Hopefully this provides more portable values for Infinity and NaN in Perl.

git-svn-id: https://swig.svn.sourceforge.net/svnroot/swig/trunk@10616 626c5289-ae23-0410-ae9c-e8d60b6d4f22

Examples/test-suite/perl5/li_typemaps_runme.pl

@@ -37,24 +37,29 @@ batch('ulong', 0, 1, 12, 0xffffffff);
 batch('uchar', 0, 1, 12, 0xff);
 batch('schar', -0x80, 0, 1, 12, 0x7f);
 
-# IEEE 754 machine, please!
+{
-batch('float',
+	use Math::BigInt qw();
+	# the pack dance is to get plain old NVs out of the
+	# Math::BigInt objects.
+	my $inf = unpack 'd', pack 'd', Math::BigInt->binf();
+	my $nan = unpack 'd', pack 'd', Math::BigInt->bnan();
+	batch('float',
 	  -(2 - 2 ** -23) * 2 ** 127,
 	  -1, -2 ** -149, 0, 2 ** -149, 1,
 	  (2 - 2 ** -23) * 2 ** 127,
-  'nan');
+	  $nan);
-{ local $TODO = "shouldn't some Inf <=> float work?";
+	{ local $TODO = "float typemaps don't pass infinity";
-  # I'm going to guess that it could work reasonably as
+	  # it seems as though SWIG is unwilling to pass infinity around
-  # NV Inf => float Inf
+	  # because that value always fails bounds checking.  I think that
-  # float Inf => NV NaN
+	  # is a bug.
-  # but this needs some thought.
+	  batch('float', $inf);
-  batch('float', 'inf');
+	}
-}
+	batch('double',
-batch('double',
 	  -(2 - 2 ** -53) ** 1023,
 	  -1, -2 ** -1074, 0, 2 ** 1074,
 	  (2 - 2 ** -53) ** 1023,
-  'nan', 'inf');
+	  $nan, $inf);
+}
 batch('longlong', -1, 0, 1, 12);
 batch('ulonglong', 0, 1, 12);
 SKIP: {