[Zope-Coders] core dump in Zope 2.7 test suite

[Tim Peters]

[Jeremy]
> I'm using CVS python built a couple of days ago.

Hmm.  Since nobody else has said "me too" (I haven't tried yet -- sorry),
maybe that's the problem.  There have been at least 4 checkins to
unicodeobject.c since the 2.3 release claiming to fix leaks, and those
always have a chance of introducing worse problems.

> I ran the tests again omitting the PageTemplate tests and found
> another failure...
>
> testUnIndexError
> (Products.PluginIndexes.PathIndex.tests.testPathIndex.TestCase) ... ok
> testing splitter functionality ... Segmentation fault (core dumped)

So lookin' more and more like random memory trashage.  What happens if you
run without any Unicode operations <wink>?

In the earlier debug output:

$9 = {_ob_next = 0x0, _ob_prev = 0x0, ob_refcnt = 0,
      ob_type = 0x8130160, length = 1, str = 0x41903130,
      hash = -1, defenc = 0x0}

that certainly means the object has been released already.

MAL's memory-mgmt is pretty elaborate.  It might pay to short-circuit all
the Unicode freelist stuff; then the debug pymalloc would kick in and spray
usefully useless bytes into freed memory, which often triggers errors closer
to the cause.  I'll note in passing that the comments for
KEEPALIVE_SIZE_LIMIT end with

   Note: This is an experimental feature ! If you get core dumps when
   using Unicode objects, turn this feature off.

I have a sinking feeling that he wouldn't have written that if he hadn't
seen core dumps.