[Piglit] [PATCH 4/5] glx_ext_create_context_es2_profile: Verify that indirect-rendering is impossible

[Ian Romanick]

On 06/12/2012 02:56 PM, James Jones wrote:
> On 6/12/12 2:47 PM, Ian Romanick wrote:
>> On 06/12/2012 02:35 PM, James Jones wrote:
>>> On 6/12/12 2:25 PM, Ian Romanick wrote:
>>>> From: Ian Romanick<ian.d.romanick at intel.com>
>>>>
>>>> The spec doesn't forbid indirect rendering with OpenGL ES 2.0.
>>>> There's no protocol defined, so it seems impossible that this could
>>>> ever work.
>>>>
>>>> NVIDIA's closed-source driver fails this test.  An indirect-rendering
>>>> ES2 context is created.  I have not verified whether this context is
>>>> actually usable, or it is garbage.
>>>
>>> Yeah, I didn't test, but in theory it "just works" with the existing
>>> GLX protocol, so I saw no reason to disable it.  Since GLX owns the
>>> GL command protocol specification, not GL or GLES, and this extension
>>> is creating contexts using GLX, I see no reason this should be
>>> considered a required error.
>>
>> So... how do commands like glReleaseShaderCompiler, glShaderBinary, and
>> glGetShaderPrecisionFormat, which have no GLX protocol, work?
>
> They work just as good as the subset of GL commands that don't have
> protocol defined today which many implementations (including ours) allow
> contexts to be created for. I'm not saying it's correct that our

That seems dodgy. :)

> implementation currently allows creation of GLES contexts that only
> partially work (or maybe not at all for all I know), but I don't think
> it's necessarily correct to require ALL implementations to fail either.
> Presumably they could have their own in-house protocol in place.
>
> For example, we implement "Beta" protocol for many GL operations
> that  don't have ARB-defined protocol, because we have customers that want
> indirect rendering support now and the ARB approval process often drags
> on forever. It only works when using both our client and server, both
> with beta protocol support enabled, but it is a usable solution and
> doesn't violate the spec in any way that I know of.

Right.  I knew you guys did that, and I can't think of anything wrong 
with that.  That's why I didn't make a similar test for desktop GL. 
(see also http://lists.x.org/archives/xorg-devel/2012-June/031602.html).

I think I'll rework the test to use the context that was created.  If 
that works, the test will pass.  Does that seem like a fair compromise?