[Python-Dev] Python and POSIX (original) (raw)

Marc Recht recht at netbsd.org
Tue Sep 16 06:59:01 EDT 2003

You should take a look at Solaris, then. Depending on whether you define XPG macros (I forgot which ones specifically), the socket functions mutate from BSD (e.g. connect(3BSD)) to XPG functions (connect(3XPG)). This is implemented by exposing

_# pragma redefineextname connect xpgconnect or some such in socket.h, if the XPG macro is defined. I had a quick glance at the Solaris 8 headers and it seems that's influenced by EXTENSIONS, too. So it should be the same as without the macros.

Likewise, on HP-UX, you select one of the three (!) curses libraries on the system through these macros. Oh. Didn't know about HP-UX. That's weird.

So, why not split pyconfig.h up into an external and internal header? That would be possible if one would a) provide a patch, and OK. I justed wanted to verify if there's enough interest (and no general rejection) in such a patch and if anyone else is already working on it.

b) could demonstrate that, on all relevant systems, lack of such macros cannot possible have an effect on proper operation of extension modules. That's a bit harder since I don't have access to all "relevant" systems. I could test on the open-source x86 platforms, though. And IIRC CompileFarm has Solaris, too. And I believe that - since these changes came up with Python 2.3 (with the exception of _POSIX_1_SOURCE and _POSIX_SOURCE) - that it will break not much or anything at all if at least _XOPEN_SOURCE are moved _XOPEN_SOURCE_EXTENDED to an internal header (or an internal part of pyconfig.h). (As a first step.)

But, you're right there shouldn't be a rash decision and this needs much testing. I think I'll just go on and provide a patch as basis for further discussion.

Regards, Marc

More information about the Python-Dev mailing list