Re: mingw32 and sockets

[Simon Josefsson]

Bruno Haible <address@hidden> writes:

> Simon Josefsson wrote:
>> Would it be possible to create those three POSIX header files if they
>> don't exist on the system, and if winsock2.h do exist?  The created
>> files would simply #include <winsock2.h>.
>
> Yes, that's the way I would go.

It seems doing that this is not sufficient, the functions doesn't
conform to POSIX behavior.

We could provide replacement functions, perhaps inspired by the
re-implementations in plibc, see
http://cvs.sourceforge.net/viewcvs.py/plibc/plibc/.

However, that sounds like real work, and unpaid Win32 work isn't my
kind of thing, so I'll leave that part for somebody else.  Providing
the POSIX headers seem like good starting point, though, some Win32
functions may be close-enough so that large parts of a program works.
I'll see if the original reporter is interested in carrying on this,
to eventually make GnuTLS built through mingw32 work.  I have no idea
how difficult it would be.

>> Creating them would be done
>> like getopt_.h and others.  The complication is that this has to be
>> done in a sub-directory somehow, so that #include for sys/socket.h
>> works.
>
> Sure. It makes the Makefile.am snippet a little more complicated than for
> the getopt_.h case, because you need a 'distclean-local' rule that does a
> 'rmdir sys'. But then it will work as smoothly as getopt_.h.

Yup.

>> Perhaps this should be considered a mingw32 bug instead?
>
> It would be if the mingw people reach out for POSIX compliance. Do they?

The answer seem to be no.  So ideally, I think the incompatibilities
should be solved by gnulib.

Thanks,
Simon