[Mesa-dev] [PATCH 3/6] glx: Move vertex array protocol state into the indirect backend

[Adam Jackson]

On Tue, 2017-11-14 at 13:15 -0800, Ian Romanick wrote:
> On 11/14/2017 12:13 PM, Adam Jackson wrote:
> > diff --git a/src/glx/indirect_glx.c b/src/glx/indirect_glx.c
> > index cfae12f6c0..b552b5768a 100644
> > --- a/src/glx/indirect_glx.c
> > +++ b/src/glx/indirect_glx.c
> > @@ -148,9 +148,21 @@ indirect_bind_context(struct glx_context *gc, struct glx_context *old,
> >     sent = SendMakeCurrentRequest(dpy, gc->xid, tag, draw, read,
> >  				 &gc->currentContextTag);
> >  
> > -   if (!IndirectAPI)
> > -      IndirectAPI = __glXNewIndirectAPI();
> > -   _glapi_set_dispatch(IndirectAPI);
> > +   if (sent) {
> > +      if (!IndirectAPI)
> > +         IndirectAPI = __glXNewIndirectAPI();
> > +      _glapi_set_dispatch(IndirectAPI);
> > +
> > +      /* The indirect vertex array state must to be initialised after we
> > +       * have setup the context, as it needs to query server attributes.
> > +       */
> > +      __GLXattribute *state = gc->client_state_private;
> > +      if (state && state->array_state == NULL) {
> > +         glGetString(GL_EXTENSIONS);
> > +         glGetString(GL_VERSION);
> > +         __glXInitVertexArrayState(gc);
> > +      }
> > +   }
> 
> This is where this code used to be, but commit d57c85c1 moved it.  Does
> this not re-regress things?  I guess wrapping it in 'if (sent)' seems
> like it ought to be enough to prevent the original problem.

Eep, good catch. I think this would indeed re-regress things. We're
calling ::bind before __glXSetCurrentContext - in fact with the dummy
context bound - which means those glGetStrings would probably fizzle.

I'll dig into this a bit further, hopefully a real piglit run will
shake out this and anything else remaining.

- ajax