Re: Please gb ktexteditor/armhf
Hi!
> On Sep 12, 2018, at 2:31 PM, Julien Cristau <jcristau@debian.org> wrote:
>
> [cc += debian-arm]
>
>> On 09/09/2018 11:15 PM, Pino Toscano wrote:
>> Hi,
>>
>> the 5.49.0-2 build of ktexteditor failed because two unit tests
>> SIGBUS'ed. OTOH, armel worked, and my tests on the abel armhf porterbox
>> worked fine. (While on harris GCC ICEd really a lot, and I gave up
>> after the 4 ICE in a row...).
>>
>> So please giveback ktexteditor_5.49.0-2/armhf, hoping it was some kind
>> of transient failure...
>>
> I doubt it's transient failure (it failed again), more likely to be the
> fact that arm-arm-01 is arm64 hardware. You may want to try amdahl's
> armhf chroot rather than abel. The arm porters may be able to help too.
It fails on sparc64 as well which is a good indicator that it’s an alignment issue although the crash on sparc64 looks differently.
Generally, code that uses pointer type casts is usually the cause for unaligned access and SIGBUS. Pointer type casts should always be avoided since they can always lead to undefined behavior.
On x86, unaligned access is usually less a problem but it can trigger crashes when using extensions like SSE which has stricter alignment requirements.
I suggest debugging the crash on a porterbox. Unaligned accesses can be normally easily tracked down with gdb.
Adrian
Reply to: