buildbot try does the wrong thing with mercurial queues

[dsallings]

We'd like to be able to use buildbot try and mercurial queues. It's a pretty simple change to make:

Instead of asking for the tree identity, ask for the identity of the qparent. If there's not a qparent, ask for the current identity (as is done now). Then always diff against the discovered patch level.

A patch is attached that seems to do the job well.

[dsallings]

try with mq support

[Ben]

Seems reasonable to me, any objections to that one ?

For the non Mercurial aware, this patch means that if you have any pending patch (as opposed to normal changeset) on your repository, the diff is done with those actual pending patches. If not, revert back to the current modif on the working dir.

[dustin]

 http://repo.or.cz/w/buildbot.git?a=shortlog;h=refs/remotes/dustin/bug180

(note: I have no way to test this, as I don't use hg -- I'm going on Ben's "seems reasonable")

[Ben]

Note that a test case would be nice ...

I'll try to combine one tonight.

[Ben]

I cannot make my testsuite working ...

So I leave the ticket open, untill *someone* fix the testsuite or the patch ....

Here is where I am so far (see patch actually)

[Ben]

draft of testcase

[Ben]

to finish for today, it's more a wish than a defect ...

[dustin]

OK, this patch is available in integrated form at:

 http://repo.or.cz/w/buildbot.git?a=shortlog;h=refs/remotes/dustin/bug180

Since this is a "draft", I haven't pulled it into the dev tree yet, and will un-mark it for review.

===============================================================================
[ERROR]: buildbot.test.test_vc.Mercurial.testTryMq

Traceback (most recent call last):
  File "/home/dustin/devel/projects/buildbot/t/buildbot/buildbot/test/test_vc.py", line 345, in check
    + ": see logs for stdout")
exceptions.RuntimeError: command ['/usr/bin/hg', 'qnew', '-f', 'patch.diff'] finished with exit code 255: see logs for stdout

By the way, dsallings, what is your full name and email address, so I can give you credit?

[dsallings]

Great, thanks. Dustin Sallings <dustin@…>

[dsallings]

I just realized this was showing an error. I haven't used hg with buildbot in a while, but I was using it with this recipe. Is it actually failing? If so, does the log show anything?

[dustin]

looks like 'qnew' is too new for my version of hg?

[I] dev-util/mercurial
     Available versions:  1.0.1-r2 ~1.0.1-r3 1.0.2 {bash-completion bugzilla cvs darcs emacs git gpg subversion test zsh-completion}
     Installed versions:  1.0.2(02:02:29 12/03/08)(-bash-completion -bugzilla -emacs -gpg -test -zsh-completion)
     Homepage:            http://www.selenic.com/mercurial/
     Description:         Scalable distributed SCM

[dsallings]

You have to enable mq before the commands work:

~/.hgrc:

[extensions] hgext.mq =

[dsallings]

Sorry, poorly formatted.

[extensions]
hgext.mq=

[dustin]

Hmm, I don't really know what that means, but is there some way to detect that and either skip the tests or give a helpful failure message in that case? That might deserve another bug.

For now, I see this, which I don't really know how to diagnose.

===============================================================================
[FAIL]: buildbot.test.test_vc.Mercurial.testTryMq

Traceback (most recent call last):
  File "/home/dustin/devel/projects/buildbot/t/buildbot/buildbot/test/test_vc.py", line 964, in _do_getpatch_trunkhead_2
    self.failUnlessIn("try",ss.patch[1])
  File "/home/dustin/devel/projects/buildbot/t/buildbot/buildbot/test/test_vc.py", line 382, in failUnlessIn
    self.failUnless(string.find(substring) != -1, msg)
twisted.trial.unittest.FailTest: did not see the expected substring 'try' in string ''
-------------------------------------------------------------------------------
Ran 1 tests in 4.782s

By the way, I have cloned the buildbot repo at  http://github.com/djmitche/buildbot. If you want to meet me in #buildbot (freenode), maybe we can figure out a way to make this patch process easier.

[dsallings]

hg help | grep qnew

[dustin]

OK, I added a fix to detect the mq extension and skip if it's not present, at

 http://github.com/djmitche/buildbot/tree/bug180

However, with the proper .hgrc, this is still failing..

[dustin]

hmm, this may be due to an earlier commit. Running 'git bisect' now.

[dsallings]

This is just under a year old. I pretty much don't use hg anymore for anything, but the functionality was very useful when I was.

That is to say, I don't know how useful I'll be in getting it up-to-date, as I don't use it anymore, and last time I tried updating local git changes over cvs, there were too many conflicts.

It might be good to reach out to more hg users and see if anyone can at least tell if it still works.

[dustin]

did-not-see-expected-test.log

[dustin]

gotcha, thanks.

[marcusl]

Is it skipping the entire Mercurial tests or just the mq-part? It looks like it's the entire shebang, which is a bit worrying, no?

Also, how about using the repository's hgrc file ( i.e .hg/hgrc) to just enable the extension for our test repo?

Just my 0.02$.

[dustin]

This is still failing for me; I just rebased it onto master.

 http://github.com/djmitche/buildbot/tree/bug180

[marcusl]

I've merged with current master and made a small mod that only skips the tryMqTest. It fails here too, but at least the failure is more localized and more test work without mq-enabled.

 http://github.com/marcusl/buildbot/tree/bug180

[dustin]

I've got your most recent changes merged in my bug180, but it's still failing. Just FYI :)

[dustin]

What's going on here? I pulled marcusl's bug180 branch. I have

[extensions]
hgext.mq=

in my .hgrc, and

dustin@euclid ~/code/buildbot/t/buildbot [bug180] $ hg showconfig | grep hgext
extensions.hgext.mq=

yet

[SKIPPED]: buildbot.test.test_vc.MercurialInRepo.testCheckout

Add 'hgext.mq =' to the '[extensions]' section of ~/.hgrc

We should really either get this merged or give up..

[marcusl]

Haven't worked on for.. well.. 8 months.

I'm not using mq either, so I've never gotten around to scratch this itch properly.

[dustin]

next release, then :)

[dustin]

bump..

[burcin]

attachment:try-mq.diff works for me, after manually applying the patch to tryclient.py. Could this be merged for the next release?

The first hunk of  https://github.com/marcusl/buildbot/commit/50c3be07aa646a3668e38503e3abd25d1bb00879 seems to remove the code that checks if mq is available. Perhaps reverting this would fix the tests.

[marcusl]

The commit was made to only disable the testTryMq() test, not the entire Mercurial test case, when mq is not enabled. (Since I and many others didn't use MQ, Mercurial tests weren't run at all.)

That was the intention at least. Looking at it now, it seems as if MercurialHelper?._qnew_supported might be set (or not), and the commit removed some code which would update that value again.

The right thing would be to make sure that _qnew_supported has a correct value always, and not to rely on the unit tests to check and set it. (Could be a property that's set lazily inside MercurialHelper?, on first access....)