[Python-Dev] refleak status

[Michael Hudson]

After a week or so of frantically searching for these things, I sort of 
forgot about the reference leak hunt I started in Python's tests.  A 
week ago I ran my hacked regrtest.py again, made some notes, tried & 
failed to send them to python-dev, went on holiday, came back, digged a 
little more and am trying to send again...

real leaks i think i've fixed (am syncing up to CVS now, will test 
overnight).

     test_class leaked [66, 66, 66, 66] references
     test_hotshot leaked [111, 111, 111, 111] references

real leaks i cant fix:

     test_compile leaked [45, 45, 45, 45] references
         lambda a,a:0 worst offender.  confused.

red herrings:

     test_codeccallbacks leaked [2, 2, 2, 2] references
         callback registry
     test_mutants leaked [-1489, 1520, -755, -183] references
         randomized test

suspected bugs in _Py_RefTotal tracking:

     test_descr leaked [2, 2, 2, 2] references
         something to do with resurrection
     test_gc leaked [18, 18, 18, 18] references
         test_trashcan ... not a real leak, it seems

suspected red herrings:

     test_cpickle leaked [6, 3, 2, 1] references
     test_logging leaked [82, 82, 82, 82] references
         (bet this is registry stuff of one sort or another)
     test_threadedtempfile leaked [-74, 3, 3, 161] references
     test_unicode leaked [7, 7, 7, 7] references
         registries again

apparent refleaks that still need digging:

     test_minidom leaked [1, 1, 1, 1] references
     test_new leaked [2, 2, 2, 2] references
     test_pkg leaked [8, 8, 9, 9] references
     test_pkgimport leaked [5, 6, 6, 6] references
     test_sax leaked [1762, 1762, 1762, 1762] references

Wouldn't like to claim anything about the following tests:

6 tests failed:
     test_dircache test_doctest test_getopt test_multifile
     test_warnings test_zipimport
27 tests skipped:
     test_al test_bsddb test_bsddb3 test_cd test_cl test_curses test_dl
     test_email_codecs test_gdbm test_gl test_imgfile test_largefile
     test_linuxaudiodev test_locale test_mpz test_nis
     test_normalization test_ossaudiodev test_pep277 test_poll
     test_socket_ssl test_socketserver test_sunaudiodev test_timeout
     test_urllibnet test_winreg test_winsound
Those skips are all expected on darwin.

(test_threaded_import deadlocks, so add that to the list)

Should investigate why the failing tests fail when run repeatedly and
check that they're not hiding ref leaks.

Cheers,
mwh