git does not retry if the fetch fails

[dustin]

On the metabuildbot, we're seeing

fatal: read error: Invalid argument

on Windows XP slaves. I think this is just due to windows being a lame excuse for an operating system, but Buildbot should work around it by retrying. The retry parameter is set to (5,5) on the metabuildbot, but no retry occurs.

[dustin]

I can't replicate this in master..

This may be due to the old buildbot version on those slaves.

[dustin]

Well, the slaves are running 0.8.1, which has the retry behavior. The Twisted version is 10.1.0. I added

diff --git a/slave/buildslave/commands/vcs.py b/slave/buildslave/commands/vcs.py
index b0e8f7f..7dd33d5 100644
--- a/slave/buildslave/commands/vcs.py
+++ b/slave/buildslave/commands/vcs.py
@@ -1019,7 +1019,7 @@ class Git(SourceBase):
         command = ['fetch', '-t', self.repourl, '+%s' % self.branch]
         self.sendStatus({"header": "fetching branch %s from %s
"
                                         % (self.branch, self.repourl)})
-        return self._dovccmd(command, self._didFetch)
+        return self._dovccmd(['false']+command, self._didFetch)
 
     def _didClean(self, dummy):
         # After a clean, try to use the given revision if we have one.

locally to try to replicate the problem (on Linux), and I see this being retried. Making the same change on the slave did not retry.

I upgraded the cm-bbot-xp-NNN slaves to buildbot-slave-0.8.3, and all seems well. So I'm calling this fixed for now.