DSA queries for VAOs are incomplete.

[NicolBolas]

GL 4.5/ARB_DSA has some oddities in how it accesses vertex array object data.

First, there's this, from the definition of glGetVertexArrayIndexed:

An INVALID_VALUE error is generated if index is greater than or equal to the value of MAX_VERTEX_ATTRIBS

OK, fine... except that this function also takes GL_VERTEX_BINDING_OFFSET, which is supposed to be indexed by binding index, not attribute index. So GL_MAX_VERTEX_ATTRIBS is the wrong limit to use.

The other problem is that this API does not provide DSA queries for the following (most of which are vertex binding values rather than vertex attribute values):

• The buffer object bound to the given binding. Neither GL_VERTEX_ATTRIB_ARRAY_BUFFER (which would fetch the buffer used by the attribute index's binding index) nor a direct GL_VERTEX_BINDING_BUFFER DSA query is provided.
• The binding stride. Remember: GL_VERTEX_ATTRIB_STRIDE is different from GL_VERTEX_BINDING_STRIDE. Indeed, no DSA functions exist which set the GL_VERTEX_ATTRIB_STRIDE; glVertexArrayVertexBuffer(s) is defined to only set the binding stride.
• The binding divisor.
• The binding index for an attribute index.

Basically glGetVertexArrayIndexed cannot completely replace the need for binding the VAO and using glGetVertexAttribIiv/glGetIntegeri_v.

An interesting note is that ARB_direct_state_access explicitly says that glGetVertexArrayIndexed should be added to the "Get Command" in table 23.4 for GL_VERTEX_BINDING_STRIDE. So clearly someone thought that this function should be able to access the binding stride at least.

[pdaniell-nv]

Yikes, it does seems all this is not specified. FWIW the NVIDIA driver implements all the missing queries above. There must have been a disconnect between the GL_ARB_vertex_attrib_binding spec, EXT_dsa spec, ARB_dsa spec and the core spec.

Someone will need to figure out all the missing bits for OpenGL 4.6 to complete this specification.

[pdaniell-nv]

Addressing the issue internally here: https://gitlab.khronos.org/opengl/API/issues/108

[oddhack]

Fixed internally - this will show up in the next public spec update.

[oddhack]

Fixed in the 2019-10-22 spec updates (hopefully - this was a fairly complex issue, please check).