Bug #130720 for Math-BigInt: 1.999817 breaks the tests of CryptX, Math-BigInt-Lite, and Math-Counting

Wed Oct 16 21:09:11 2019 ANDK [...] cpan.org - Ticket created

Subject:

1.999817 breaks the tests of CryptX, Math-BigInt-Lite, and Math-Counting

As per subject. Involved distros: MIK/CryptX-0.064.tar.gz GJACKLAM/Math-BigInt-Lite-0.18.tar.gz MENE/Math-Counting-0.1305.tar.gz A sample fail report for CryptX: http://www.cpantesters.org/cpan/report/d7f2f25c-ed85-11e9-84a2-b83f75811fd8 (at the moment not reachable). Sample fail reports for Math-BigInt-Lite and Math-Counting are currently not publicly available, since cpantesters has a major outage.

Thu Oct 17 10:28:07 2019 peter.john.acklam [...] gmail.com - Correspondence added

Thanks for notifying me! I tested Math-BigInt-1.999817 against Math-BigInt-Lite, so that is a fail that suprises me. Anyway, I will look into this.

Thu Oct 17 10:28:08 2019 The RT System itself - Status changed from 'new' to 'open'

Fri Oct 18 04:50:16 2019 peter.john.acklam [...] gmail.com - Correspondence added

All failed tests in CryptX, Math-BigInt-Lite, and Math-Counting are due to a change in the bpow() method (and the overloaded ** operator) regarding non-integer results. When a Math::BigInt operation results in a non-integer, the output is either truncated (i.e., rounded towards zero) or upgraded, if upgrading is enabled. The bpow() method did not adhere to this and returned NaN for things like 2**(-3). Compare this to explicitly computing 1/8, which returned 0. In 1.999817 this inconsistency was removed. So now, 2**(-3) and 1/(2**3) give the same result, i.e., 0. I consider this change in Math-BigInt to be a bug fix. I suggest the test suits for CryptX, Math-BigInt-Lite, and Math-Counting are updated to reflect this improvement of Math-BigInt.

Fri Oct 18 22:52:55 2019 ANDK [...] cpan.org - Correspondence added

Could you find out why your own testing with Math-BigInt-Lite did not fail? Maybe there's something hidden below the surface?

Sat Oct 19 02:24:47 2019 peter.john.acklam [...] gmail.com - Correspondence added

On Fri Oct 18 22:52:55 2019, ANDK wrote: Show quoted text

> Could you find out why your own testing with Math-BigInt-Lite did not > fail? Maybe there's something hidden below the surface?

I was unfortunately testing against my private Math-BigInt-Lite repo, which had been modified since the 0.18 release to reflect the changes to bpow()/**. When I test against the 0.18 release I see all the same failed tests.

Wed Oct 23 11:40:47 2019 gene [...] cpan.org - Correspondence added

Thanks for the heads up. Fixed in Math-Counting-0.1306 now. - GENE