[Public WebGL] OT: exception handling

Mark Callow [email protected]
Sun Apr 8 19:05:38 PDT 2012

On 07/04/2012 01:13, Florian Bösch wrote:
> I think the whole idea of context lost is perverted and broken. If a
> system process runs out of memory. It gets sigtermed, if it oversteps
> its memory, it gets sighalted, no reasonably programming scheme would
> do things like silently free memory you've been using. It's why
> virtual memory managers, swap, protected memory etc. have been
> invented oh roundabouts 1970 and been thoroughly implemented on all
> personal computing hardware to around 1985, to being fully supported
> by any OS by around 1995. GPUs context lost is a barbaric regression
> to the glory days of 1950 or thereabouts in this regard.
> On Fri, Apr 6, 2012 at 6:04 PM, Benoit Jacob <[email protected]
> <mailto:[email protected]>> wrote:
>     On mobile devices at least, we have no choice. First, EGL contexts
>     get lost when (that's why EGL_CONTEXT_LOST exists) so the problem
>     exists anyway and WebGL context lost/restored events are just
>     giving content a chance to recover.
> I agree that you have no choice on current mobile hardware, and it's
> not the failing of WebGL to try to handle it. It is still a
> stupendously bad idea from a hardware/programming point of view, but
> unfortunately one we can't do anything about until in oh, 20 or 50
> years or so when GPUs have arrived in the age of real programmable
> machines.

Back in 1988 we virtualized IRIS GL contexts on the Silicon Graphics
IRIS 4D GT. The hardware supported 16 contexts; if more were needed they
would be swapped in and out.

Unfortunately there is no requirement in today's OpenGL for drivers to
do this. Since there is no way to get a complete dump of the GL state
the only way for an application/OS to do it for itself is to shadow the
GL state.

BTW EGL_CONTEXT_LOST stems from the days of devices such as early
Qualcomm hardware where OpenGL ES ran partially on the DSP and when that
was needed to process an incoming call - EGL_CONTEXT_LOST. I agree it is
a very inelegant design.



-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://khronos.org/pipermail/public_webgl_khronos.org/attachments/20120409/25cc4a46/attachment.html>

More information about the public_webgl mailing list