[Buildbot-devel] win32: prob with cvs -D checkouts

[Brian Warner]

> Guess nobody else is having this problem.  As a workaround, is there a 
> way to force the master to just use old-style "-r HEAD" checkouts for 
> just one builder?

Hey, sorry for the delay. And sorry for the bug! That's a wacky one.

Yeah, there is an override switch called 'alwaysUseLatest' that tells the
Source step to stop being clever and just use -rHEAD (or whatever is
appropriate for the VC system). Add it as a keyword argument when you create
the CVS object:

 steps = [s(step.CVS, cvsroot="foo", mode="clobber", alwaysUseLatest=True),
          s(step.Configure),
          ...]


> > It appears that the email-RFC-style date format is not parseable by 
> > cvsnt versions before 2.0 but, on after updating cvsnt, it appears to 
> > me (best I can tell) that the date string is not properly quoted for 
> > the shell and the 2.0 client barfs on the individual components.  This 
> > may be a magical difference between win python and unix python (os x 
> > for me) b/c it's the same buildbot code on both platforms but appears 
> > to behave differently.

It might have to do with the way the 'cvs' command is spawned. Under unix,
where we've got exec(), we can specify an array of argument strings directly,
so there's no quoting/dequoting loop to go through. The win32 spawnProcess()
mechanism may not have that luxury, so it might be doing args.join(" "),
which will lose if any of your arguments have spaces.

I'm not sure where the windows spawnProcess() code is at these days, there
were some folks at PyCon last week talking about how broken it was, but then
there were some others who said no, it's just fine.

Would you throw a bug onto the tracker for this one? It's possible that the
recently-released Twisted-2.0 may influence it (but note that I haven't done
much testing of buildbot against 2.0, YMMV).

cheers,
 -Brian