[Buildbot-devel] Questions around build request merging

[Dustin J. Mitchell]

On Thu, Dec 27, 2012 at 10:46 AM, Axel Hecht <l10n.moz at gmail.com> wrote:
> I guess the hard problem is, in particular for DAG-based version control
> systems, how do I find out which revision to use to include those N requests
> in a build. It's kinda boring for single-head VCs, obviously, and it works
> well for projects which try to stick to a head per branch. The problem stays
> that if you don't know the topology of your VC, you're facing a hard
> problem, and it doesn't matter much if you move that around.

Right, and this gets us a step closer to being able to handle that.  I
think it's pretty simple, actually: if source stamp A is an ancestor
of B, they're collapsible, and Buildbot should use B.  If not, then
they're not collapsible.  But I don't plan to implement that in 0.9.0.

As for the blame list, I see that as a derived characteristic -
something that can be determined after the fact based on what actually
happened.  Furthermore, a user might want different blamelist
calculations for IRC, email, etc., which would be impossible to do in
advance.

I also figure that anyone who wants to know "who broke the build" and
is also using queue collapsing should expect Buildbot to only give an
approximate answer anyway: queue collapsing throws out information.

As for Jared's point about tracking what was collapsed into what -- I
think that's useful state to have available for reporting about a
build.  It's also a good source of information for the blamelist.

Dustin