Dzmitry Malyshau

We discussed this a bit in the office hours. It's possible that some context is lost, given that this was previously discussed 5+ weeks ago, but the general consensus is:...

IIRC, @dneto0 also preferred this to be an error. So it's pretty much "needs specification" now.

Added to the pile of things "depends on constexpr semantics"

The problem with (1) is that WGSL, unlike MSL/HLSL/CUDA, doesn't have separate `int`/`uint`/etc types. Our types are `i32`/`u32`/etc, so blindly following this popular option would be inconsistent with WGSL today....

Mappable buffers can only be used for copies, and nothing else. At least with the current spec. Both mapping and copies require 4-byte alignment on offsets and sizes. This means,...

We discussed this in #naga:matrix.org land and didn't find the change to be appealing. Here is a piece I wrote previously, internally: --- We think it would be a mistake...

Action item from the [virtual F2F](https://docs.google.com/document/d/1eiDkPsGbrrYjoiHKBTSH_KspYmqJj3ZWbjXuoMAk_rk/edit#heading=h.oihcpxdw9n4p): reach out to ISVs like BabylonJS asking if they are ok with `minBufferBindingSize` being a strong requirement on BGLs.

@kainino0x the change to optional doesn't affect the semantics. The issue is still there: if the binding type depends on, say, texture component type, do we require it to be...

Yes, the AI was for `minBufferBindingSize` because if the answer is "we can make it a requirement", then we can align this with the other fields, and no further action...

Well, not all issues can be helped by an immediate discussion :) The point of this issue is to highlight and track the inconsistency in the specification, so that we...