[PyPy-issue] [issue524] intermittent seg fault

D.S. McNeil pypy-dev-issue at codespeak.net
Tue Apr 27 17:59:01 CEST 2010

New submission from D.S. McNeil <dsm053 at gmail.com>:

r74045.  Linux 2.6.18-8.el5 x86_64, compiled w/32-bit python-2.6.5, gcc 4.1.1.
pypy-c w/jit.

Resulting binary works nicely, except I'm seeing a seg fault on code which is
incredibly sensitive.  That is, it comes and goes with slight changes in the
code, but it's entirely deterministic.  I've attached the resulting code, which
consistently segfaults for me:

[mcneil at smpserv goal]$ ./pypy-c ~/sf9.py 100000
# getting gonal for sides= 3
# getting gonal for sides= 4
# getting gonal for sides= 5
# getting gonal for sides= 6
# getting gonal for sides= 7
# getting gonal for sides= 8
working up to 8 for range 100000 1000000
have 140391 terms to consider
left unbuilt: 3906
s =  8 could not build 100190
s =  8 could not build 100824


s =  8 could not build 128208
s =  8 could not build 128894
Segmentation fault

Note that the code itself no longer makes much sense since most of it's been
removed, and it contains references to variables no longer used.. but removing
them can get rid of the segfault.  Almost every further simplification does the
same.  Some change the point at which it breaks instead.

In fact, if I run pypy-c from the home directory using "pypy-c sf9.py 100000"
instead, then I segfault at 128208 instead of 128894.

Even running the code under gdb gets rid of the segfault in Heisenbug fashion,
so I'm worried that it'll be hard to reproduce for anyone else.  I'm happy to
run whatever diagnostics would be useful.

effort: ???
files: sf9.py
messages: 1702
nosy: dsm, hpk, pypy-issue
priority: bug
release: ???
status: unread
title: intermittent seg fault

PyPy development tracker <pypy-dev-issue at codespeak.net>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: sf9.py
Type: text/x-python
Size: 1165 bytes
Desc: not available
URL: <http://mail.python.org/pipermail/pypy-issue/attachments/20100427/7ed3b10b/attachment.py>

More information about the Pypy-issue mailing list