[Buildbot-devel] Problem with a slave updating via git

[Tim Hatch]

I installed a new builder today with updated versions of
Buildbot/Twisted and I think I may have discovered a bug on WinXP.  I
was wondering if anyone else has a similar setup involving git source steps:

* Windows XP SP2 or WEPOS 1.0
* python 2.6
* buildbot 0.7.12 from the zip
* twisted 8.2.0 or 9.0.0
* msysgit 1.5.6 through the newest (1.6.5.1)
* a git source step
* building a specific new rev (i.e. using contrib/git_buildbot.py)

For me, this worked under 0.7.11p3 because it would unconditionally
fetch, before doing a reset --hard $rev.  The 0.7.12 behavior is to trap
errors doing a reset --hard $rev first, and if that happens, fetch.

The trapping works fine on Windows 7 and Linux (in cursory testing)
using the same version of Twisted.  On Windows XP and WEPOS, it appears
to return success, and the "fetching branch master from $url" never
appears (it moves on to the rev-parse step).

I tried the git reset --hard $rev manually, and it sets errorlevel=128.
 I've attached the logs I do have from this stage -- both the
web-visible portion of the log (which shows stderr listing the failure)
as well as the slave's twistd.log (which shows that it existed with
signal None, status 0, which is wrong).  Any thoughts on how to test
further?

Tim
-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: buildbot-slave-web.log
URL: <http://buildbot.net/pipermail/devel/attachments/20100126/884bf6fb/attachment.ksh>
-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: buildbot-slave-twistd.log
URL: <http://buildbot.net/pipermail/devel/attachments/20100126/884bf6fb/attachment-0001.ksh>