> Are you saying that the direct mapping requirements I quoted above have been abandoned?

Yes.

See these quotes:

Apple: "Having an isolated focus on SPIR-V is not a good perspective, we have MSL etc. to consider too"

...

Apple: "Extreme A is no interaction with SPIRV?, extreme B is very tightly coupled to SPIRV, we should find a middle point"

...

Google: "We take all targets into consideration. We can allow developers do runtime probing and maybe branch according to that. Optimization occurs in existence of uncertainty. Dealing with fuzziness is a fact of life and we should give the tools to developers to determine in runtime."

From https://docs.google.com/document/d/15Pi1fYzr5F-aP92mosOLtRL4...

See also:

https://github.com/gpuweb/gpuweb/pull/599

https://github.com/gpuweb/gpuweb/issues/582

https://github.com/gpuweb/gpuweb/issues/566

http://kvark.github.io/spirv/2021/05/01/spirv-horrors.html

https://github.com/gpuweb/gpuweb/issues/847#issuecomment-642...

https://news.ycombinator.com/item?id=24858172

The process has been very implementation driven, with Naga, the Rust tooling that does the whole conversion shenanigans, growing and changing its IR and semantics steadily. It's ironic that a group that is usually very adverse to having a single implementation of things is now pointing to Naga and arguing that we don't need an easy conversion step because we have all the compiler infrastructure build already.

The group has slowly redefined "bijective" to mean "easily compilable", which would be pretty hilarious, if it weren't so sad.

I agree Apple is the bad guy here and even the public discussions were a clown show, who knows how much worse it was behind closed doors. But I’d rather have an inelegant standard than none at all, and I think the problems with WGSL can be worked around.