subversion-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Ivan Zhakov <i...@visualsvn.com>
Subject Re: svn commit: r1740320 - in /subversion/trunk/subversion: include/svn_client.h libsvn_client/conflicts.c svn/conflict-callbacks.c tests/libsvn_client/conflicts-test.c
Date Thu, 21 Apr 2016 16:37:50 GMT
On 21 April 2016 at 17:34, Stefan Sperling <stsp@apache.org> wrote:
> On Thu, Apr 21, 2016 at 02:04:08PM -0000, stsp@apache.org wrote:
>> Author: stsp
>> Date: Thu Apr 21 14:04:08 2016
>> New Revision: 1740320
>>
>> URL: http://svn.apache.org/viewvc?rev=1740320&view=rev
>> Log:
>> Add a conflict resolution option for dir/dir "incoming add vs local
>> obstruction upon merge". This option merges the two directories.
>>
>> Again, the implementation is not atomic, yet.
>> And it doesn't always work as expected.
>> Add two XFAIL regression tests which illustrate known problems.
>
> I'll need some help here to decide how to proceed.
>
> Briefly, the problems I'm running into are:
>
>  - It is not possible to merge an "only-adds" delta from rN-1 to rN for
>    a path created in rN. Essentially, this is the same issue as we had
>    for 'svn diff -cN' for a node created in rN, which was fixed at some point.
>    So this is a case where svn diff -cN works, but svn merge -cN doesn't.
>    I now need svn merge -cN to work in this case (see the 1st of 3 tests
>    added in this commit).
>
>  - I'm not sure to what extent the resolver should be responsible for
>    mergeinfo. Should it assume that existing mergeinfo in a working copy
>    remains valid when further merges are performed to resolve a tree
>    conflict? I guess not. In the case I'm looking at, the merge only
>    produces a delta if run with --no-ancestry (why?) which disables
>    mergeinfo recording. It is possible to construct cases where the lack
>    of additional mergeinfo recording seems wrong (see the 3rd of 3 tests
>    added in this commit).
>
> Does anyone have suggestions about these problems?
>
> Should the resolver be using the standard merge code at all in this case?
> Perhaps that's the wrong approach?
>
> Implementation aside, I do think the option to merge the two directories
> makes sense, even if they are ancestrally unrelated.
May there are some implementation problems, but I think merging two
directories makes sense: it's real world case during some refactoring.

-- 
Ivan Zhakov

Mime
View raw message