Should this library use zerocopy rather than declarative parsing.

[Frostie314159]

@george-hopkins @kevinbader @ProfFan
I hope it's okay, that I tagged you here, but you're the only people I know, that use the library. I've been working on a Wi-Fi stack for the ESP32 written in Rust, based on the research conducted by the esp32-open-mac project and I noticed, that the declarative approach can generate a lot more code, which in some situations may be unnecessary and in general generates quite a lot of overhead. After taking a look at smoltcp's parsing code, I began thinking about using zerocopy imperative reading and writing. It would help me greatly to hear your opinion on this. I'll create a branch, where I experimentally implement zerocopy for some frame types.

Some elements, like the RSNE could benefit greatly from this, since the code gen for it is currently a mess.

[ProfFan]

I think smoltcp styled zero copy reading/writing is a better approach at least in embedded programs. If done nicely the API can actually be easy to use as well. The current API requires a copy in pwrite/pread every time...

[Frostie314159]

That was my thought process too, and I think, that we can model certain frames (e.g. Authentication) a lot better this way.

[ProfFan]

I found out another benefit of the smoltcp approach: the xxxRepr structs are const-friendly because they don't take any buffers, which probably allows more compiler optimization.