https://gpuweb.github.io/gpuweb/#abstract-opdef-validating-inter-stage-interfaces

I noticed a few inconsistencies with the spec when reading through the interface validation code in wgpu-core.

• User-defined variables all count as 4 components towards the limit (not what ty.dim.num_components() returns).

  Each user-defined output of descriptor.vertex consumes 4 scalar components.

  For each user-defined input of descriptor.fragment there must be a user-defined output of descriptor.vertex that location, type, and interpolation of the input.

• If the topology is "point-list" the component limit for vertex outputs needs to be reduced by 1.

  If descriptor.primitive.topology is "point-list":

  Decrement maxVertexShaderOutputComponents by 1.

• For fragment inputs, the front_facing, sample_index, and sample_mask builtins all count as 1 component towards the limit.

• There is now a maxInterStageShaderVariables (also mentioned in Further spec compliance #2170):

  The location of each user-defined output of descriptor.vertex must be < device.limits.maxInterStageShaderVariables.

I also had a couple questions on things aren't necessarily issues but weren't obvious to me when reading the code:

• At the point where validation is done:

  wgpu/wgpu-core/src/validation.rs

  Line 1255 in 53dd49f

  Varying::Local { ref iv, .. } => iv.ty.dim.num_components(),

  are all Varying::Local variables "user-defined"? (in terms of how the spec uses "user-defined")
• (Now tracked by Validation of inter-stage variable types does not match the spec #9143): Matching types between stages is checked via frag_input_ty.is_subtype_of(vert_output_ty) (here). E.g. for a particular location this would allow vec4<f32> on the vertex side and f32 on the fragment side. Whether this is allowed isn't clear to me when looking at the line in the spec:

  For each user-defined input of descriptor.fragment there must be a user-defined output of descriptor.vertex that location, type, and interpolation of the input.