dfilter: Add infinity and NaNs to the float token

Parse INF and INFINITY (case insensitively) and NAN and
NAN(char_sequence) as strtod does, bringing back support for
floating point infinity and nan literals which was accidentally
removed in 881dec9

Also add a note about NaN comparisons (right now we make NaNs
compare equal with everything, including non-NaNs.)

Add some tests with infinity and NaN

epan/dfilter/scanner.l

@@ -501,6 +501,18 @@ HexExponent		([pP][+-]?[[:digit:]]+)
 	return set_lval_float(yyextra, yytext, yytext);
 }
 
+(?i:inf)(?i:inity)? {
+	/* Infinity. */
+	update_location(yyextra, yytext);
+	return set_lval_float(yyextra, yytext, yytext);
+}
+
+(?i:nan)(\([[:alnum:]_]*\))? {
+	/* NaNs (including quiet NaNs). */
+	update_location(yyextra, yytext);
+	return set_lval_float(yyextra, yytext, yytext);
+}
+
 :[[:xdigit:]]+	{
 	/* Numeric prefixed with ':'. */
 	update_location(yyextra, yytext);

epan/ftypes/ftype-double.c

@@ -118,6 +118,12 @@ val_divide(fvalue_t * dst, const fvalue_t *a, const fvalue_t *b, char **err_ptr
 static enum ft_result
 cmp_order(const fvalue_t *a, const fvalue_t *b, int *cmp)
 {
+	/* In C, NaNs compare unordered with everything, including the same NaN.
+	 * The below makes all NaNs compare equal with everything. We could make
+	 * NaNs ordered by having negative NaNs below -inf and positive NaNs above
+	 * inf. C23 adds the totalorder function (which distinguishes NaNs from
+	 * each other by payload and also distinguishes -0 from 0.)
+	 */
 	if (a->value.floating < b->value.floating)
 		*cmp = -1;
 	else if (a->value.floating > b->value.floating)

epan/proto.c

@@ -522,6 +522,7 @@ static const char *reserved_filter_names[] = {
 	"false",
 	"nan",
 	"inf",
+	"infinity",
 	NULL
 };
 

test/suite_dfilter/group_double.py

@@ -89,3 +89,22 @@ def test_le_2(self, checkDFilterCount):
     def test_le_3(self, checkDFilterCount):
         dfilter = "icmp.resptime <= 492"
         checkDFilterCount(dfilter, 0)
+
+    def test_inf_1(self, checkDFilterCount):
+        dfilter = "icmp.resptime < inf"
+        checkDFilterCount(dfilter, 1)
+
+    def test_inf_2(self, checkDFilterCount):
+        dfilter = "icmp.resptime > -infinity"
+        checkDFilterCount(dfilter, 1)
+
+    def test_inf_3(self, checkDFilterCount):
+        # A protocol can't have the name inf or infinity, but a field can
+        # This is just to check that the filter compiles without error
+        dfilter = "dvmrp.infinity == 255"
+        checkDFilterCount(dfilter, 0)
+
+    def test_nan(self, checkDFilterCount):
+        # XXX - We compare NaNs oddly
+        dfilter = "icmp.resptime == nan"
+        checkDFilterCount(dfilter, 1)