Re: new macro: AC_C_RESTRICT

[Akim Demaille]

Paul Eggert said:

> Jim Meyering <address@hidden> writes:
>
>> Comments/suggestions?
>
> Looks good, thanks.  However, could you please check for "__restrict"
> before "__restrict__"?  GCC supports either spelling, but "__restrict"
> is shorter and older DEC Alpha C compilers support only "__restrict".
> The handwritten "configure" script for MPlayer
> <http://www.mplayerhq.hu> checks for "__restrict" first, so this
> should be reliable.
>
> Also, please check for "_Restrict" as well.  That is the only spelling
> supported by Sun WorkShop 6 update 2 C.

Wow...  There is already that many different names...

> Jim Meyering <address@hidden> writes:
>
>> I don't like the name very much, since it implies C-only support,
>> while the macro is useful/necessary also with C++ compilers.
>
> Aren't the other AC_C_ macros in that camp too?

Yes, indeed.  That's why I would like to suggest another framework,
which I think Autoconf should have followed anyway: config.h should
encode the result of the system audit only, and not the actions.
In the present case, I would much prefer if the macro could AC_DEFINE
HAVE_C_RESTRICT or HAVE_C___RESTRICT, or maybe AC_DEFINE
C_RESTRICT as undefined, "restrict", "__restrict" etc. and then, let
system.h make the #define.  Similarly, I don't think it was wise
to have config.h #define the missing types.

Of course this is introducing a brand new interface in Autoconf.
But I think it is a good one as:
- it is cleaner for other languages
  (remember the complaints of Fortran people who see such results
  in their config.h)
- it is cleaner for C vs. C++ etc.
  (C_RESTRICT and CXX_RESTRICT are different, it is up to the user
   to use them correctly).