Re: [Monotone-devel] buildbot 0.7.5 and mtn 0.28+?

[Nathaniel J. Smith]

On Mon, Dec 25, 2006 at 02:12:22AM -0500, Shawn Samuel wrote:
> Has anybody updated the monotone buildbot steps to support mtn and
> whatever changes have been made to the automate interface? I looked at
> the most recent buildbot release (0.7.5), and it has buildbot steps
> for monotone that still reference "monotone" as the executable. I'm
> wondering just how far from working this code is, and if there is any
> more recent code out there that I could start with to get buildbot
> 0.7.5 running against mtn?

Oh, goodness, I sent Brian a patch for monotone support a year+ ago,
but I didn't realize he ever actually integrated it.  You're
definitely right that the version upstream won't work anymore as is.

AFAIK, the current status is:
  -- upstream buildbot must have the original version of my patch,
     which is now against an obsolete version of monotone, but may
     have been updated to actually work with current upstream, which
     has been undergoing a lot of changes
  -- we have a locally hacked copy of buildbot at
       http://venge.net/monotone/monotone-buildbot-2.tar.bz2
     which is what we are actually running.  It is based on some
     now-ancient cvs snapshot of buildbot, but does work with modern
     monotone.  OTOH, it doesn't actually work terrifically _well_
     with modern monotone, since I just made some horrific kluges to
     keep it limping along.  E.g., to get the list of modified files
     in each change, it doesn't actually parse the revision, it just
     pulls out all double-quoted strings.
  -- at this point there are several python programs with much better
     monotone interface code than the stuff I kluged up in buildbot,
     for instance, viewmtn and the mtn trac module.  This could be
     pulled out and re-used in buildbot (with presumably some tweaking
     to make it Twisted-friendly).
  -- I think someone said something on monotone-devel at some point
     about updating the buildbot integration stuff for their own use,
     but we never actually saw any of the code from that; possibly
     they still have it.

So, in summary, there is an old and klugy thing that works, and if you
want anything better than that, you'll have to do some work to update
the integration glue, and ideally get Brian to take it upstream.  It's
important work, though, and probably not too hard.  (Let me know if
you try this, and if you need any pointers.)

Cheers,
-- Nathaniel