[Buildbot-devel] Re: upcoming release

[Brian Warner]

> Would someone be willing to take a look at the following bug report
> http://sourceforge.net/mailarchive/forum.php?thread_id=6705497&forum_id=32301
> 
> and tell me if the problem is addressed in 0.6.3?  Thanks

In a word, I'm not sure. The root cause is probably a mkdir or chdir which
fails, and I'd hazard a guess that it's one of those windows things where
spaces in a directory name cause other tools to break.

The failure in mkdir or chdir or whatever should have caused the step to
stop, but instead it kept running. I don't think anything has changed in the
step-failure-detection code since 0.6.2, so if there is a bug there, it may
still be present.

This one paragraph is an issue that should be fixed in CVS now. It was a
problem in the buildslave.[1]:

 * I typed "buildbot start c:\bb".  I ignored a stack trace that
   indicated a failure deleting c:\bb\full\builder (No such file or
   directory), since I didn"t understand it.


I noticed your config file used a factory defined as follows:

 f1 = factory.BuildFactory([s(step.ShellCommand, command="dir c:\\")])

Could you try using command=["dir", "c:\\"] and see if that make a
difference? The preferred way to define commands is with a list of arguments,
rather than requiring /bin/sh (or its windows equivalent) to parse the
command. Dunno if it'll make a difference, but it seems worth a try.

Also, could you send me the slave's twistd.log output for the failing
command? I'd like to see the stack trace that produced the "directory name is
invalid" error.

thanks,
 -Brian

[1]: incidentally, does anyone know why sf.net's mail archiver replaces quotes
with double quotes? It's annoying and lossy