bisect does not drill down into merge commits

[SirVer]

Bzr merge does not drill down into mergecommits that merged one branch into another. It is therefore only manually possible to detect errors in commits that came into the history via merging. This is a major downside of using bzr's bisect vs for example git's bisect.

[SirVer (sirver)]

I do not want to be pushy here, but a bisect plugin for a distributed version control system *must* support merge commits imho. Otherwise it is rather a bisect plugin for bzr-svn then for bzr. An importance of "Critical" would be more appropriate.

[Gustaf Thorslund (gthorslund)]

I've put three test cases in a branch at lp:~gthorslund/bzr-bisect/bug-539937 (and related it to his bug). Two of them will fail. I think it's because of this bug, but review of the test cases are welcome. So far no progress in fixing the bug, but if I do I'll push it to the same branch.

[Martin Pool (mbp)]

On 23 October 2010 03:54, Gustaf Thorslund <email address hidden> wrote:
> I've put three test cases in a branch at lp:~gthorslund/bzr-
> bisect/bug-539937 (and related it to his bug). Two of them will fail. I
> think it's because of this bug, but review of the test cases are
> welcome. So far no progress in fixing the bug, but if I do I'll push it
> to the same branch.

If you want you can propose the branch for merging with those cases
marked as KnownFailure and a link to the bug, to checkpoint your work.

--
Martin

[Gustaf Thorslund (gthorslund)]

Martin,

Thanks for the hint. I've done so now.

/Gustaf

[Xavier (Open ERP) (xmo-deactivatedaccount)]

This seems to work OK in the latest trunk (as I was pleasantly surprised to discover), shouldn't this bug be closed?

[Gustaf Thorslund (gthorslund)]

Xavier,

Do you have some test that didn't work before and that does work now? I've seen at least some case where it works now, and some where it doesn't. The test cases I have added still fails. Try:

$ bzr selftest -v -s bp.bisect

So far not figured out what's failing.

/Gustaf

[Xavier (Open ERP) (xmo-deactivatedaccount)]

> Do you have some test that didn't work before and that does work now?

Nope, it's just that I started using bzr bisect recently, and had quite a long session with it today at work, and it drilled down into merged branches just fine. Granted, these were not too complex branches.

[Gustaf Thorslund (gthorslund)]

I think the problem with "corner" cases covered by some small test cases. In my cases the merge is in revision 2. One looking for a change in the merge point, and another at revision 1.2 in the sub-tree.