Tracking issue for num_bits_bytes stabilization

[aturon]

The BITS and BYTES constants for e.g. i32 are currently unstable, due to lack of clarity around:

• Should these be associated constants of the type instead?
• Is there a better way to get at these via e.g. size_of?

cc @eddyb

[eddyb]

BYTES would be directly redundant with any kind of size_of computation in constants, so I'd be against keeping it, if the latter is on the horizon.
BITS on the other hand, is indeed BYTES * 8 in all the cases we currently provide it in, but it might be more ergonomic.
Do we have (even rough) usage numbers?
Technically, anyone can implement their own trait with associated constants, so a straight-forward replacement would be available where necessary, even if std doesn't provide it and size_of still doesn't work in constant.

[alexcrichton]

We've gotten this far without these being stable, let's just axe them.

[SimonSapin]

Does this include usize::BYTES and isize::BYTES? mem::size_of doesn’t work in const contexts.

[alexcrichton]

🔔 This issue is now entering its final comment period for deprecation 🔔

although ping @rust-lang/compiler, I hear claims that there were discussions of these constants at the work week, and there may be something we're interested of!

---

@SimonSapin yeah this'd be those constants (they're unstable today anyway).

[SimonSapin]

Is there a replacement that works in const contexts (e.g. array sizes) and does not involve multiple definitions with #[cfg(target_pointer_width = "64")]?

[SimonSapin]

A single std::ptr::BYTES constant (or some other name/location) would work, although making size_of a const fn somehow would be better.

[alexcrichton]

There isn't a replacement that I currently know of, although if this is used widely throughout the nightly ecosystem we can leave them for now until there's a suitable replacement.

[andrewtj]

I've been using std::usize::BYTES in const contexts. Doing a #[cfg(target_pointer_width = …)] dance instead wouldn't be much of a hassle but a more succinct alternative would be nice.

[tari]

Could size_of be made const so it becomes usable in const contexts even after removal of BYTES? I agree that these constants are of limited utility, but a #[cfg] dance is not very nice to work with.

[eddyb]

@tari It can be, but it requires more work on constants (the actual computation only depends on the pointer size and target-specific alignments).

[alexcrichton]

The libs team discussed this in triage recently and the decision was to deprecate

[alexcrichton]

Closed by #30943

[SimonSapin]

This was removed without any replacement :(

[SimonSapin]

To clarify, since the first message of this thread mentions i32, I only care about one const for the size of usize/isize/pointers, since that’s the only one that changes per platform.