Avoid wrong results for power() with NaN input on some platforms.

Per spec, the result of power() should be NaN if either input is NaN.
It appears that on some versions of Windows, the libc function does
return NaN, but it also sets errno = EDOM, confusing our code that
attempts to work around shortcomings of other platforms.  Hence, add
guard tests to avoid substituting a wrong result for the right one.

It's been like this for a long time (and the odd behavior only appears
in older MSVC releases, too) so back-patch to all supported branches.

Dang Minh Huong, reviewed by David Rowley

Discussion: https://postgr.es/m/75DB81BEEA95B445AE6D576A0A5C9E936A73E741@BPXM05GP.gisp.nec.co.jp

1 files changed, 4 insertions, 3 deletions

diff --git a/src/backend/utils/adt/float.c b/src/backend/utils/adt/float.c
index 6522c0816ef..88215d910a5 100644
--- a/src/backend/utils/adt/float.c
+++ b/src/backend/utils/adt/float.c
@@ -1566,14 +1566,15 @@ dpow(PG_FUNCTION_ARGS)
 	 * pow() sets errno only on some platforms, depending on whether it
 	 * follows _IEEE_, _POSIX_, _XOPEN_, or _SVID_, so we try to avoid using
 	 * errno.  However, some platform/CPU combinations return errno == EDOM
-	 * and result == Nan for negative arg1 and very large arg2 (they must be
+	 * and result == NaN for negative arg1 and very large arg2 (they must be
 	 * using something different from our floor() test to decide it's
 	 * invalid).  Other platforms (HPPA) return errno == ERANGE and a large
-	 * (HUGE_VAL) but finite result to signal overflow.
+	 * (HUGE_VAL) but finite result to signal overflow.  Also, some versions
+	 * of MSVC return errno == EDOM and result == NaN for NaN inputs.
 	 */
 	errno = 0;
 	result = pow(arg1, arg2);
-	if (errno == EDOM && isnan(result))
+	if (errno == EDOM && isnan(result) && !isnan(arg1) && !isnan(arg2))
 	{
 		if ((fabs(arg1) > 1 && arg2 >= 0) || (fabs(arg1) < 1 && arg2 < 0))
 			/* The sign of Inf is not significant in this case. */