Bug#807038: Perl 5.22.1-RC3 is now available!

[Craig A. Berry]

On Fri, Dec 4, 2015 at 12:24 PM, Niko Tyni <ntyni at debian.org> wrote:
> On Fri, Dec 04, 2015 at 10:28:50AM -0600, Craig A. Berry wrote:
>> On Fri, Dec 4, 2015 at 6:42 AM, Niko Tyni <ntyni at debian.org> wrote:
>
>> > t/op/sprintf2 ................................................. # Failed test 1463 - at op/sprintf2.t line 671
>> > #      got '0x1p+1020
>> > # '
>> > # expected /(?^:^0x1.0{522}1p\+1020$)/
>
>> FWIW, on Darwin PPC these tests are skipped:
>>
>> ok 1463 # skip darwin doublekind=4
>> ok 1464 # skip darwin doublekind=4
>> ok 1465 # skip darwin doublekind=4
>> ok 1466 # skip darwin doublekind=4
>> ok 1467 # skip darwin doublekind=4
>> ok 1468 # skip darwin doublekind=4
>>
>> What does Debian PPC show for:
>>
>> % ./perl -Ilib -V:doublekind
>> doublekind='4';
>
> doublekind='4'; looks like those tests get run only on linux
> with that doublekind.

Right.  I hadn't noticed the linux-only part before.  These tests came
in with a fix for a problem reported on Debian PPC64:

<http://perl5.git.perl.org/perl.git/commit/5488d3733162ee806bb5f5c55694e8beaaf7b1cc?f=t/op/sprintf2.t>

<https://rt.perl.org/Public/Bug/Display.html?id=125633>

To me the configuration in that ticket looks awfully similar to the
configuration that now sees test failures, but I could be missing
something.

On 32-bit Darwin PPC (where the tests are skipped), I see the same
output running the test code as is in your log.  E.g., your log has:

t/op/sprintf2 ................................................. #
Failed test 1451 - at op/sprintf2.t line 671
#      got '0x1p+1020
# '
# expected /(?^:^0x1.0{522}1p\+1020$)/

and I get:

./perl -Ilib -e 'printf("%La\n", (2**1020) + (2**-1072));'
0x1p+1020


> I'm attaching 'perl -V' output for all the failing platforms: powerpc
> s390x and ppc64.  All are big-endian, and the latter two are 64-bit. All
> have doublekind=4.
> --
> Niko Tyni   ntyni at debian.org