Please download subprocess.zip and extract the archive. Run - 

perl test.pl

you will notice 'invalid handle' error on duplicating STD_INPUT_HANDLE in python. This is observed in Windows XP SP3. Vista SP2 and higher platforms doesn't show this failure when I checked.

"test.pl" spawn a python process "test.py" using CREATE_NEW_PROCESS_GROUP flag. Before it does, it close STDIN and create a listening socket. Perl under the hood, convert the socket handle to file descriptor using open_osfhandle(..) call. When it does this, the C runtime set the standard input handle to this descriptor  for console app. [ See msvcr* osfinfo.c code - SetStdHandle( STD_INPUT_HANDLE, (HANDLE)value )  ]. As a result, when you get the STD_INPUT_HANDLE, you will get the new one. Duplicating this handle, hence throw the invalid handle error in Windows XP SP3. 

When I looked at MSDN for DuplicateHandle(..), it is mentioned not to duplicate socket handle using this call. So I suspect something going wrong in XP platform. To fix this issue, I have come up with a patch - python-2143_patch. Please take a look and let me know your comments.

-Karthik

I'm sorry, but I genuinely can't see what you're trying to say here. If you believe that there's a bug in Python's standard library, can you show a reproducible *Python* testcase for it, please (using the pywin32 modules if that helps) and state clearly what you think is wrong.

Hi Tim,

Thanks for your reply.

I added test case using perl and python since it was easy to reproduce using perl socket module and show the issue happening with python's subprocess.py. There is definitely an action required in subprocess.py to catch 'invalid handle error' and my attached patch handle this situation.

I can come up with a test case like perl does to create a listening socket and map the socket handle to a file descriptor. You will see this issue *only* when the open_osfhandle(..) is used to map the handle to a file descriptor.

-Karthik

On 13/11/2012 20:57, Karthk Rajagopalan wrote:
> I added test case using perl and python since it was easy to
> reproduce using perl socket module and show the issue happening with
> python's subprocess.py. There is definitely an action required in
> subprocess.py to catch 'invalid handle error' and my attached patch
> handle this situation.

So, because subprocess.call includes a call to DuplicateHandle(stdin),
that call will fail if stdin is currently a handle to a socket.
Is that right?

Your patch fails to apply cleanly to the hg tip, because Gregory P.
Smith has been making changes in this very piece of code over the last
few days. I'll add him as interested to the call as someone who can make
a call as to the suitability of your patch.

> I can come up with a test case like perl does to create a listening
> socket and map the socket handle to a file descriptor. You will see
> this issue *only* when the open_osfhandle(..) is used to map the
> handle to a file descriptor.

If this is to go ahead, we will need a test case which can run as part
of the Python stdlib tests, and perl is not part of that stdlib!

Hi Tim, 

Thanks for your reply.

Yes, DuplicateHandle(..) seem to fail if the handle is to a socket under XP SP3. 

Can you point me to the guidelines about submitting patch so I can merge my change in main branch and upload it?

We build python from source and doesn't use pywin32 as of now. So I downloaded them yesterday to come up with a test case in python which you can use in your test bed.

-Karthik

Start here: http://docs.python.org/devguide/

In particular: http://docs.python.org/devguide/patch.html

Sorry I can't help on the windows side of things.  I don't have access to any windows systems.

Karthk, if you can run up an up-to-date patch and a test I'm willing to review and commit it, otherwise this one will lie quiet.

Is XP under Python support now?  I'm sure I've read that it isn't but can't find a reference.