GNOME Bugzilla – Bug 735273
orc-0.4.21 fails tests on i686-linux
Last modified: 2017-11-30 11:54:57 UTC
Created attachment 284281 [details] Both tests failing We've never had errors with 0.4.19. I tested on two machines: both fail on the "test-limits" tests, and one fails on "exec_opcodes_sys".
Created attachment 284282 [details] Log of one test failing
Is this is a duplicate of bug #728129 then?
The "exec_opcodes_sys" part seems same to me, but I see no mention of "test-limits" in there. I have no idea what that test does, and it provides no output.
Could you try and git bisect it from 0.4.19 just to see if it's something in orc that changed (and not the toolchain leading to different results because of operations of undefined behaviour)?
OK, I bisected it, using an always-binary-equal sandbox environment to build+test it. "Surprisingly", the test starts to fail in the commit introducing it :-) 988914384acdf30eab22226d79de3541e7e6f02d If I had a clue this test was introduced in-between those releases, I could save quite some time.
Are you compiling with gcc (which version?) or clang?
It's gcc-4.8.3.
Ok, so a different bug than bug #728129. Can you run "orc-bugreport" and attach the output here?
Orc 0.4.21 - integrated testing tool Active backend: sse L1 cache: 32768 L2 cache: 262144 L3 cache: 3145728 Family/Model/Stepping: 6/37/5 CPU name: Intel(R) Core(TM) i3 CPU M 370 @ 2.40GHz Compiler options: sse2 sse3 ssse3 sse41 sse42 Opcode test: No errors detected. BTW, I hope that builds do not assume orc will be run on the same type of CPU as it was compiled.
They don't, orc detects the CPU at runtime and emits code for that specific CPU.
Does it still fail with latest orc ? Quite a few fixes related to undefined behaviours went in.
I tested with newer toolchain now, but 0.4.21 still fails and 0.4.28 succeeds.