[Buildbot-devel] GitPoller fixes in 0.8.3p1

[Dustin J. Mitchell]

On Thu, Jan 13, 2011 at 2:07 PM, Matisse Enzer <menzer at apple.com> wrote:
>> That happens when the fetch fails, for whatever reason.  Amber has
>> suggested that a better solution would be to use remote branches, so
>> e.g., we'd log master..origin/master instead of HEAD..FETCH_HEAD.  I
>> think that might improve the situation a little bit (after a failed
>> fetch, you'd just get zero revisions).

I just fixed this particular problem - the git poller in 'master' now
tracks $branch against origin/$branch, which is more reliable than the
fairly ephemeral HEAD and FETCH_HEAD

> OK - I'll see if I can add some code to show me why the fetch is failing.
> I admit I am considering switching to post-receive hooks instead of polling. (I have not used them because i don't like the idea that the git repo needs to know about buildbot.)

We're using post-receive hooks in the metabuildbot because they're
easy to set up with github.  In general, if that's possible, I
recommend it, since it doesn't require a git repository on the master,
etc.  Still, GitPoller *should* work and work well!

Dustin