[Python-checkins] Python Regression Test Failures refleak (80)

[Neal Norwitz]

On 6/13/06, Georg Brandl <g.brandl at gmx.net> wrote:
> Michael Hudson wrote:
> > Neal Norwitz <neal at metaslash.com> writes:
> >
> >> test_builtin leaked [8, 8, 8] references
> >> test_exceptions leaked [40, 40, 40] references
> >
> > So who broke refcounting then?  This is pretty new and must be
> > somewhere fairly core, hopefully someone remembers changing something
> > there?  I don't remember any specific checkins though.
>
> FWIW, it would perhaps be a good idea for the refleak mail
> to include something like sys.subversion.

The data is available in the docs.python.org/dev/results directory.
If you want to copy the results into the mail it shouldn't be too
hard.  It's all in Misc/build.sh.

The problem in this case was I was doing some testing in the same dir
and had outstanding changes that screwed up ref counting.  There are
always a few outstanding changes.  I can think of 4 of the top of my
head:

 build.sh - don't ignore any leaky test
 regrtest.py - ignore test_tcl
 test_socket and test_threadedtempfile - sleeps to avoid the ref count
problems.  this gives threads time to shutdown and seems to be
working.  the test_socket needs to be improved, i will probably check
in the tempfile fix.  it's kinda ugly.  i had sent a patch a while
ago.  it's the same thing with the bug fixed (forgot to increment
count).

n