[GSOC] strace testing in docker at travis-ci.org

[Philippe Ombredanne]

On Thu, May 12, 2016 at 1:55 PM, Yun-Chih Chen <b03902074 at ntu.edu.tw> wrote:
>> My advice: do not run docker. I think this is a setting in the Travis file.
>> Otherwise the number of indirections becomes a tad too large:
>> Travis may itself run in VMs -> which runs containers --> in which
>> your run QEMU ....
>> A few too many moving parts IMHO.
>>
>> So your goal is to limit the number of these and remove docker from
>> the equation.
>
> Thanks for the advice.  Unwrapping these layers is a good idea.  But I
> think it's still helpful to know why tests fail on those
> architectures.  I had another repo that tests strace with
> QEMU+schroot, although many QEMU segfaults are still present.

Once you have docker out of the equation, and other config issues
out of the way, a test failure is definitely a bug.
It needs to be investigated and eventually fixed...
It could be either a bug strace OR a bug in the strace test for this
architecture
(such as trying something that is not supported at all by that arch)
This was somehow expected that you would discover new bugs :)

Now, before diving in more exotic archs, I think you should first get
something stable with more common ones. Then you can tackle the
off-broadway archs.

>> As for the failure to build on various architectures, are you sure the
>> toolchain you use is compiled with support for these? It looks like it
>> from the Travis matrix though.
>
> I can confirm that "gcc-4.6" ( in wheezy docker ), "gcc-4.9" ( in
> jessie docker ) and "libc6-dev" are indeed installed within docker and
> are not from Travis's Trusty box.  Please let me know if I'm mistaken.
> ( :

You are not.
-- 
Cordially
Philippe Ombredanne