RFE: do not use `int32`

[DemiMarie]

int32 is boxed in OCaml.

[talex5]

Is this showing up on a profile somewhere? I'd expect allocation due to this to be insignificant, but I haven't measured it.

[DemiMarie]

Nope! I just noticed it while reviewing the code.

[DemiMarie]

Is this showing up on a profile somewhere? I'd expect allocation due to this to be insignificant, but I haven't measured it.

Is it? There are lots of int32s in Wayland protocol requests and events.

[talex5]

I often find when I profile something that the problems turn out to be somewhere I didn't expect, so it would be good to confirm there isn't something more urgent to fix first. I recently wrote a blog post about profiling OCaml programs (https://roscidus.com/blog/blog/2024/07/22/performance/) which lists some tools that I've found useful, if anyone wants to have a go at it.

[DemiMarie]

I’m concerned that this might show up as uniformly slow code, due to cache misses and other problems. The proper solution is for OCaml to support unboxed int32s, like GHC Haskell and MLton do.

[DemiMarie]

This could be implemented using the Immediate64 attribute.