hgbuildbot fails to signal merges properly causing incorrect builds

[dsallings]

I'm not entirely sure whether the fault lies in PBChangeSource or hgbuildbot.py, but I've worked around it in the latter.

Basically, when a changeset comes in with no files, buildbot doesn't consider it and will build from a changeset that arrived before it. This is very undesirable.

To work around this, I've special-cased a no-file changeset to say it changed the file merge which at least makes it work in my environment.

I don't like this all that much, but it's a discussion point.

[paulegan]

I've been using a similar patch to address this issue. It seems that the git change source has the same problem (and suggested fix): #249. I've attached a patch which includes a fix for this issue (and #328) as well as refactoring the code to make it a little more clear.

[marcusl]

Patch to make the inrepo-branch patch compile with hg 1.0

[marcusl]

I've added a updated patch that works with the new api in mercurial 1.0.2. (workingctx is not a member function of localrepo anymore + len(foo) replaces foo.count() in some occasions)

I hope you're able to make sense of this.

[paulegan]

I've been using a similarly updated patch since upgrading. It should be noted that the new api applies to Mercurial 1.1, and not 1.0.2. Both old & new apis can be supported with a try/except fall-back.

PS: in 1.1 you can also get workingctx with repo[None].

[paulegan]

Refactor and bug fixes for hgbuildbot.py

[marcusl]

Replying to paulegan:

I've been using a similarly updated patch since upgrading. It should be noted that the new api applies to Mercurial 1.1, and not 1.0.2. Both old & new apis can be supported with a try/except fall-back.

Ok. Sorry about that. I thought I had updated to the 1.0.2 tag in my hg-stable checkout.

PS: in 1.1 you can also get workingctx with repo[None].

I noticed that. It felt a bit clearer the way I did it though. No matter.

[dustin]

This is at:

 http://github.com/djmitche/buildbot/tree/bug181.

marcusl, do you want to do more with this patch?

[dustin]

(btw, tests are passing for me)

[marcusl]

I think it's as good as it gets. It does put the branch name in the changes, so it should be ok.

Note that we don't have a hg-inrepo test (see #187) yet.

[dustin]

ok, it's in :)