.org/ds/viewMessage.do?dsForumId=462&dsMessageId=2384755
Re: Tree conflict fix in 1.6.x STATUS
Posted by Julian Foad <ju...@btopenworld.com>.
Stefan Sperling wrote:
> Stephen Butler noted this problem, too. See r38803.
Ah, ok. (I wish I'd emailed two hours ago when I first noticed something
was awry, instead of delving into it :-( )
- Julian
------------------------------------------------------
http://subversion.tigris.org/ds/viewMessage.do?dsForumId=462&dsMessageId=2384954
Re: Tree conflict fix in 1.6.x STATUS
Posted by Stefan Sperling <st...@elego.de>.
On Tue, Aug 18, 2009 at 08:45:34PM +0100, Julian Foad wrote:
> The reason seems partly to be that the test tries the wrong merge
> command:
> [[[
> svn merge file:///[...]/branch file:///[...]/A [...]/A
> --- Merging differences between repository URLs into 'svn-test-work/working_copies/merge_tests-133/A':
> C svn-test-work/working_copies/merge_tests-133/A/B/E
> C svn-test-work/working_copies/merge_tests-133/A/mu
> C svn-test-work/working_copies/merge_tests-133/A/D/G/pi
> C svn-test-work/working_copies/merge_tests-133/A/D/H
> Summary of conflicts:
> Tree conflicts: 4
> ]]]
>
> which, although its output looks like what we expect, is actually a
> nonsensical two-URL merge.
Stephen Butler noted this problem, too. See r38803.
Stefan
------------------------------------------------------
http://subversion.tigris.org/ds/viewMessage.do?dsForumId=462&dsMessageId=2384949
Re: Tree conflict fix in 1.6.x STATUS
Posted by Julian Foad <ju...@btopenworld.com>.
> * r38000, r38006, r38099, r38100, r38181
> Don't break the working copy when a tree-conflict features an incoming
> replace.
> Justification:
> My colleagues and some clients are really annoyed by this problem.
> Although it's been fixed, it is not in 1.6.x yet, which is met with
> little understanding. Releasing 1.6.5 without this patch would
> probably be some bad publicity...
> Notes:
> If a merge has a delete followed by an add on the same node, and
> also that node is modified locally, svn tried to create two tree-
> conflicts on the same node. That hit an assertion and bailed out,
> breaking the working copy.
> The cmdline test for this issue is merge_replace_causes_tree_conflict
> in merge_tests.py, should become nr. 133.
> (My `make check' with these revisions merged to 1.6.x is clean)
> Votes:
> +1: neels, stsp
In my testing, the output of "svn status" during the test run shows
incorrect tree conflict information:
[[[
M C 3 1 jrandom svn-test-work/working_copies/merge_tests-133/A/B/E
> local add, incoming add upon merge
M C 3 1 jrandom svn-test-work/working_copies/merge_tests-133/A/mu
> local obstruction, incoming add upon merge
M C 3 1 jrandom svn-test-work/working_copies/merge_tests-133/A/D/G/pi
> local obstruction, incoming add upon merge
M C 3 1 jrandom svn-test-work/working_copies/merge_tests-133/A/D/H
> local obstruction, incoming delete upon merge
]]]
These should all say "local modification" and "incoming delete" (or,
ideally, "incoming replacement").
The reason seems partly to be that the test tries the wrong merge
command:
[[[
svn merge file:///[...]/branch file:///[...]/A [...]/A
--- Merging differences between repository URLs into 'svn-test-work/working_copies/merge_tests-133/A':
C svn-test-work/working_copies/merge_tests-133/A/B/E
C svn-test-work/working_copies/merge_tests-133/A/mu
C svn-test-work/working_copies/merge_tests-133/A/D/G/pi
C svn-test-work/working_copies/merge_tests-133/A/D/H
Summary of conflicts:
Tree conflicts: 4
]]]
which, although its output looks like what we expect, is actually a
nonsensical two-URL merge. The correct merge command would be:
svn merge $URL/branch A
When I try this, the output is:
[[[
--- Merging r2 through r4 into 'svn-test-work/working_copies/merge_tests-133/A':
C svn-test-work/working_copies/merge_tests-133/A/B/E
C svn-test-work/working_copies/merge_tests-133/A/mu
/home/julianfoad/src/subversion-1.6.x/subversion/libsvn_client/merge.c:525: (apr_err=155016)
svn: Attempt to add tree conflict that already exists
]]]
This indicates that the problem is not fixed for node-kind-changing
replacements. However, if the file-replaces-file and dir-replaces-dir
cases are fixed, that is still a major improvement.
If I disable the kind-changing replacement tests, the status after the
merge is:
[[[
C C 3 1 jrandom svn-test-work/working_copies/merge_tests-133/A/B/E
> local add, incoming add upon merge
M C 3 1 jrandom svn-test-work/working_copies/merge_tests-133/A/mu
> local obstruction, incoming add upon merge
]]]
This still doesn't look right: the property status shouldn't show as 'C'
when there's a tree conflict, and I think the tree conflict type should
be "local modification" in both cases.
Perhaps as an intermediate step we can fix the code for same-kind
replacements and shunt the kind-changing replacement tests to a separate
Xfailing test.
Unfortunately I can't do any of this by tomorrow.
It's tempting to say that any improvement is better than none and this
changes a crash into an tree conflict report which, though incorrect, is
better. However, I can't recommend this for back-port in its current
state.
- Julian
------------------------------------------------------
http://subversion.tigris.org/ds/viewMessage.do?dsForumId=462&dsMessageId=2384930
Re: Tree conflict fix in 1.6.x STATUS
Posted by Julian Foad <ju...@wandisco.com>.
On Tue, 2009-08-18, Mark Phippard wrote:
> Julian,
>
> Just curious, are you planning to review this backport? The 1.6.5
> tarball should be rolling tomorrow and it would be good to get it in.
Thanks for the heads-up. I'll take a look.
- Julian
> * r38000, r38006, r38099, r38100, r38181
> Don't break the working copy when a tree-conflict features an incoming
> replace.
> Justification:
> My colleagues and some clients are really annoyed by this problem.
> Although it's been fixed, it is not in 1.6.x yet, which is met with
> little understanding. Releasing 1.6.5 without this patch would
> probably be some bad publicity...
> Notes:
> If a merge has a delete followed by an add on the same node, and
> also that node is modified locally, svn tried to create two tree-
> conflicts on the same node. That hit an assertion and bailed out,
> breaking the working copy.
> The cmdline test for this issue is merge_replace_causes_tree_conflict
> in merge_tests.py, should become nr. 133.
> (My `make check' with these revisions merged to 1.6.x is clean)
> Votes:
> +1: neels, stsp
------------------------------------------------------
http://subversion.tigris.org/ds/viewMessage.do?dsForumId=462&dsMessageId=2384766