Bug#710336: some observations
Hi,
On Sat, Jun 01, 2013 at 01:22:18PM +0200, Rene Engelhard wrote:
> 09:35 < suihkulokki> _rene_: do you have any idea why graphite2 ftbfs's on arm
> 09:35 < suihkulokki> _rene_: I take it is due to unaligned access since it's also failing on sparc?
> [...]
> 10:07 < _rene_> suihkulokki: was never reaally debugged...
> 10:07 < _rene_> suihkulokki: can be, but why does armhf then build?
> [...]
> 10:17 < suihkulokki> _rene_: armv7+ can do unaligned memory access
> 10:17 < _rene_> ah, hmm
> 10:18 < suihkulokki> this issue has become more interesting now that texlive-bin is blocked by lack of libgraphite2-dev
> 10:18 < _rene_> lol
> 10:18 < _rene_> ok
> 10:19 < _rene_> good that he enabled it knowing it's not there on all archs
> 10:19 < _rene_> if you tell me what to check I can do :)
> 10:19 < _rene_> s/it/it on all archs/
> 10:20 -!- davi [davi@wilbur.25thandclement.com] has joined #debian-arm
> 10:20 < _rene_> (or, if you want, NMU it with my permission)
> 10:22 < suihkulokki> _rene_: if you have an armv5 machine, you can test setting /proc/cpu/alignment to 3 and see if the testsuite passes
> 10:23 < suihkulokki> if you don't have one, I (or someone else here) will look at it later
> 10:23 < _rene_> suihkulokki: sheevaplug
> 10:23 -!- mones_ is now known as mones
> 10:23 < suihkulokki> sheeva is armv5, that should do
> [...]
just noticed I missed a part here:
10:29 < suihkulokki> setting it to 5 will make the code SIGBUS where the unaligned access being made
10:29 < hrw> morning everyone
10:29 < _rene_> will do those tree and put the logs (at least the 5 one) up for download
> 10:41 < _rene_> suihkulokki: with 3 I also get the testuite failing
> 10:53 < _rene_> fails on all three but with a bit of differences
> 10:53 < _rene_> suihkulokki: zyklop.dyndns.org:~/rene/graphite2_1.2.1-1_armel.build.default
> 10:53 < _rene_> suihkulokki: zyklop.dyndns.org:~/rene/graphite2_1.2.1-1_armel.build.3
> 10:53 < _rene_> suihkulokki: zyklop.dyndns.org:~/rene/graphite2_1.2.1-1_armel.build.5
And that is http://zyklop.dyndns.org/~rene/<file>. Damn cut'n'waste back then.
Regards,
Rene
Reply to: