bug#18237: Small fix for MSDOS

[Reuben Thomas]

On 10 August 2014 19:02, Eli Zaretskii <eliz@gnu.org> wrote:

> Date: Sun, 10 Aug 2014 17:40:24 +0100
> From: Reuben Thomas <rrt@sc3d.org>
>
> The following patch helps in the case when a sockets library is installed;
> in my case, ls080b.zip from the DJGPP site:
>
> === modified file 'src/msdos.c'
> --- src/msdos.c    2014-08-09 16:12:33 +0000
> +++ src/msdos.c    2014-08-09 21:51:56 +0000
> @@ -4016,7 +4017,7 @@
>  #endif
>
>
> -#ifndef HAVE_SELECT
> +#ifndef HAVE_SYS_SELECT_H
>  #include "sysselect.h"
>
>  /* This yields the rest of the current time slice to the task manager.
>
>
> The reasoning is that having select doesn't imply having the sys/select.h
> functionality that the following code emulates.
>
> Does that look OK to install?

You didn't tell enough details, so I don't understand why the change
is needed, and why it is OK not to test HAVE_SELECT at all here.  (I
don't have ls080b.zip installed.)  Please tell what problems did you
see with the original code, and let's take it from there.

libsocket is a sockets implementation: http://rich.phekda.org/richdawe/lsck/lsck_dl.htm

When it is installed, configure correctly detects that select is available, and sets HAVE_SELECT. However, libsocket has no sys/select.h, nor does it implement sys_select.

Hence, at present, when I build using ./configure && make for DOS with libsocket installed, HAVE_SELECT is defined to 1, so the MS-DOS implementation of sys_select in msdos.c is not compiled. Later, the linker is unable to find sys_select, and the build fails.

I think you're also saying that I should test both HAVE_SELECT && HAVE_SYS_SELECT_H therefore? You're probably right, I (incorrectly?) assumed that if HAVE_SYS_SELECT_H is true, then HAVE_SELECT would be too.

--
http://rrt.sc3d.org