Kevin Jacobs added the comment: Couldn't resist looking into this more and I must admit that I am a bit stumped. If instead of returning a value from the cache fast path in _PyType_Lookup, I modified the code to store the value and assert that it matches with the non-cached result from traversing the mro. That assertion never fails. However, test_incomplete works with this change that ideally should not alter the semantics of _PyType_Lookup.
More so, if I selectively don't cache the name "next" (with the caching fast path reenabled), test_incomplete works. For test_incomplete, _PyType_Lookup on "next" always return NULL from both the cached and non-cached paths. I'm beginning to suspect there is some black magic going on. None of the relevant reference counts drop to zero, the same return value (NULL) is generated, but the slow patch seems to be doing something with a side effect that makes it work. Wish I had more time to debug this, but I'm already overdrawn until Wednesday... _____________________________________ Tracker <[EMAIL PROTECTED]> <http://bugs.python.org/issue1700288> _____________________________________ _______________________________________________ Python-bugs-list mailing list Unsubscribe: http://mail.python.org/mailman/options/python-bugs-list/archive%40mail-archive.com
