On Sun, Apr 18, 2010 at 09:04:45PM +0300, Niko Tyni wrote: > Package: perl > Version: 5.12.0-1 > Severity: important > User: debian-p...@lists.debian.org > Usertags: perl-5.12-transition > > The powerpc build failed: > > Failed 5 tests out of 1752, 99.71% okay. > ../cpan/Math-BigInt/t/bare_mbi.t > ../cpan/Math-BigInt/t/bigintpm.t > ../cpan/Math-BigInt/t/sub_mbi.t > op/inc.t > op/pack.t > > The failures are almost certainly related to -Duselongdouble, as > they didn't show up on 5.12.0~rc3-1, which only had -Duse64bitint.
At least op/pack.t also fails on 5.10.1 with -Duselongdouble, so this is not a 5.12 regression. I see the 128-bit long doubles are quite special on ppc32: they are actually a sum of two 64-bit doubles, which is not IEEE compliant [1] and "has serious arithmetic anomalies" [2]. While this should certainly be forwarded upstream, I doubt we can do anything else for 5.12.0 than disable -Duselongdouble on powerpc. [1] http://en.wikipedia.org/wiki/Long_double [2] http://gcc.gnu.org/ml/gcc-help/2006-10/msg00138.html -- Niko Tyni nt...@debian.org -- To UNSUBSCRIBE, email to debian-bugs-dist-requ...@lists.debian.org with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org
