search

gpuweb / gpuweb

Where the GPU for the Web work happens!
https://webgpu.io
Other
4.86k stars 319 forks source link

[wgsl] Consider allowing `_` ident for struct members #1310

Open kainino0x opened 3 years ago

kainino0x commented 3 years ago

It would have the special meaning of "unused" and it wouldn't be valid to use any ._ member from any code. Additionally, it would be valid to have multiple _ that would normally collide.

We could do this in other places as well (function args in particular), but struct members are most important if you have to add struct members in order to insert padding.

Prior art comes from numerous functional-adjacent languages like Rust (which does not actually allow it in structs) and Haskell/OCaml (in matches and functions iirc, probably not in records).

If we don't want to do this now, the safe thing is to disallow identifiers of _ so that we can use the syntax later. Rust does this.

kvark commented 3 years ago

I definitely agree that we should not allow _ for the struct field names. The semantic of "I don't need a name" could be useful in some cases, but we expect the most frequent case to be the same struct definition shared between different entry points (in which case _ isn't needed). We could easily move this to post-MVP.

dneto0 commented 3 years ago

I agree this is a good idea.

dj2 commented 3 years ago

Note, an identifier of _ is already disallowed by the spec. The IDENTIFIER is defined as [a-zA-Z][0-9a-zA-Z_]* which requires a leading character.

litherum commented 3 years ago

Can you write a short program where_ in a struct would be valuable in addition to offset/span/align?

(Because we aren't a functional language, and don't have lambdas or function pointers, I don't see much value in doing this for function parameters today. Given our current type system, no WGSL programmer is going to want to curry, or write a function that expects a callback of a particular signature.)

kvark commented 3 years ago

Can we move this to post-MVP? Seems marginally useful and non-essential. We have bigger fish to fry atm.

kainino0x commented 3 years ago

Can you write a short program where_ in a struct would be valuable in addition to offset/span/align?

It's sort of an alternative to offset/span/align. Potentially.

Can we move this to post-MVP? Seems marginally useful and non-essential. We have bigger fish to fry atm.

Like I said in the OP, all we really have to do is decide whether to reserve the syntax. Dan pointed out it's already an invalid identifier, so the syntax is already reserved. So yes.