On Fri Feb 18 12:51:20 2011, DOUGDUDE wrote:
Show quoted text> I was looking through the ActiveState build reports
> (
http://code.activestate.com/ppm/JE/) and see that the tests always
> seems to fail under only Windows 32-bit (64-bit seems OK). A CPAN
> testers report with the same failures is at
>
http://www.cpantesters.org/cpan/report/6e686ced-8278-1014-99d0-f8152553c65e
> (the Mac OS X and Solaris failures are due to the tests taking more than
> 300s, apparently).
>
> I'm going to see if I can get a hold of a 32-bit Windows machine to see
> if I can reproduce this and figure out what the problem is (I just have
> 64-bit ones). If you have any insights to what it may be, I'd be happy
> to hear as it may make finding the problem easier :)
I knew about the ActivePerl failures, but I didn’t know that there was a 64-bit vs. 32-bit
distinction for ActivePerl builds, or that JE worked on any version of ActivePerl.
The problem is that it does not support nan and inf. It seems to have to do with what compiler
is used. (See also <
http://rt.perl.org/rt3/Ticket/Display.html?id=78330>.)
I’m not sure it’s worth the effort to implement nan and inf in pure Perl. (If someone writes a
patch, though, that would change things.) Everyone who has brought up the issue so far has
been willing to switch to Strawberry Perl (which just works).
Basically, (32-bit) ActivePerl does not return nan (or -1.#IND) for 0+sin 9**9**9, but -1 or
something else. Similarly for 9**9**9+1.
