homepage

Using WinXP I compiled python 3.2 from the current sources using Visual C++ 2010 Express.
Running
rt -v test_socket
resulted in the attached output.
The tests testSmallReadNonBlocking and testWriteNonBlocking have errors, and an assertion fails for test_connect and test_create_connections.

current revision = svn revision 86553 

The core problem seems to be that the errno module has mismatching values on your Python build (for ECONNREFUSED and EWOULDBLOCK, at least, which explains all 4 failures). On a Windows 7 VM here:
>>> errno.ECONNREFUSED
10061
>>> errno.EWOULDBLOCK
10035

As this was requested on IRC, I put the Visual C++ 2010 Express solution and project files here as well. The patch "pcbuild_vs2010.patch" should be applied to the PCbuild directory.

Daniel: If you need VS2008, you can get it here: #2008-Visual-CPP">http://www.microsoft.com/express/Downloads/#2008-Visual-CPP 

Kotan: please perform the following procedure:
1. Open pcbuild.sln
2. In the pythoncore project, open errnomodule.c
3. find the reference to EWOULDBLOCK and WSAEWOULDBLOCK
4. for each one, "go to definition"
5. report here what header files Visual Studio thinks they are defined in, and what values they have.

#define EWOULDBLOCK 140
#define WSAEWOULDBLOCK 10035L
The editor suggests that EWOULDBLOCK is already defined, therefore its code is used, i.e. 140.
#define ECONNREFUSED 107
#define WSAECONNREFUSED 10061L
The E* are defined in errno.h and the WSA* in WinError.h. It seems that in my case the POSIX compliant headers are used.

OK, so what's the path to errno.h? In the errno.h included in my copy of VS 2010, the errno values only go up to 42 (80 if you count STRUNCATE), and EWOULDBLOCK is not defined in this header file.

c:\Program Files\Microsoft Visual Studio 10.0\VC\include\errno.h
The content of my errno.h:
...
#define EILSEQ 42
#define STRUNCATE 80
#endif
#endif
/* Support EDEADLOCK for compatibility with older MS-C versions */
#define EDEADLOCK EDEADLK
/* POSIX SUPPLEMENT */
#define EADDRINUSE 100
#define EADDRNOTAVAIL 101
...

Oops, I was confused. My copy is VS 2008, of course. So that's another incompatible change in VS 2010. Tcl, Ruby, and curl are ahead of us, here:
http://redmine.ruby-lang.org/issues/show/3092
http://sourceforge.net/tracker/index.php?func=detail&aid=3019634&group_id=10894&atid=310894
http://www.mail-archive.com/curl-library@cool.haxx.se/msg05164.html
I think we can work around that by switching the order of preference: use WSA constants if available, and errno.h values otherwise.
Kotan: can you please search the CRT sources whether these "POSIX" values are used anywhere?

Only in the "system_error" file. (http://msdn.microsoft.com/en-us/library/ee372194.aspx) Seem to be a C++0X standard header file.
http://msdn.microsoft.com/en-us/library/5814770t.aspx
talks about the Posix codes only for compatibility reasons.

> Only in the "system_error" file. (http://msdn.microsoft.com/en-us/library/ee372194.aspx) 
> Seem to be a C++0X standard header file.
(I assume you have actually scanned the crt\src folder as well)
Then it's indeed safe to ignore.
> http://msdn.microsoft.com/en-us/library/5814770t.aspx
> 
> talks about the Posix codes only for compatibility reasons.
Unfortunately, they managed to break compatibility with that change,
instead of introducing it.

Ok I did switched the preference for all "new" defines inside VC++ errno.h to use the WSA* alternatives if available.
Now test_socket passes for me, and the test_asyncore which did block before, now passes without a problem.

Confirming this patch fixes the test_aynsc* tests in my VS10 build. Shouldn't it swap all WSA* defines to protect against this in the future, though? Alternatively, should the check for WSA* codes existing be in Lib\asyncore.py?

I can also confirm that this patch corrects the problem with test_asyncore for Python 2.7 built with VS2010.

FYI: this would likely be handled through #13210. I have a conversion sandbox started at http://hg.python.org/sandbox/vs2010port/ and am working through fixing test failures after the initial conversion.

Added a patch as used by CCP in production. Covers more WSA cases.

This has been fixed with the proper 2010 support