[Public WebGL] Non-binding context creation attributes, specifically 'stencil'

Glenn Maynard [email protected]
Mon Apr 16 21:34:26 PDT 2012


On Mon, Apr 16, 2012 at 10:45 PM, Jeff Gilbert <[email protected]> wrote:

> This implementation-defined behavior causes the above demo to render as
> intended in Chrome, but fails to do so in Firefox. If we were to have a
> strict or adverse reference renderer, this demo would also fail to
> consistently render as intended there. Since the two implementations are
> (in as far as this demo is concerned) spec-compliant, I believe this is a
> spec issue.
>

I don't agree that this is inherently a bug in the spec.  It's by
design--as the cost of supporting hardware of varying capabilities--that
it's possible for an application to function on one spec-compliant
implementation and not on another.

(That's not to say we shouldn't do this, of course.)


> I move that we require any (implicit or explicit) negative requests for
> 'antialias', 'depth', and 'stencil' to guarantee they will not be supplied
> with that respective buffer type, similar to 'alpha'. Because 'stencil'
> defaults to false, this means that in the absence of a positive request for
> a stencil buffer, the context will not receive one.
>

I do agree with this.  These are hints is because some hardware may not
support all rendering features (or all combinations of features).  It seems
unlikely that a GPU would fail to support *not* using one of these
features, so there should be no reason to allow implementations to ignore a
false value in the context creation options.  (In other words, while some
hardware may not be able to satisfy a "true" request, they should always be
able to satisfy "false".)  Doing so pays the cost of decreased interop
without any improvement to hardware compatibility.

I'm not sure if this is necessarily possible with "antialias", however.
nVidia drivers allow users to force antialiasing on globally, so browsers
may be unable to disable it.  I'm not sure, though; this may only apply to
rendering to display buffers and not when rendering to offscreen buffers.
(If you're using WebGL to do math rather than render graphics, you may
really want to force antialiasing off...)

-- 
Glenn Maynard
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://khronos.org/pipermail/public_webgl_khronos.org/attachments/20120416/5a8468fb/attachment.html>


More information about the public_webgl mailing list