tempfile crashes

[theller]

BPO	623464
Nosy	@tim-one, @theller
Files	posixmodule.c.diff: Patch for posixmodule.c

^{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/theller'
closed_at = <Date 2002-11-07.16:12:12.000>
created_at = <Date 2002-10-15.09:54:30.000>
labels = ['OS-windows']
title = 'tempfile crashes'
updated_at = <Date 2002-11-07.16:12:12.000>
user = 'https://github.com/theller'

bugs.python.org fields:

activity = <Date 2002-11-07.16:12:12.000>
actor = 'theller'
assignee = 'theller'
closed = True
closed_date = None
closer = None
components = ['Windows']
creation = <Date 2002-10-15.09:54:30.000>
creator = 'theller'
dependencies = []
files = ['632']
hgrepos = []
issue_num = 623464
keywords = []
message_count = 15.0
messages = ['12781', '12782', '12783', '12784', '12785', '12786', '12787', '12788', '12789', '12790', '12791', '12792', '12793', '12794', '12795']
nosy_count = 3.0
nosy_names = ['tim.peters', 'nnorwitz', 'theller']
pr_nums = []
priority = 'normal'
resolution = 'fixed'
stage = None
status = 'closed'
superseder = None
type = None
url = 'https://bugs.python.org/issue623464'
versions = ['Python 2.3']

[theller]

tempfile.NamedTemporaryFile(".zip") crashes with an
access violation. Win2k, SP2.

[nnorwitz]

Logged In: YES
user_id=33168

Note: You probably want:
tempfile.NamedTemporaryFile(suffix=".zip")

I tried this under Linux and get: OSError: [Errno 22]
Invalid argument

So this appears to be windows specific. Exactly which line
is causing the crash? os.fdopen?

[theller]

Logged In: YES
user_id=11105

It crashes in fdopen - looks like a bug in MS runtime library:

(in file vc98\crt\src\fdopen.c)
fdopen calls _tfopen(), creates a FILE *stream, and if the first
character in mode is not r, w, or a, sets 'stream' to NULL to
signal an error. Then it calls _unlock_str(stream), which
crashes.

[nnorwitz]

Logged In: YES
user_id=33168

We don't check anywhere else for valid mode chars. We would
have to fix Modules/posixmodule.c and Modules/socketmodule
if we wanted to check. Thomas, do you think a checks should
be added or close this as a 3rd party problem?

[theller]

Logged In: YES
user_id=11105

IMO we should check for a valid mode. There is a large
surprise if python crashes with an access violation when the
innocent user does innocent things like this. But this is just
*my* opinion.

[nnorwitz]

Logged In: YES
user_id=33168

Sounds good to me. Care to work on a patch with tests?

[tim-one]

Logged In: YES
user_id=31435

We generally don't check for valid mode characters
because C allows implementations to extend the standard
set, there's no way to query the platform about which mode
extensions it supports, and we don't want to block users
from using platform-specific mode extensions.

It would be OK by me if we insisted that the first character
be in "rwa", but checking more than that is off limits. Note
that Windows simply ignores all of the mode chars starting
with the first one it doesn't recognize:

>>> f = open('example.txt', 'what a crock this mode string 
is!')
>>> f
<open file 'example.txt', mode 'what a crock this mode 
string is!' at 0x00651870
>
>>>

That was the same as passing "w".

[theller]

Logged In: YES
user_id=11105

Here's a patch for Modules/posixmodule.c, which fixes this
particular issue, conforming to Tim's requirements.

[tim-one]

Logged In: YES
user_id=31435

Cool! One thing: drop the Windows #ifdef. I asked Guido,
and he agrees that Python wants to enforce this requirement
on all platforms.

[theller]

Logged In: YES
user_id=11105

Great, will do this tomorrow. What about socketmodule.c,
which Neal mentioned?

[tim-one]

Logged In: YES
user_id=31435

If socketmodule isn't crashing, I'd leave it alone.

[nnorwitz]

Logged In: YES
user_id=33168

Thomas, I belivee the code below will check if the other
cases crash.

# test posixmodule.popen(), 
# would need check around line 2835
# as first code in popen()
import posix
# you may need a command that works, 
# I'm guessing cmd.exe is ok
posix.popen('cmd.exe', '.zip')

# test socketmodule.sock_makefile(), 
# would need check around line 1572
# after PyArg_ParseTuple
from socket import *
s = socket(AF_INET, SOCK_STREAM)
s.makefile('.zip')

[theller]

Logged In: YES
user_id=11105

I couldn't get it to crash with Neil's code, so I checked it in
without the Windows #ifdef: posixmodule.c, rev 2.271.

BTW: I always assume we don't need NEWS entries for bug
fixes, is this correct?

[tim-one]

Logged In: YES
user_id=31435

Thanks! You certainly need a NEWS item for this, and
possibly even a doc change, since requiring that mode
strings begin with certain letters is a new language
requirement. However unlikely, *somebody* may have code
out there that will break because of it.

[theller]

Logged In: YES
user_id=11105

Sigh, even more work.
In this case, it cannot be called a bugfix anymore...