--nextPart1326586.FVkT8LQjiv
Content-Type: text/plain; charset=iso-8859-1
Content-Transfer-Encoding: 8Bit

Raymond Meyer wrote:
> Well I Googled and turns out others hade similar problems with isinf on
> Solaris. Something to do with GCC optimising calls to isinf during
> configure tests, so that it returns 0 and it appears as if isinf is
> implemented. I'll look at it in more details when I've got time.

Does attached patch fixes the problem for you?

If you are building from release tarball, just search for isnan(f) and
isinf() from configure script and insert return before them.


regards,

-- 
Hasso Tepper 



--nextPart1326586.FVkT8LQjiv
Content-Type: text/x-diff; name="isnan-isinf.pathc"
Content-Transfer-Encoding: 8Bit
Content-Disposition: attachment; filename="isnan-isinf.pathc"

Index: kjs/configure.in.in
===================================================================
--- kjs/configure.in.in	(revision 661966)
+++ kjs/configure.in.in	(working copy)
@@ -16,7 +16,7 @@
     LIBS="-lm"
     AC_MSG_CHECKING([for isnan with <math.h>])
     AC_TRY_LINK(
-	[#include <math.h>], [float f = 0.0; isnan(f)],
+	[#include <math.h>], [float f = 0.0; return isnan(f)],
 	[AC_MSG_RESULT(yes)
 	    AC_DEFINE([HAVE_FUNC_ISNAN], [1], [Define if you have isnan])],
 	AC_MSG_RESULT(no)
@@ -29,7 +29,7 @@
     LIBS="-lm"
     AC_MSG_CHECKING([for isinf with <math.h>])
     AC_TRY_LINK(
-	[#include <math.h>], [float f = 0.0; isinf(f)],
+	[#include <math.h>], [float f = 0.0; return isinf(f)],
 	[AC_MSG_RESULT(yes)
 	    AC_DEFINE([HAVE_FUNC_ISINF], [1], [Define if you have isinf])],
 	AC_MSG_RESULT(no)


--nextPart1326586.FVkT8LQjiv--