I observed the same yesterday.
One We fixed today with the testcase.
I was working with Senthil to write a testcase for the other one to prove and fix the issue(which I could see in my local test data).
We got stuck in testing a rollback of a merge(-rX:X-1), Senthil is working on it once it is ready will commit the fix.
From: Daniel Rall [mailto:firstname.lastname@example.org]
Sent: Fri 7/6/2007 12:15 AM
To: Kamesh Jayachandran
Subject: Merging of the svn:megeinfo property
On Wed, 04 Jul 2007, email@example.com wrote:
> Fix the problem: Two people from 2 different working copies at revision rX
> trying to do a merge of rY and rZ independently from /branch/b1 to /trunk.
> Assume the first person commits(merge of rY), when second guy tries to
> do the commit(merge of rZ) he would get a out of date error. So second
> guy would do 'svn up'. svn up does not merge the 'svn:mergeinfo' correctly
> It just keeps rZ.
> * subversion/libsvn_wc/props.c
> (svn_wc__merge_props): Record the merged svn:mergeinfo on the target.
> --- trunk/subversion/libsvn_wc/props.c (original)
> +++ trunk/subversion/libsvn_wc/props.c Wed Jul 4 11:50:49 2007
> @@ -830,6 +830,8 @@
> SVN_ERR(combine_forked_mergeinfo_props(&to_val, from_val,
> to_val, pool));
> + apr_hash_set(working_props, propname,
> + APR_HASH_KEY_STRING, to_val);
> *state = svn_wc_notify_state_merged;
Kamesh, there are two other places in svn_wc__merge_props() where I
think we may need to modify working_props. The attached patch blindly
applies your same fix (from 25646 and r25653) to those two places.
However, as described in the inline comment, we may need to do
something smarter. 'merge_tests.py 56' fails with this patch.
Received on Thu Jul 5 21:29:23 2007