str: Mark unstable round_char_boundary feature functions as const

[okaneco]

Mark floor_char_boundary, ceil_char_boundary const
Simplify the implementations, reducing the number of arithmetic operations

It seems unnecessary to do the lower/upper bounds calculations and extra slicing when we can jump straight to inspecting the bytes, assuming the underlying data is valid UTF-8.

Tracking issue #93743

[rustbot]

r? @Mark-Simulacrum

rustbot has assigned @Mark-Simulacrum.
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

[clarfonthey]

I wouldn't necessarily say this "simplifies" the implementation since a lot of this is relative to what the optimised output is, and the new implementation may not be as clear to the optimizer that at most four iterations happen. However, I think that it's a valid implementation and making it const is nice.

[clarfonthey]

Personally, I think that simply adding this as a condition to the loop itself might be helpful (and similarly above) to ensure that the optimiser knows this precondition as well. Not sure how it'd affect codegen, but it feels like the safest approach compared to, for example, assert_unchecked.

[okaneco]

Compiler explorer links
floor_char - https://rust.godbolt.org/z/PKr1GP1rs
ceil_char - https://rust.godbolt.org/z/nMqPT4rYY

Some surface notes:
The while loops don't have panic paths.
For the iterators, the floor version unrolls but the ceil remains a loop.

I filed an issue because of noticing something in the original implementation #139759.
The saturating_sub from the index still required a bounds check, but that's been fixed and should be in the bump to LLVM 21.

[Mark-Simulacrum]

@bors r+

Seems reasonable enough and ~equally clear to me.

[bors]

📌 Commit 7f7d343 has been approved by Mark-Simulacrum

It is now in the queue for this repository.