Overloading atan2 occurs in Math::BigInt. It's obvious that bignum is getting the order of arguments wrong. That smells like an improper use of overload in the BigInt module.
Going on that theory, lets try patching Math::BigInt to account for the third argument to an overloaded binary operator.
Looking at Math-Bigint-1.74, the most recent, we see.
use overload {
# . . . other operators
'atan2' => sub { atan2($_[0]->numify(),$_[1]) },
# . . . the rest of them
};
Clearly, no attention is paid to the argument order. Here's a patch:
--- BigInt.pm~ Sat Jan 1 10:57:35 2005
+++ BigInt.pm Sun Jan 23 22:04:36 2005
@@ -75,7 +75,9 @@
'cos' => sub { cos($_[0]->numify()) },
'sin' => sub { sin($_[0]->numify()) },
'exp' => sub { exp($_[0]->numify()) },
-'atan2' => sub { atan2($_[0]->numify(),$_[1]) },
+'atan2' => sub { $_[2] ?
+ atan2($_[1],$_[0]->numify()) :
+ atan2($_[0]->numify(),$_[1]) },
# are not yet overloadable
#'hex' => sub { print "hex"; $_[0]; },
That is not applicable with patch as is, since I've converted tabs to spaces for web presentation. Use it as a guide to hand-edit the file.
A quick grep for "atan2" in t/ reveals that the only tests of atan2 occur in fallback.t. The tests there are not too rigorous, only testing atan2 for equal values of the arguments, and testing twice for the same values. I'll patch that, too.
--- fallback.t~ Thu Dec 11 14:17:58 2003
+++ fallback.t Sun Jan 23 22:36:17 2005
@@ -28,7 +28,7 @@
}
print "# INC = @INC\n";
- plan tests => 8;
+ plan tests => 9;
}
# The tests below test that cos(BigInt) = cos(Scalar) which is DWIM,
+but not
@@ -46,10 +46,11 @@
ok (exp($bi), exp(1));
ok (atan2($bi,$bi), atan2(1,1));
-my $bf = Math::BigInt->new(1);
+my $bf = Math::BigInt->new(0);
-ok (cos($bf), cos(1));
-ok (sin($bf), sin(1));
-ok (exp($bf), exp(1));
-ok (atan2($bf,$bf), atan2(1,1));
+ok (cos($bf), cos(0));
+ok (sin($bf), sin(0));
+ok (exp($bf), exp(0));
+ok (atan2($bi,$bf), atan2(1,0));
+ok (atan2($bf,$bi), atan2(0,1));
With these changes to lib/Math/BigInt.pm and t/fallback.t, Math::BigInt builds correctly and passes all tests.
With that installed, now I get for your problem,
$ perl -e'use bignum;print atan2(1,0), " ", atan2(0,1),$/'
1.5707963267949 0
$
in complete agreement with reality.
I'll review Math::BigInt for any other non-commutative binary opertors, and post the patches to rt.cpan.org.
|