Re: #if vs. #ifdef -- again (was: AC_CHECK_SIZEOF failing on undefined HAVE_STDINT_H with -Werror)

[nash]

Wouldn't it be preferrable to use #if so that a config.h token could
be an expression that variously evaluates to T/F depending on other
stuff? For example,

In config.h somewhere:
 #define HAVE_WORKING_MMAP (defined(HAVE_MMAP) && defined(HAVE_MUNMAP))

In the program:
 #if HAVE_WORKING_MMAP
 ... stuff ...
 #else
 ... different stuff  ...
 #endif

Stupid example but the point is that using #ifdef wouldn't work for
this. Seems like a program checking a config.h identifier shouldn't
really have to know whether it evaluates to an expression, or not.


-nash

On 8/15/06, Stepan Kasal <address@hidden> wrote:
> Hello,
>
> On Tue, Aug 15, 2006 at 03:45:13PM +0200, Ralf Wildenhues wrote:
> > Neither can I, but there has been a move in some GNU packages (Autoconf,
> > Gnulib) from #ifdef to #if, which I've never understood.  If we want a
> > move back, we need to know the incentives that caused it in the first
> > place, and see whether the reasons for it are (still) valid, to avoid
> > bouncing back and forth.
>
> that's why asked here.  Let's wait for an answer from Paul.
>
> > I believe the feedback loop on some of the more obscure issues w.r.t.
> > Autoconf can easily execed a couple of years.  Not sure if this is one
> > of them.
>
> Well, the move from #ifdef to #if happened sometimes around 2000.
>
> Autoconf 2.50 already used #if in all code (w/ 1 exception).
> Autoconf 2.13 used #ifdef in the code almost exclusively (w/ one
> exception), but the manual used #if.
>
> Stepan
>
>
> _______________________________________________
> Autoconf mailing list
> address@hidden
> http://lists.gnu.org/mailman/listinfo/autoconf