Re: [bug-gnulib] sys/socket.h on mingw32 vs socklen (resend)

[Bruno Haible]

Simon Josefsson wrote on 2005-12-17:
> The sys_socket module below will create a sys/socket.h file, primarily
> for mingw32, but it could be extended for other systems or missing
> sys/socket.h features in the future.
> 
> This would solve the problem of accessing sys/socket.h stuff in
> applications.

I like this idea. So we can get rid of #if HAVE_SYS_SOCKET_H in many places.

> Index: lib/socket_.h
> ===================================================================
> RCS file: lib/socket_.h
> diff -N lib/socket_.h
> --- /dev/null 1 Jan 1970 00:00:00 -0000
> +++ lib/socket_.h     16 Dec 2005 15:04:41 -0000
> @@ -0,0 +1,26 @@
> +/* Provide a sys/socket header file for systems lacking it (read: mingw32).
> +   Copyright (C) 2005 Free Software Foundation, Inc.
> +   Written by Simon Josefsson.
> +
> +   This program is free software; you can redistribute it and/or modify
> +   it under the terms of the GNU General Public License as published by
> +   the Free Software Foundation; either version 2, or (at your option)
> +   any later version.
> +
> +   This program is distributed in the hope that it will be useful,
> +   but WITHOUT ANY WARRANTY; without even the implied warranty of
> +   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
> +   GNU General Public License for more details.
> +
> +   You should have received a copy of the GNU General Public License
> +   along with this program; if not, write to the Free Software Foundation,
> +   Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA.  */
> +
> +#ifndef _SYS_SOCKET_H
> +#define _SYS_SOCKET_H
> +
> +#if HAVE_WINSOCK2_H
> +# include <winsock2.h>
> +#endif

Comments should indicate for which platforms this file is meant to be useful
and on which platforms you expect <winsock2.h> to be present. (Currently only
mingw, I think?)

Also, this should include <ws2tcpip.h>, so that socklen_t is defined
(and a type redefinition error is avoided when the user includes <ws2tcpip.h>).

> +
> +#endif /* _SYS_SOCKET_H */
> Index: m4/sys_socket_h.m4
> ===================================================================
> RCS file: m4/sys_socket_h.m4
> diff -N m4/sys_socket_h.m4
> --- /dev/null 1 Jan 1970 00:00:00 -0000
> +++ m4/sys_socket_h.m4        16 Dec 2005 15:04:41 -0000
> @@ -0,0 +1,19 @@
> +# sys_socket_h.m4 serial 1
> +dnl Copyright (C) 2005 Free Software Foundation, Inc.
> +dnl This file is free software; the Free Software Foundation
> +dnl gives unlimited permission to copy and/or distribute it,
> +dnl with or without modifications, as long as this notice is preserved.
> +
> +dnl From Simon Josefsson.
> +
> +AC_DEFUN([gl_HEADER_SYS_SOCKET],
> +[
> +  AC_CHECK_HEADERS_ONCE([sys/socket.h])
> +  if test $ac_cv_header_sys_socket_h = yes; then
> +    SYS_SOCKET_H=''
> +  else
> +    AC_CHECK_HEADERS_ONCE([winsock2.h])
> +    SYS_SOCKET_H='sys/socket.h'
> +  fi
> +  AC_SUBST(SYS_SOCKET_H)
> +])

Looks good. Fine.

> Index: modules/sys_socket
> ===================================================================
> RCS file: modules/sys_socket
> diff -N modules/sys_socket
> --- /dev/null 1 Jan 1970 00:00:00 -0000
> +++ modules/sys_socket        16 Dec 2005 15:04:41 -0000
> @@ -0,0 +1,36 @@
> +Description:
> +A <sys/socket.h> for systems lacking it.
> +(Currently only useful for mingw32, it will simply include winsock2.h.)
> +
> +Files:
> +lib/socket_.h
> +m4/sys_socket_h.m4
> +
> +Depends-on:
> +
> +configure.ac:
> +gl_HEADER_SYS_SOCKET
> +
> +Makefile.am:
> +BUILT_SOURCES += $(SYS_SOCKET_H)
> +EXTRA_DIST += socket_.h
> +
> +# We need the following in order to create <sys/socket.h> when the system
> +# doesn't have one that works with the given compiler.
> +sys/socket.h: socket_.h
> +     mkdir sys

Create the directory only when it does not exist. (Otherwise the user is
blocked if he happens to press Ctrl-C after the sys directory has been
created but before sys/socket.h is present.)

> +     cp $(srcdir)/socket_.h address@hidden
> +     mv address@hidden $@
> +MOSTLYCLEANFILES += sys/socket.h sys/socket.h-t
> +
> +mostlylocal-clean:

Should be mostlyclean-local, not mostlylocal-clean.

> +     rmdir sys

Remove the directory only when it exists and is empty. (Some systems create
files in your directories without being asked for, e.g. .DS_Store on MacOS X.)

> +
> +Include:
> +#include <sys/socket.h>
> +
> +License:
> +LGPL
> +
> +Maintainer:
> +Simon Josefsson

OK, good.

> However, it does not solve the problem of accessing sys/socket.h
> inside autoconf tests, and in particular from the socklen
> module.
> 
> Currently the socklen test fail on mingw32:
> 
> checking for socklen_t... no
> checking for socklen_t equivalent...
> configure: error: Cannot find a type to use in place of socklen_t
> 
> because:
> 
> configure:3814: i586-mingw32msvc-gcc -c -g -O2 conftest.c >&5
> conftest.c:23:26: sys/socket.h: No such file or directory
> 
> I can think of three solutions:
> 
> 0) In socklen, default the type to 'int' instead of aborting.  This
>    would be compatible with mingw32, they use 'typedef int socklen_t'.
>    The patch below implement this.
> 
> 1) Fix the socklen M4 test to check for the existence of socket.h,
>    winsock2.h and ws2tcpip.h and include the appropriate headers when
>    needed.  This has the problem that the application will have to
>    include the non-POSIX headers too, if they need socklen_t.  The
>    patch below implement this.
> 
> 2) Create the gnulib sys/socket.h directly in the M4 macro, and set up
>    CFLAGS to include the gnulib directory during ./configure.  This
>    has the problem that all modules invoked before the sys_socket
>    module will not work.  It seem wrong for those modules to depend on
>    "sys_socket", because sys/socket.h is a POSIX header which most
>    tests could assume.

A fourth solution is

3) Set some configure variables during sys_socket_h.m4 that are used in the
socklen.m4 tests.

Ad 0): I don't like this, because socklen_t values are passed through a
pointer, and if the argument type of getpeername() is not 'int *', the
program will break horribly. Don't make the autoconf test less reliable than
it is now.

Ad 1): Having socklen.m4 check for ws2tcpip.h is probably unavoidable, since
it is this header file which defines socklen_t on mingw, and winsock2.h doesn't.
But let the sys_socket_h module do the rest, and let user programs use
only <sys/socket.h>. I don't like the two gnulib modules be so tied together,
but can't see a better solution.

Ad 2): Creating a file directly from within an m4 macro looks wrong. Reasons:
- Filenames don't belong in .m4 files.
- The removal of the file must be triggered by "make distclean", i.e. from a
  Makefile. Then it's more symmetrical if the file is created by the same
  Makefile.

> I think 0 and 1 should be done, together with installing the
> sys_socket module.

I think a mix of 1) and 3) is the best.

> Possibly, socklen should not test for ws2tcpip.h.

I think it needs to.

Bruno