Implement ptr_cast_array

[scottmcm]

ACP: rust-lang/libs-team#602
Tracking Issue: #144514

[rustbot]

r? @thomcc

rustbot has assigned @thomcc.
They will have a look at your PR within the next two weeks and either review your PR or reassign to another reviewer.

Use r? to explicitly pick a reviewer

[okaneco]

I noticed this in the libs-api thread before it was accepted but didn't want to be off-topic.
I was wondering how many uses of unsafe ptr casts are really just working around lack of const casting traits.

I think all of chunk methods could also be a fixme(const-hack) for TryFrom not being const before the const trait initiative. So the unsafe could also be removed here eventually.

I thought the other cast_array uses were more motivating outside of this module.

[scottmcm]

True, there's a bunch of different ways this could be done. Certainly there's also options like self.as_chunks().0.first() that would work for this function.

I don't think these chunk methods would be TryFrom, though, because the &[T; N] as TryFrom<&[T]> intentionally requires an exact length match. (That's aligned with the general guidance that "from" stuff gives the "same" thing, and having them be eq needs the lengths to be the same.)

But yeah, .get(..N)?.as_array() (https://doc.rust-lang.org/nightly/std/primitive.slice.html#method.as_array) would be an option if get worked in const. (That's still two checks in the MIR, though, so dunno if it's better that way or not.)

EDIT: Oh, wait, not, the as_chunks version doesn't work because that panics for N == 0. But yes, lots of options. Even split_at_checked(N)?.0.as_array(), I suppose, which doesn't need to wait for const-trait.

[okaneco]

I was thinking of TryInto (unless that has the same issues and I'm just mistaken/misremembering). But as you said, now we have a lot of other options.