[Paraview] Bug 0006714: pvbatch requires X server

[Berk Geveci]

Something is fishy here. We obviously ran examples both on Xt3 and
BlueGene and generated screenshots without X. So, there is a
difference between pvbatch compiled for those platforms and the one
you compiled. I'll discuss this with Pat and Utkarsh tomorrow.

-berk

On Mon, Apr 7, 2008 at 5:35 PM, Kent Eschenberg <eschenbe at psc.edu> wrote:
> Hi Utkarsh,
>
>  Sorry its taken so long to try this. Summary: nothing changed - it still
> fails exactly as before.
>
>  You might want to take a close look at your environment when you test it.
> When you log onto a parallel system I think you still provide a way for it
> to reach an X server. My parallel system also has no X server but it still
> got back to the X server of my workstation via the ssh connection which sets
> the environment variable DISPLAY. When I ran the program I would see windows
> appearing on the screen for a second. However, when I unset the environment
> variable DISPLAY it crashed.
>
>  More importantly, I tried to emulate the situation would would have in a
> true batch environment such as we have on the XT3. There, jobs are submitted
> with the qsub command. What is submitted is a script. qsub is not used on
> the Linux cluster where I am testing this but I can emulate it with the "at"
> command. Example: "at -f cut.job NOW" submits file cut.job for immediate
> execution. That file is simple:
>
>  mpiexec -np 2 pvbatch --use-offscreen-rendering cut.py >& cut.log
>
>  My default environment variables LD_LIBRARY_PATH and PATH select the
> correct version of pvbatch, i.e., the compiled from CVS with MPI. However,
> when submitted this way, the environment variable DISPLAY is not set. The
> above job fails with no output because of the bad X server connection.
>
>  I've attached one of the Python scripts I've been using for testing. I
> might not be setting something in it that is needed.
>
>  Cheers,
>  Kent
>
>
>
>  Utkarsh Ayachit wrote:
>
> > Hi Kent,
> >
> > Attached is patch. With this patch (and OSMesa build) I could do:
> >
> > mpirun -np 2 ./bin/pvbatch --use-offscreen-rendering  /tmp/foo.py
> >
> > on a machine with no X. I still get the  "bad X server connection." from
> the OpenGLExtensionManager -- which I guess you fix using Pat's suggestions.
> However, in spite of the error message, things should work fine.
> >
> > Let me know if this patch works, if so, I'll commit it to the CVS.
> >
> > Utkarsh
> >
>