clone 501825 -1 reassign -1 perl 5.10.0-16 retitle -1 sv_chop() alignment problems on sparc severity -1 important thanks
On Wed, Oct 15, 2008 at 11:39:52PM +0300, Niko Tyni wrote: > On Fri, Oct 10, 2008 at 09:00:21PM +0200, Martin Zobel-Helas wrote: > > Package: libjson-xs-perl > > Version: 2.23-1 > > Severity: serious > > > > There was an error while trying to autobuild your package: > > > > > Automatic build of libjson-xs-perl_2.23-1 on schroeder by sbuild/sparc > > > 99.99 > > > Build started at 20081010-1421 > > > > t/19_incr...............dubious > > > Test returned status 0 (wstat 10, 0xa) > > > DIED. FAILED test 697 > > > Failed 1/697 tests, 99.86% okay Here's a status update on this; real life is currently leaving me little hack time, so it's proceeding quite slowly. It doesn't apply to Lenny, though, so it's not very urgent. As I wrote earlier, this is an alignment problem on sparc where gcc -O2 is using double word instructions on a struct that should be 64-bit aligned but isn't. The alignment is lost in libperl's sv_chop() function. As far as I understand, this happens because of the 'OOK hack', which upgrades the SV into an SvPVIV, moving the value into a non-aligned location. I think it's clear that this is a bug in the Perl core rather than JSON-XS itself, so cloning and reassigning. The build failure can be worked around by lowering the optimization level on sparc to -O1 or -O0, disabling the use of double-word instructions. I've reproduced the bug with a 'pristine' non-Debian Perl 5.10.0. It's gone in bleadperl; bisecting shows it was fixed with change 32836, which reimplements the OOK hack in a different way that doesn't upgrade the SV anymore. It's still present in the current maint-5.10 branch, as of change 34715. I'm currently working on a minimal testcase for a proper Perl bug report. Ideally, this would be a simple C program using libperl and cutting JSON-XS from the loop. I've discussed this a bit with Marc Lehmann (Cc'd), the JSON-XS author. Unfortunately bugs.debian.org is silently discarding his messages because their envelope-sender is [EMAIL PROTECTED], matching FROM_DAEMON, so the correspondence is not visible in the bug log. -- Niko Tyni [EMAIL PROTECTED] -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
