[gfortran, patch] provide isfinite as a macro

[François-Xavier Coudert <fxcoudert at gmail dot com>]

Right now, libgfortran.h provides a fallback implementation of
isfinite() when it is not available from the system. This
implementation has prototype "int isfinite (double)", which is wrong:
isfinite is supposed to be a macro, accepting float, double and long
double types. Attached patch fixes this problem, and removes the
fallback use of finite(), which can only accept doubles.

This patch fixes a problem with large real(16) numbers on solaris. It
is regtested on sparc-solaris, and the only other platform where it
could go wrong is cygwin, which i'll regtest this evening.

OK for mainline and 4.0 (given that cygwin regtests correctly)?

FX

:ADDPATCH libgfortran:

Attachment: isfinite.ChangeLog
Description: Binary data

Index: libgfortran/libgfortran.h
===================================================================
RCS file: /cvsroot/gcc/gcc/libgfortran/libgfortran.h,v
retrieving revision 1.27
diff -p -u -r1.27 libgfortran.h
--- libgfortran/libgfortran.h   7 Jul 2005 22:07:53 -0000       1.27
+++ libgfortran/libgfortran.h   3 Aug 2005 14:15:49 -0000
@@ -179,17 +179,11 @@ typedef off_t gfc_offset;
    alternatives, or bail out.  */
 #if (!defined(isfinite) || defined(__CYGWIN__))
 #undef isfinite
-static inline int
-isfinite (double x)
-{
 #if defined(fpclassify)
-  return (fpclassify(x) != FP_NAN && fpclassify(x) != FP_INFINITE);
-#elif defined(HAVE_FINITE)
-  return finite (x);
+#define isfinite(x) (fpclassify(x) != FP_NAN && fpclassify(x) != FP_INFINITE)
 #else
-#error "libgfortran needs isfinite, fpclassify, or finite"
+#define isfinite(x) (((x) == (x)) && ((2 * (x) != (x)) || ((x) == 0)))
 #endif
-}
 #endif /* !defined(isfinite)  */
 
 /* TODO: find the C99 version of these an move into above ifdef.  */