Bug #3813 for DateTime: test 20infinite.t also failed in win32 (w2k) A/S 5.8.0 version 0.1702

[guest - Fri Sep 19 00:32:24 2003]: Show quoted text
line 61) Show quoted text
line 61) Show quoted text
If negative epoch times fail on some OSs (such as Win32) and tests are skipped, perhaps this test should be skipped as well? Otherwise the test still fails on some Win32 machines.

[autarch@urth.org - Tue Mar 8 17:50:06 2005]: Show quoted text
Not if the tests don't pass and the module isn't installed becasue of that. How many more people should have to see a failing module test and go here to find this ticket which is now more than two years old? Just asking... /J

[autarch@urth.org - Mon Nov 21 09:56:01 2005]: Show quoted text
Ok, sorry if I came off as whiny. I don't know if it's any excuse, but my recation came from the impression that you didn't care about the problem, which I thought was a shame since the module is too useful to be hindered by this. My bad. Anyway, I poked around a little and found two things, both related to the tests itself (I think): 1) I wonder if the magic in Test::More isn't playing tricks with the overloading, which makes it unsuitable to use is() to compare the numbers. If 20infinite.t, line 61 uses cmp_ok with '==' instead, it seems to work better. cmp_ok( $deltas{$_}, '==', $nan, "infinity - infinity = nan ($_)" ); 2) The second thing is the Windows specific fix mentioned on line 17 # for some reason, Windows only gets NaN if abs() is used I'm not sure which Perl version/OS version etc that came from, but using winxp and Perl "v5.8.6 built for MSWin32-x86-multi-thread", it works correctly if I _don't_ use the abs() (so that comment was very helpful!) my $nan = $posinf - $posinf; This second thing of course needs to be tested on a few more combinations before anything can be said for certain, but for the original poster it was broken with w2k + 5.8.0. So I guess the question is whether these two things actually work, like for real usage of the module outside of the test suite, or if the infinite stuff is still broken for real on win32. I don't feel competent enough to answer that question, not knowing the module well enough. There's a patch at the end. If it doesn't break anything else and you decide to apply it, that comment about Windows should perhaps be rephrased to mention the change, in case the previous approach worked for previous versions of something (OS/Perl). /J Index: t/20infinite.t =================================================================== --- t/20infinite.t (revision 1) +++ t/20infinite.t (working copy) @@ -15,7 +15,7 @@ my $posinf = 100 ** 100 ** 100; my $neginf = -1 * $posinf; # for some reason, Windows only gets NaN if abs() is used -my $nan = abs( $posinf - $posinf ); +my $nan = $posinf - $posinf; # infinite date math { @@ -58,7 +58,7 @@ my @compare = $is_win32 ? ( qw( days seconds ) ) : ( qw( days seconds nanoseconds ) ); foreach (@compare) { - is( $deltas{$_}, $nan, "infinity - infinity = nan ($_)" ); + cmp_ok( $deltas{$_}, '==', $nan, "infinity - infinity = nan ($_)" ); } my $new_pos = $pos->clone->add( days => 10 );