Re: MS-Windows build broken in Fmake_network_process

[Helmut Eller <eller.helmut@gmail.com>]

* Eli Zaretskii [2010-03-26 16:22+0100] writes:

> This change:
>
>     revno: 99750
>     author: Helmut Eller <eller.helmut@gmail.com>
>     committer: YAMAMOTO Mitsuharu <mituharu@math.s.chiba-u.ac.jp>
>     branch nick: trunk
>     timestamp: Thu 2010-03-25 17:48:52 +0900
>     message:
>       Call `select' for interrupted `connect' rather than creating new socket (Bug#5173).
>
> breaks the MS-Windows build:
>
> The compiler emits a warning, and the linker errors out:
>
>   gcc -I. -c -gdwarf-2 -g3  -mtune=pentium4 -O2     -Demacs=1 -DHAVE_CONFIG_H -I../nt/inc -DHAVE_NTGUI=1 -DUSE_CRT_DLL=1 -o oo-spd/i386/process.o process.c
>   process.c: In function `Fmake_network_process':
>   process.c:3663: warning: passing arg 4 of `getsockopt' from incompatible pointer type
>
>   oo-spd/i386/temacs1.a(process.o)(.text+0x3297): In function `Fmake_network_process':
>   D:\gnu\bzr\emacs\trunk\src/process.c:3663: undefined reference to `getsockopt@20'
>
> The compiler warning is because the prototype on Windows is:
>
>   int getsockopt(SOCKET, int, int, char*, int*);
>
> The linker error is because we would need to link against yet another
> library to get this function.  But I don't think we should do that.
> I'm actually bewildered why this code:
>
> 	    int len = sizeof xerrno;
> 	    eassert (FD_ISSET (s, &fdset));
> 	    if (getsockopt (s, SOL_SOCKET, SO_ERROR, &xerrno, &len) == -1)
> 	      report_file_error ("getsockopt failed", Qnil);
>
> was used unconditionally when a very similar code in
> wait_reading_process_output is clearly marked with a comment saying
> not to use it except on GNU/Linux:
>
>   #ifdef GNU_LINUX
> 		/* getsockopt(,,SO_ERROR,,) is said to hang on some systems.
> 		   So only use it on systems where it is known to work.  */
> 		{
> 		  int xlen = sizeof(xerrno);
> 		  if (getsockopt(channel, SOL_SOCKET, SO_ERROR, &xerrno, &xlen))
> 		    xerrno = errno;
> 		}
>   #else
>
> Would you please provide an alternative code (similar to what the
> #else branch does in wait_reading_process_output) that will not use
> getsockopt?

getsockopt is not Linux specific; it's pretty much part of the BSD
socket API and is available on Windows too.  Why don't you just link it
in which would simplify wait_reading_process_output too?  Why do the
Unix ports have to pay the price for the Window port?

Helmut