Add a new regression test, which checks that we handle fp overflow correctly,

and produce a correct infinity.

Currently, this tests fails on 68060 (060sp is to blame) and 88100 processors,
and maybe more.

1 files changed, 32 insertions, 0 deletions

diff --git a/src/regress/lib/libc/ieeefp/infinity/infinity.c b/src/regress/lib/libc/ieeefp/infinity/infinity.c
new file mode 100644
index 0000000000..0f60b4762b
--- /dev/null
+++ b/src/regress/lib/libc/ieeefp/infinity/infinity.c
@@ -0,0 +1,32 @@
+/*      $OpenBSD: infinity.c,v 1.1 2004/01/15 18:53:24 miod Exp $       */
+/*
+ * Written by Miodrag Vallat, 2004 - Public Domain
+ * Inspired from Perl's t/op/arith test #134
+ */
+
+#include <math.h>
+#include <signal.h>
+
+void
+sigfpe(int signum)
+{
+        /* looks like we don't handle fp overflow correctly... */
+        _exit(1);
+}
+
+int
+main(int argc, char *argv[])
+{
+        double d, u;
+        int i;
+
+        signal(SIGFPE, sigfpe);
+
+        d = 1.0;
+        for (i = 2000; i != 0; i--) {
+                d = d * 2.0;
+        }
+
+        /* result should be _positive_ infinity */
+        return ((isinf(d) && copysign(1.0, d) > 0.0) ? 0 : 1);
+}