python2.3 buildbot, mostly working
C. Titus Brown
during the PyCon talks today, I fixed most of the issues with the 2.3
buildbot, and incorporated Namshin's os.command fix for the subprocess
issue in downloader. You can see the patches in the buildbot-fixes
branch,
http://github.com/ctb/pygr/commits/buildbot-fixes/
The most complicated patches were for a bug in Python 2.3's shelve.Shelf
class, which tries to access self.writeback and self.dict even when
those attributes may not have been set (for e.g. a failed 'open').
There's still one annoying and weird error,
http://lyorn.idyll.org:8019/builders/Python23/builds/14/steps/run%20tests/logs/stdio
where Python is segfaulting at the end of runtest.py. Someone should
verify that in their own python2.3 installation; I doubt it's specific
to our install on lyorn, but it could be.
cheers,
--titus
--
C. Titus Brown, c...@msu.edu
Christopher Lee
I cleaned up and pushed the buildbot-fixes to master. Get the latest
code from the public repo.or.cz repository. Comments below.
On Mar 29, 2009, at 2:34 PM, C. Titus Brown wrote:
> The most complicated patches were for a bug in Python 2.3's
> shelve.Shelf
> class, which tries to access self.writeback and self.dict even when
> those attributes may not have been set (for e.g. a failed 'open').
I cleaned up these fixes, because they needlessly propagated a bad
pattern from Python's shelve module. Our code is much cleaner as a
result. I don't understand why the Python folks haven't cleaned this
up themselves -- the obvious solution is so much simpler and more
robust! For more details see:
http://code.google.com/p/pygr/issues/detail?id=81
-- Chris
C. Titus Brown
-> I cleaned up and pushed the buildbot-fixes to master. Get the latest
-> code from the public repo.or.cz repository. Comments below.
Awesome, thanks!
-> On Mar 29, 2009, at 2:34 PM, C. Titus Brown wrote:
->
-> > The most complicated patches were for a bug in Python 2.3's
-> > shelve.Shelf
-> > class, which tries to access self.writeback and self.dict even when
-> > those attributes may not have been set (for e.g. a failed 'open').
->
->
-> I cleaned up these fixes, because they needlessly propagated a bad
-> pattern from Python's shelve module. Our code is much cleaner as a
-> result. I don't understand why the Python folks haven't cleaned this
-> up themselves -- the obvious solution is so much simpler and more
-> robust! For more details see:
->
-> http://code.google.com/p/pygr/issues/detail?id=81
I'll look at it tonight or over the weekend -- thanks for doing this!
Note, buildbots updated; I had to remove blast_tests again, but other
than that annoying crash bug in python2.3/pygr, they're green:
git://repo.or.cz/pygr.git
...actually, there's some sort of annoying mysql crosstalk bug. Grr.
I'll deal with YOU later, mister mysql...
Christopher Lee
On Mar 29, 2009, at 2:34 PM, C. Titus Brown wrote:
> There's still one annoying and weird error,
>
> http://lyorn.idyll.org:8019/builders/Python23/builds/14/steps/run%20tests/logs/stdio
>
> where Python is segfaulting at the end of runtest.py. Someone should
> verify that in their own python2.3 installation; I doubt it's specific
> to our install on lyorn, but it could be.
I see the same Python 2.3.3 segfault at the end of the tests running
on linux 2.4.20.
-- Chris
C. Titus Brown
-> On Mar 29, 2009, at 2:34 PM, C. Titus Brown wrote:
->
-> > There's still one annoying and weird error,
-> >
-> > http://lyorn.idyll.org:8019/builders/Python23/builds/14/steps/run%20tests/logs/stdio
-> >
-> > where Python is segfaulting at the end of runtest.py. Someone should
-> > verify that in their own python2.3 installation; I doubt it's specific
-> > to our install on lyorn, but it could be.
->
-> I see the same Python 2.3.3 segfault at the end of the tests running
-> on linux 2.4.20.
Darn.
I'm going to guess it's a Pyrex bug. I'll have to track it down the ol'
fashioned binary search way.