Paranoia about unordered comparisons in IEEE float math. If we are

given values that compare as unordered, make sure we reply that they
are equal, which is better than giving an arbitrary answer --- at least
it doesn't depend on which one is passed as which arg.

src/backend/utils/adt/float.c

@@ -8,7 +8,7 @@
  *
  *
  * IDENTIFICATION
- *	  $Header: /cvsroot/pgsql/src/backend/utils/adt/float.c,v 1.71 2001/05/03 19:00:36 tgl Exp $
+ *	  $Header: /cvsroot/pgsql/src/backend/utils/adt/float.c,v 1.72 2001/06/02 17:12:12 tgl Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -566,10 +566,10 @@ float4_cmp_internal(float4 a, float4 b)
 	{
 		if (a > b)
 			return 1;
-		else if (a == b)
-			return 0;
-		else
+		else if (a < b)
 			return -1;
+		else
+			return 0;
 	}
 }
 
@@ -662,10 +662,10 @@ float8_cmp_internal(float8 a, float8 b)
 	{
 		if (a > b)
 			return 1;
-		else if (a == b)
-			return 0;
-		else
+		else if (a < b)
 			return -1;
+		else
+			return 0;
 	}
 }