[Python-checkins] Python Regression Test Failures refleak (1)

[Neal Norwitz]

On 3/6/06, Brett Cannon <brett at python.org> wrote:
> On 3/6/06, Neal Norwitz <neal at metaslash.com> wrote:
> > test_capi leaked [-143, 0, 0] references
> > test_charmapcodec leaked [-54, 0, 0] references
> > test_cmd_line leaked [-15, 15, 0] references
> > test_compiler leaked [42, 408, 16] references
> > test_generators leaked [255, 255, 255] references
> > test_quopri leaked [17, 0, 0] references
> > test_socket leaked [0, 0, 197] references
> > test_threadedtempfile leaked [3, 2, 2] references
> > test_threading_local leaked [42, 42, 42] references
> > test_urllib2 leaked [80, -130, 70] references
>
> Should we be worrying about the ones that do not leak any references
> consistently in an extreme manner?  Such as test_capi or
> test_cmd_line; they seem to have some import overhead or something
> that spikes references but they stabilize and do not leak anything in
> the end.  And test_socket is just weird.  =)

If the number is inconsistent, it might mean there's no leak.  The
inconsistency is due to cacheing or some module somewhere else in the
system holding references.  Ideally all these cases would be handled
(see the nested function cleanup() in regrtest.py).

If you find places that references are being stored and should be
cleaned up when running -R, please fix regrtest.py or send me a
message and I'll try to fix it up.

n


n