On Thu, Nov 11, 2010 at 11:03:09AM +0100, Macieira Thiago (Nokia-MS-Qt/Oslo) 
wrote:
> Replaying merges isn't very easy, as evidenced by the little use that
> rebase -m has.
>
well, it's somewhat laborious and easy to botch up, thus usually not
worth it. but given the constraints of the CI system it's fairly easy to
automate it in a reliable way. maybe i should prototype a
git-replay-merge.

> It also depends on the merges only happening in one direction and
> requires that no conflict resolution happened in those merges.
> 
that's simply not true as demonstrated in the cherry-pick substitution
sequence a few posts back. the idea is to start a merge, and in case of
conflict throw away all changes and then re-apply the diff against the
target branch from the original merge commit (including merge conflict
resolutions).

> A merge that was successful at one point can fail when rebased/remerged.
>
of course, but that's no different from any other commit in the queue.
it gets automatically set to revise and resubmit, and you repeat it.
of course, you would have rerere.enabled=true (rerere.autoupdate=true is
also very helpful), so you'd only need to resolve the new conflicts.
that's in fact what i do to avoid a second merge when the target branch
"runs away" while i'm resolving conflicts.
_______________________________________________
Opengov mailing list
Opengov@qt-labs.org
http://lists.qt-labs.org/listinfo/opengov

Reply via email to