Tracking issue for RFC 2412, "The optimize attribute"

[Centril]

This is a tracking issue for the RFC "The optimize attribute" (rust-lang/rfcs#2412).

Steps:

• Implement the RFC. Partially done, see Tracking issue for RFC 2412, "The optimize attribute" #54882 (comment).
• Adjust documentation (see instructions on forge)
• Stabilization PR (see instructions on forge)

Unresolved questions:

• Should we also implement optimize(always)? optimize(level=x)?
  • Left for future discussion, but should make sure such extension is possible.
• Should there be any way to specify what global optimization for speed level is used in
  conjunction with the optimization for speed option (e.g. -Copt-level=s3 could be equivalent to
  -Copt-level=3 and #[optimize(size)] on the crate item);
  • This may matter for users of #[optimize(speed)].
• Are the propagation and unused_attr approaches right?
• The RFC specifies we should emit an unused_attributes warning when misapplied, but similar attributes emit an error instead. Should this emit an error too?

[Centril]

@nagisa I've added all unresolved questions from the RFC for completeness; the first 2 may already be resolved, if you feel that's the case; tick the boxes.

[nagisa]

• Should we also implement optimize(always)? optimize(level=x)?
  • Left for future discussion, but should make sure such extension is possible.

Marking as resolved just because these never came up during the RFC discussion, especially since the attribute design does not prohibit adding these as a feature later on.

---

I’m assigning myself for now as I intend to implement this. Failing to dedicate time for this within two or so weeks I pledge to write up some instructions.

[nagisa]

I’ve implementation for optimize(size) now and am looking into ways to implement the harder side – optimize(speed). I might end up splitting the work into two PRs depending on how things go.

[Eh2406]

Would this allow a crate (proptest, quickcheck) that is performance sensitive, but often used in debug builds, to specify that it should always be built with release?

[nagisa]

The implementation for optimize(size) and optimize(speed) has now landed. Currently RFC is not fully implemented as the attribute propagation mechanic specified in the RFC is missing from the implementation. Nevertheless the actual functionality is present and can be toyed with.

Another thing that I’ve realised just now is that I never tested and is implemented incorrectly currently is "unused_attribute". The lint does not fire for the attribute applied to non-function items. This should also be fixed before any sort of stabilisation can proceed.

My recent observation is that optimize(none) is fairly often used by people to debug functions in an otherwise optimized build. It may be sensible to add it as well at some point.