Re: the nvidia/osmesa dilema

[John W. Eaton]

On 05/04/2016 11:49 AM, Mike Miller wrote:

> I think patches would be welcome to make the configure script
> automatically disable OSMesa if we can detect this situation somehow.

The fix probably needs to be in Octave itself since binaries built on a 
system with working OpenGL libraries/drivers/cards may run on systems 
that fail.  But that's just a workaround for the real problem.

As Rik said at some point, it seems likely that we are doing something 
wrong in the way we are calling OpenGL functions but it is not obviously 
wrong or it would fail on all systems.  When I was debugging, I thought 
maybe we weren't correctly allocating some OpenGL buffer(s), but I 
couldn't see where that was happening, and it seemed that all the 
necessary buffers were created/allocated/initialized.  And (at least 
with the address sanitizer) it was failing for me when using either FLTK 
or Qt, so the initialization would have to be wrong for both toolkits. 
I suppose that's possible, but it seemed to me to be one more pointer 
that we are doing something wrong and that the problem is not the way Qt 
initializes the OpenGL system.

I haven't been able to duplicate this problem directly so it is 
difficult if not impossible for me to debug.  So I either need someone 
to show me a way to reproduce the bug(s) or do the debugging themselves, 
at least to the point of finding out precisely where the crash is 
happening, if not why it is happening.

jwe