Description of problem:

I came across a strange problem in Python which can be reproduced only if the test case is executed via `make':

# make tests
 cd ../.. && /bin/sh /usr/src/redhat/BUILD/libxml2-2.6.26/missing --run automake-1.9 --gnu  python/tests/Makefile
 cd ../.. && /bin/sh ./config.status python/tests/Makefile 
config.status: creating python/tests/Makefile
## running Python regression tests
python: Objects/obmalloc.c:953: PyObject_Free: Assertion `pool->ref.count > 0' failed.

The test case is thread2.py from libxml2 sources.

The test case actually writes "OK" and is expected to quit (according to the code):

# Memory debug specific
libxml2.cleanupParser()
if libxml2.debugMemory(1) == 0:
    print "OK"
else:
    print "Memory leak %d bytes" % (libxml2.debugMemory(1))
    libxml2.dumpMemory()


But it spits the failed assertion.

Version-Release number of selected component (if applicable):

libxml2-2.6.26-2.1.12.el5_7.1.x86_64
libxml2-2.6.26-2.1.12.el5_7.1.i386
python-2.4.3-44.el5_7.1.x86_64

How reproducible:

always

Steps to Reproduce:
1. /usr/bin/python thread2.py in Makefile (see Makefile.am in /usr/src/redhat/BUILD/libxml2-2.6.26/python/tests/).

Additional info:

It happens on x86_64 but I suspect i386 is also affected.