Re: [RFC/AIX] xm-aix4.h - get rid of #undef fd_set & select#include

[Kevin Buettner <kevinb at redhat dot com>]

On Sat, 7 Aug 2004 21:26:01 -0700
Joel Brobecker <brobecker@gnat.com> wrote:

> config/xm-aix4.h contains the following definitions:
> 
> | /* The following text is taken from config/rs6000.mh:
> |  * # The IBM version of /usr/include/rpc/rpc.h has a bug -- it says
> |  * # `extern fd_set svc_fdset;' without ever defining the type fd_set.
> |  * # Unfortunately this occurs in the vx-share code, which is not configured
> |  * # like the rest of GDB (e.g. it doesn't include "defs.h").
> |  * # We circumvent this bug by #define-ing fd_set here, but undefining it in
> |  * # the xm-rs6000.h file before ordinary modules try to use it.  FIXME, IBM!
> |  * MH_CFLAGS='-Dfd_set=int'
> |  * So, here we do the undefine...which has to occur before we include
> |  * <sys/select.h> below.
> |  */
> | #undef fd_set
> | 
> | #include <sys/select.h>
> 
> On our systems (AIX 4.3.2 and 5.1), this type is defined in time.h
> 
>     typedef struct
>     #ifdef _ALL_SOURCE
>     fd_set
>     #endif /* _ALL_SOURCE */
>     {
>       long fds_bits[__NUM_ENTRIES];  /* bitmask for open file descriptors */
>     } fd_set;
> 
> I couldn't find config/rs6000.mh, nor the original comment. So I wonder
> if we wouldn't be able to delete these two.
> 
> I can give it a try, but would I miss something?

I'm all in favor of giving it a try.  I suspect that the the above was
intended as a workaround for a much older version of AIX.  (FWIW, I
can't find the original comment either, but the xm-aix4.h comment and
associated hackery was added in 1996.)

Kevin