Re: FYI: 17-gary-evalmp-libgmp-autoconfiscation.patch

[Akim Demaille]

| Index: config/gmp.m4
| ===================================================================
| RCS file: /cvsroot/m4/m4/config/gmp.m4,v
| retrieving revision 1.3
| diff -u -p -u -r1.3 gmp.m4
| --- config/gmp.m4 2001/09/20 08:49:31 1.3
| +++ config/gmp.m4 2001/09/21 00:12:58
| @@ -16,44 +16,52 @@
|  # Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA
|  # 02111-1307  USA
|  
| -# serial 4
| +# serial 5
|  
|  m4_define([_AC_LIB_GMP],
| -[AC_ARG_WITH(gmp,
| -[  --without-gmp           don't use GNU multiple precision arithmetic 
library],
| -[use_gmp=$withval], [use_gmp=yes])
| -

Hm... Considering that I would also like to find some general
scheme/approach to macros for such libraries, considering that I would
like to promote generic macros (in the sense of `valid for several
packages') vs dedicated macros, I would have preferred keeping the
--with.


+#if !USE_GMP
 
+M4INIT_HANDLER (mpeval)
+{
+  const char s[] = "libgmp support was not compiled in";
+
+  if (obs)
+    obstack_grow (obs, s, strlen(s));
+}
+
+#else /* USE_GMP */

This I don't understand too well.  Why building a fake module instead
of not building it.

The current module interface is lacking, IMHO, one feature: the
ability to ask whether a module *exists*.  Or a means to load and
catch error when loading fails.

But then, if you create a broken module, there is no use.

Also, why should an error message such `as libgmp support was not
compiled in' be in regular output instead of stderr?