Re: Instability in Hydra build

[Rik]

On 04/08/2016 10:57 AM, Mike Miller wrote:
> On Fri, Apr 08, 2016 at 10:31:53 -0700, Rik wrote:
>> 4/7/16
>>
>> All,
>>
>> The Hydra build seems to have been flipping back and forth between success
>> and failure somewhat randomly.  The failure is always a segfault during
>> 'make check' in oct-parse.in.yy-tst.  Usually there isn't much information,
>> but one of the runs caught this
>>
>>   libinterp/parse-tree/oct-parse.in.yy-tst ....................terminate
>> called after throwing an instance of 'std::length_error'
>>   what():  basic_string::_S_create
>> panic: Aborted -- stopping myself...
>> attempting to save variables to 'octave-workspace'...
>> save to 'octave-workspace' complete
>>
>> The failures seem to have become more consistent lately
>> (https://hydra.nixos.org/jobset/gnu/octave-default/all?page=1).  One thing
>> of note is that they all occur on i686 (32-bit) versus the x86_64
>> platform.  Any ideas?  Do we need to try and backout csets like gnulib::pipe?
> I think it's been going on longer than the introduction of gnulib::pipe.
> Eyeballing the build history, the first failure like this seems to be
> around 2016-03-14.
>
>   https://hydra.nixos.org/build/33236900
>
> This build was made with revision
>
>   http://hg.savannah.gnu.org/hgweb/octave/rev/de155ca994d4
>
> I didn't make any changes to the hydra recipe at that time.
>
> Since then it's been segfaulting on and off, but always on the i686
> build like you said. I wonder if it would be worth doing some of the
> leak investigation you and jwe have been doing on a 32-bit build.
>

Maybe so.  Is there an simple way to configure an i686 build on an x86_64
machine (assuming all 32-bit libraries are downloaded)?  Or is it easier to
use a virtual machine, install an i686 OS, and then go from there?  What
are people using for virtual machines these days?  I've used VMware and
Virtual Box in the past, but it is still often a struggle.

--Rik