There are segfaults on x86 FreeBSD 7.2 3.x. It looks like it's a regression introduced by r80876.

---

Build 556:
...
test_coding
test_hash
test_strtod
test_float
test_runpy
*** Signal 11

Build 555:
...
test_int
test_smtplib
test_asynchat
test_runpy
*** Signal 11

Build 554:
...
test_lib2to3
test_pickle
test_xml_etree
test_sys
[34438 refs]
[34470 refs]
[34728 refs]
[34468 refs]
*** Signal 11

Build 553:
...
test_robotparser
test_plistlib
test_signal
test_runpy
*** Signal 11

Build 552:
...
test_importlib
test_crypt
test_sort
test_runpy
*** Signal 11

Build 551:
...
./python -E -c 'import sys ; from sysconfig import get_platform ; print(get_platform()+"-"+sys.version[0:3])' >platform
[34466 refs]
./python -Wd -E -bb ./Lib/test/regrtest.py -uall -rwW -l 
== CPython 3.2a0 (py3k:81134, May 13 2010, 14:25:32) [GCC 4.2.1 20070719 [FreeBSD]]
==   FreeBSD-7.2-RELEASE-i386-32bit-ELF
==   /usr/home/db3l/buildarea/3.x.bolen-freebsd7/build/build/test_python_22056
Using random seed 3201291
test_exceptions
*** Signal 11

Build 550:
...
test_codecencodings_tw
test_zipfile
test_import
test_exceptions
*** Signal 11

Build 549:
...
test___future__
test_smtpnet
test_exceptions
*** Signal 11

(...)

Build 530 (r80876): ... signal 11
(Build 526..529 failed with svn exceptions)
Build 525 (r80775) didn't crashed.

I ran a test --without-threads: test_runpy and test_exceptions passed,
test_sys was not reached:


test_sqlite
Fatal error 'Cannot allocate red zone for initial thread' at line 384 in file /usr/src/lib/libthr/thread/thr_init.c (errno = 12)
*** Signal 6


No Signal 11. I suspect threads (again).

Would it be a good idea to run the FreeBSD bots --without-threads?

https://github.com/haypo/faulthandler/wiki can be tried on this buildbot to get more information about this issue. But the module have to be installed on this host.

Wouldn't that module have to be put into the actual source tree, since the tests run beneath the interpreter/libraries that are built for the test?

That may be what you meant, but "installed on this host" made me think I could do something external on the buildbot which I don't think would work given that the module has to be called from within the tests themselves?

-- David

> That may be what you meant, but "installed on this host" made me think
> I could do something external on the buildbot which I don't 
> think would work given that the module has to be called from within
> the tests themselves?

If you install faulthandler on "x86 FreeBSD 7.2 3.x" buildbot, I can create a special branch to add specific code.

Or I can add:

try:
    import faulthandler
except ImportError:
    pass
else:
    faulthandler.enable()

In py3k (eg. in Lib/test/support.py).

But the problem is now different: it looks like the bug was fixed, I don't see crashes anymore on  "x86 FreeBSD 7.2 3.x" buildbot :-)

There was failures on test_concurrent_futures, but Martin fixed it in #10798 (not completly (?) but it's better).

I close the issue because it looks like the crash was fixed. Reopen the issue if the crash occurs again.