Making language bindings portable over different versions of webgpu.h

[kainino0x]

webgpu.h is getting standardized, but will gain features over time. Meanwhile, language bindings like Dawn's C++ bindings will want to be portable over webgpu.h. But this means that if, say, Dawn's C and C++ headers gain feature X, but wgpu-native's C and the proposed Rust-on-C bindings gain feature Y, then those bindings layers won't be portable to the other implementation's C header.

• Maybe webgpu.h should expose macros saying which new features it has (or what version it is), so bindings in other languages can expose the same bits?
• Maybe webgpu.h should become fixed and new things should be in new headers?
• Maybe webgpu.h should be ABI-stable so the bindings can bundle their own webgpu.h while still working on other implementations? (not sure this is possible - I think it might prevent mixed use of the C and C++ APIs for example)

[kainino0x]

Maybe this can be done on an ad-hoc basis, where, say, Dawn generates two versions of its C++ bindings - one for Dawn, and one for some unnamed snapshot of a subset of webgpu.h that we know is supported where we need it. So those bindings would just take an implicit dependency on whatever was in that snapshot.

[kainino0x]

Dec 21 meeting

• (discussion)
• Probably fine to not solve this before we do ABI compatibility