Closed GoogleCodeExporter closed 9 years ago
I have seen the exact same error on the ZODB3 tests (the object database which
uses pickle protocol one to
persist Python objects).
In that case they only happened for special C extension types and I wasn't able
to craft a minimal test case
showing the problem.
Original comment by hanno...@gmail.com
on 14 Jun 2009 at 1:46
Yeah, this is a known issue with the changes we made to the cPickle module.
I'll try
to get a fix in this week.
Original comment by collinw
on 15 Jun 2009 at 6:50
Patch ready, need to test it against Django trunk@head.
Original comment by collinw
on 27 Jun 2009 at 3:02
Django trunk@head now passes at a cost of 10% unpickling performance. Oh well.
Patch
mailed for review.
Original comment by collinw
on 27 Jun 2009 at 7:24
Issue 7 has been merged into this issue.
Original comment by collinw
on 29 Jun 2009 at 10:01
Part of the problem was fixed in r686. There's still a problem with reading
from real
files, as opposed to file-like objects, that probably has to do with EOF
caching.
Django trunk@head now passes, though. ZODB still shows regressions, as does
cvs2svn
(which tests all kinds of corner cases).
Original comment by collinw
on 1 Jul 2009 at 11:18
The actual problem was that we need to hold an extra reference to each object
in the
Pickler's memo for as long as the Pickler object lives. A fixed has been
mailed, which
makes cvs2svn's test suite pass.
Original comment by collinw
on 3 Jul 2009 at 12:12
The refcount bug was fixed in r690. That sucked to find.
Original comment by collinw
on 4 Jul 2009 at 3:57
Original issue reported on code.google.com by
alex.gay...@gmail.com
on 13 Jun 2009 at 10:01