Fix for seg fault on test_re on mac osx

[sdm7g]

BPO	580869
Nosy	@mwhudson, @Yhg1s, @smontanaro, @warsaw, @jackjansen
Files	regrtest.diff stack-nocrash.txt

^{Note: these values reflect the state of the issue at the time it was migrated and might not reflect the current state.}

Show more details

GitHub fields:

assignee = 'https://github.com/jackjansen'
closed_at = <Date 2003-04-22.14:08:47.000>
created_at = <Date 2002-07-13.03:15:12.000>
labels = ['tests']
title = 'Fix for seg fault on test_re on mac osx'
updated_at = <Date 2003-04-22.14:08:47.000>
user = 'https://bugs.python.org/sdm7g'

bugs.python.org fields:

activity = <Date 2003-04-22.14:08:47.000>
actor = 'jackjansen'
assignee = 'jackjansen'
closed = True
closed_date = None
closer = None
components = ['Tests']
creation = <Date 2002-07-13.03:15:12.000>
creator = 'sdm7g'
dependencies = []
files = ['4428', '4429']
hgrepos = []
issue_num = 580869
keywords = ['patch']
message_count = 15.0
messages = ['40567', '40568', '40569', '40570', '40571', '40572', '40573', '40574', '40575', '40576', '40577', '40578', '40579', '40580', '40581']
nosy_count = 7.0
nosy_names = ['mwh', 'twouters', 'skip.montanaro', 'barry', 'nnorwitz', 'jackjansen', 'sdm7g']
pr_nums = []
priority = 'normal'
resolution = 'works for me'
stage = None
status = 'closed'
superseder = None
type = None
url = 'https://bugs.python.org/issue580869'
versions = ['Python 2.3']

[sdm7g]

    import resource
    soft, hard = resource.getrlimit(
resource.RLIMIT_STACK )
    resource.setrlimit( resource.RLIMIT_STACK, (1024 *
2048, hard))

is the python equivalent of the tcsh 'limit stack 2048'
and will
keep python from seg faulting on test_re .

( maybe wrapped in a "if os.platform == 'darwin' : " --
are there any other systems that have this problem ? )

-- Steve Majewski

[smontanaro]

Logged In: YES
user_id=44345

I tried adding the above code to regrtest.py. When I tried to
run python regrtest.py test_re I got a ValueError exception:
"not allowed to raise maximum limit". I have no trouble using
the ulimit builtin command to raise my stack size. Steve,
were you perhaps running as root when you tried this?

[smontanaro]

Logged In: YES
user_id=44345

BTW, attached is my diff against regrtest.py.

[smontanaro]

Logged In: YES
user_id=44345

BTW, attached is my diff against regrtest.py.

[smontanaro]

Logged In: YES
user_id=44345

BTW, attached is my diff against regrtest.py.

[smontanaro]

Logged In: YES
user_id=44345

Sorry for the bogus updates. Opera 6.0 Beta 1 apparently has file
upload problems. Resorting to IE...

[sdm7g]

Logged In: YES
user_id=105602

NOT running as root -- I just tried it again and verified
that it
works for me. HOWEVER: I haven't yet upgraded to Jaguar/10.2
Maybe that's the difference ?

I can post a request on the pythonmac list and ask some
others to
try it and find out if it's a 10.1.x vs 10.2 thing. -- Steve

[warsaw]

Logged In: YES
user_id=12800

I can verify the patch works on 10.1, specifically the OSX
Server in the SF compile farm.

[Yhg1s]

Logged In: YES
user_id=34209

I'd suggest adding a min(max, <new limit>) to the patch, so
that it becomes:

soft, hard = resource.getrlimit(resource.RLIMIT_STACK)
newsoft = min(hard, max(soft, 1024*2048))
resource.setrlimit(resource.RLIMIT_STACK, newsoft, hard)

And if the test suite only triggers this problem, doesn't it
make more sense in s?re.py, as Fredrik suggested ?

[Yhg1s]

Logged In: YES
user_id=34209

Oh, and the diff is kind of amusing. It seems to be
MacBinary, but 'file' on my Debian machine isn't quite sure:

/home/thomas/regrtest.diff: MacBinary III data with
surprising version number

Fortunately it's still readable :-) I guess I should go back
to "cool with a Macintosh" too.

[warsaw]

Logged In: YES
user_id=12800

Ok, I've added the Wouters Elaboration to regrtest.py and
verified it works on both 10.1 and 10.2 (the former being
the OSX machine on the SF compile farm and the latter being
my home machine). Attached is the patch (I hope :).

(Aside: why does this build python' for 10.1 but python.exe' for 10.2?)

I think this patch is fine for Py2.2.2. Today Guido said
he'd feel more comfortable with a patch to main.c than a
patch to regrtest.py, s?re.py, or site.py (if I understood
him correctly). However, I don't think there's time for
that for Py2.2.2. Perhaps we should simply commit this
simple fix so the test suite passes, and then work out a
more comprehensive patch for Py2.3?

Note that In addition, here is the output of the regrtest on
10.1 and 10.2. Is it worth spending time trying to fix
these for the Py2.2.2 release?

10.1:

170 tests OK.
2 tests failed:
test_fcntl test_largefile
21 tests skipped:
test_al test_cd test_cl test_curses test_dl
test_email_codecs
test_gdbm test_gl test_imgfile test_linuxaudiodev
test_locale
test_minidom test_poll test_pyexpat test_sax test_socket_ssl
test_socketserver test_sunaudiodev test_unicode_file
test_winreg
test_winsound
3 skips unexpected on darwin:
test_sax test_pyexpat test_locale
make: *** [test] Error 1

10.2:

171 tests OK.
1 test failed:
test_locale
21 tests skipped:
test_al test_cd test_cl test_curses test_dl
test_email_codecs
test_gdbm test_gl test_imgfile test_linuxaudiodev
test_minidom
test_nis test_poll test_pyexpat test_sax test_socket_ssl
test_socketserver test_sunaudiodev test_unicode_file
test_winreg
test_winsound
2 skips unexpected on darwin:
test_sax test_pyexpat

[mwhudson]

Logged In: YES
user_id=6656

(Aside: why does this build python' for 10.1 but python.exe' for 10.2?)

Because the sf machine has a case sensitive filesystem?

I don't see any failures on 10.1.5 (after I ran ulimit, duh).

Haven't tried this patch yet...