This issue tracker has been migrated to GitHub, and is currently read-only.
For more information, see the GitHub FAQs in the Python's Developer Guide.

classification
Title: Py_Initialize crashes when stdout has been redirected with freopen
Type: crash Stage:
Components: Windows Versions: Python 3.3, Python 3.4
process
Status: languishing Resolution: works for me
Dependencies: Superseder:
Assigned To: Nosy List: Mitchell.Stokes, amaury.forgeotdarc, brian.curtin, christian.heimes, pitrou, tim.golden
Priority: normal Keywords:

Created on 2010-10-13 04:11 by Mitchell.Stokes, last changed 2022-04-11 14:57 by admin.

Files
File name Uploaded Description Edit
py_main.c Mitchell.Stokes, 2010-10-13 04:11 Sample Program
py_main_fileno_check.c Mitchell.Stokes, 2010-10-13 18:54
py_main_dup2.c Mitchell.Stokes, 2010-10-13 18:54
Messages (8)
msg118502 - (view) Author: Mitchell Stokes (Mitchell.Stokes) Date: 2010-10-13 04:11
I have Python embedded in a C/C++ program, and I'm trying to redirect stdout to a text file. Since I'm trying to get rid of the console, I want C and Python stdout going to the text file (so, not sys.stdout = open('log.txt', 'w')). To this end, I used freopen() like so:

freopen("log.txt", "w", stdout);

to redirect stdout in C. However, when using Python 3.1, I get the following error:

Fatal Python error: Py_Initialize: can't initialize sys standard streams
OSError: [Errno 9] Bad file descriptor

followed by a crash (Runtime Error). I think this is somewhat related to this bug: http://bugs.python.org/issue1415

I've attached a sample program that illustrates the problem. I compiled it with:

gcc -o py_main.exe py_main.c -IC:\Python31\include -LC:\Python31\libs -lpython31

I had someone test the sample program on Linux, and it worked fine there, so I think it's a Windows issue.
msg118504 - (view) Author: Amaury Forgeot d'Arc (amaury.forgeotdarc) * (Python committer) Date: 2010-10-13 06:42
It looks very similar to issue6501.
Can you check if setting the environment variable
    PYTHONIOENCODING=cp1252
solves the problem?
msg118515 - (view) Author: Antoine Pitrou (pitrou) * (Python committer) Date: 2010-10-13 12:22
I don't think freopen() is the right solution, since it might open another (different) file descriptor under the hood; but Python always uses file descriptor 1 when creating sys.stderr.

I would suggest instead something such as (untested):

int tmpfd;
tmpfd = open("myfile.txt", O_CREAT | O_WRONLY);
dup2(tmpfd, 1);
msg118540 - (view) Author: Mitchell Stokes (Mitchell.Stokes) Date: 2010-10-13 16:38
Setting PYTHONIOENCODING=cp1252 does not help.

Also, dup2 is not available on Windows (at least not from what I've read). From my understanding, freopen() is supposed to be the portable way to redirect stdout.
msg118546 - (view) Author: Antoine Pitrou (pitrou) * (Python committer) Date: 2010-10-13 16:52
> Also, dup2 is not available on Windows (at least not from what I've
> read).

MSDN says dup2 is deprecated but you can use _dup2 instead:
http://msdn.microsoft.com/en-us/library/8syseb29%28v=VS.80%29.aspx

> From my understanding, freopen() is supposed to be the portable way to
> redirect stdout.

I'm not sure, but as I said Python does not use C's "stdout". Instead,
it writes directly (using write()) to the file descriptor number 1. So,
if freopen() creates a different file descriptor for the C stdout,
Python's stdout will not be redirected where you want it to be.

I'm not saying that's what happens in your program, though. You could
call _fileno(stdout) after the reopen() call to check if it's still 1 or
not.
msg118562 - (view) Author: Mitchell Stokes (Mitchell.Stokes) Date: 2010-10-13 18:54
I used _dup2() and I still got a crash. Also, if I use _fileno(stdout) after using freopen(), I get 1.

I'm uploading two more example programs.

py_main_fileno_check.c prints the fileno to the text file
py_main_dup2.c uses _dup2() to redirect stdout instead of freopen
msg118565 - (view) Author: Antoine Pitrou (pitrou) * (Python committer) Date: 2010-10-13 19:07
Thanks for the information. I'm afraid I can't be of any more help, since I'm not a Windows developer. I hope someone else can chime in.
msg192378 - (view) Author: Christian Heimes (christian.heimes) * (Python committer) Date: 2013-07-05 23:58
I can't reproduce the issue with Python 3.2 and dup2 on Linux.

$ gcc -o py_main py_main_dup2.c  -I/usr/include/python3.2mu -lpython3.2mu -lm -pthread -ldl -lutil
$ ./py_main

log.txt contains 'foo' and the debug message, too. Is this still an issue for you?
History
Date User Action Args
2022-04-11 14:57:07adminsetgithub: 54289
2013-07-05 23:58:34christian.heimessetstatus: open -> languishing
versions: + Python 3.3, Python 3.4, - Python 3.1
nosy: + christian.heimes

messages: + msg192378

resolution: works for me
2010-10-13 19:07:15pitrousetnosy: + tim.golden, brian.curtin
messages: + msg118565
2010-10-13 18:54:16Mitchell.Stokessetfiles: + py_main_dup2.c
2010-10-13 18:54:03Mitchell.Stokessetfiles: + py_main_fileno_check.c

messages: + msg118562
2010-10-13 16:52:23pitrousetmessages: + msg118546
2010-10-13 16:38:04Mitchell.Stokessetmessages: + msg118540
2010-10-13 12:22:31pitrousetnosy: + pitrou
messages: + msg118515
2010-10-13 06:42:33amaury.forgeotdarcsetnosy: + amaury.forgeotdarc
messages: + msg118504
2010-10-13 04:11:13Mitchell.Stokescreate