[vtk-developers] Gil tests on megas

[Ben Boeckel]

On Tue, Nov 17, 2015 at 12:53:09 +0100, Mathieu Westphal wrote:
> So it may be related to my fiddling with python.

So this problem has cropped up occasionally on the builder, seemingly
randomly. I don't know the root cause (IIRC, it occurs in Python2 as
well, but if we have a reproducer, tracking it down while we have it
happening reliably would be good).

> But i'm unable to reproduce it manually on megas.

I believe it is related to the "_strict" feature megas uses which sets
some environment variables:

  - 'MALLOC_PERTURB_': str(random.randint(1, 255)),
  - 'MALLOC_CHECK_': '3',
  - 'MEMCPY_CHECK_': '1',

which means that memory is getting junked somewhere. Valgrind should
have more information.

> Which fails with an Xerror
> 
> - Executing it with DISPLAY=:0 fail on image size
> - Executing it with DISPLAY=:1 simple pass the test !

DISPLAY :0 is my user's X (it is my main development machine); :1 is the
VNC server the dashboards should be using.

> - Repushing on my branch, waiting for the buildbot, same tests still fails !
> 
> May be i'm doing something not rigth here, any help apreciated.

I think valgrind is the next step.

> PS: I unvoluntarly deleted the kitware crontab, but it was pointing on a
> non existente file, so i supose it's ok.

Yeah, the crontab is probably obsolete, thanks.

--Ben