Kaveh R. Ghazi wrote:
Okay, thanks both of you for tracking this down.
So the solaris definition of isinf is not exception-safe in the presence
of nans. Try inserting this code into c99-math-double-1.c after it
includes math headers. (This is what fixincludes puts into the header
for solaris 10). See if it helps:
#undef isinf
#define isinf(x) __extension__ ({ const __typeof (x) __x_i = (x); \
__builtin_expect(sizeof(__x_i) == sizeof(float) \
? isgreater(__builtin_fabsf(__x_i),__FLT_MAX__) \
: sizeof(__x_i) == sizeof(long double) \
? isgreater(__builtin_fabsl(__x_i),__LDBL_MAX__) \
: isgreater(__builtin_fabs(__x_i),__DBL_MAX__), 0); })
It may pass isinf, but fail on a later if-abort. But hopefully this is
the only broken bit in that header on solaris.
[ultra10:gcc/gcc/testsuite] andreast% svn diff gcc.dg/c99-math-double-1.c
Index: gcc.dg/c99-math-double-1.c
===================================================================
--- gcc.dg/c99-math-double-1.c (revision 132096)
+++ gcc.dg/c99-math-double-1.c (working copy)
@@ -3,6 +3,13 @@
#include <math.h>
#include "c99-math.h"
+#undef isinf
+#define isinf(x) __extension__ ({ const __typeof (x) __x_i = (x); \
+__builtin_expect(sizeof(__x_i) == sizeof(float) \
+ ? isgreater(__builtin_fabsf(__x_i),__FLT_MAX__) \
+ : sizeof(__x_i) == sizeof(long double) \
+ ? isgreater(__builtin_fabsl(__x_i),__LDBL_MAX__) \
+ : isgreater(__builtin_fabs(__x_i),__DBL_MAX__), 0); })
int main(void)
{
The test does complete w/o abort.
Also the other two tests complete (applied the above on them too).
gmake RUNTESTFLAGS='dg.exp=c99-math-*' check
=== gcc Summary ===
# of expected passes 6
Thanks,
Andreas
