[Svnmerge] [PATCH] Bidirectional patch take 2

[Raman Gupta]

Raman Gupta wrote:
> Raman Gupta wrote:
>> Blair Zajac wrote:
>>> I tried your original patch against Giovanni's repository and needed
>>> this fix in there to get it to work.
>> That's weird, I didn't need it then, and I also don't seem to need it
>> now (using the latest versions from collabnet with no patches). I am
>> using svn 1.2.3. What version are you using?
> 
> I replicated the issue using svn 1.3.0. The attached patch should fix
> the underlying problem rather than masking the symptom. I believe the
> reason is that repeated discontinuous merges cause conflicts with the
> svnmerge-integrated property. The attached patch reverts those changes,
> if any, after each merge, which should be fine because the last
> operation is to set the svnmerge-integrated property to its correct
> value. Can anyone think of a reason not to do this?

BTW, here is the issue which I believe causes the difference between svn
1.2.3 and 1.3.0 with relation to the svnmerge-integrated property conflicts:

http://subversion.tigris.org/issues/show_bug.cgi?id=2035

The behavior between 1.2.3 and 1.3.0 is different enough that people
should be aware of this, as it may affect us again in the future. I'd
like us to think more about how these property merges in 1.3.0 are
operating, and how we can work around spurious property conflicts when
executing merges for multiple ranges.

The 1.2.3 behavior was actually a little easier as far as svnmerge was
concerned, despite the behavior being incorrect, because the
svnmerge-integrated property is always set after all is said and done,
so conflicts in that property should really never be considered.

Cheers,
Raman