Const constructors from str, from slice for ArrayString, ArrayVec

[bluss]

• Add from_str_const constructor for ArrayString

• Add from_slice_const constructor for ArrayVec

We can't make the ArrayVec constructors generic, not even where T: Copy,
so they are just templated out with macros for the integer types, quite limited.

These use "const error" for error handling - using panic would need Rust 1.57 (not yet released).

Closes #204

[aldanor]

Neat!

About compile-time const, for pre-1.57, you can use a hack like this btw:

pub(crate) struct Assert<const L: usize, const R: usize>;

impl<const L: usize, const R: usize> Assert<L, R> {
    pub const LENGTH_LT_CAPACITY: usize = R - L - 1;
}

macro_rules! assert_length_lt_capacity_const {
    ($len:expr, $cap:expr) => {
        #[allow(path_statements)]
        { Assert::<$len, $cap>::LENGTH_LT_CAPACITY; }
    }
}

fn main() {
    assert_length_lt_capacity_const!(2, 3);
    // assert_length_lt_capacity_const!(3, 3); // compile-time error
}

[bluss]

I see. That's similar to what I already added for it to work on stable, but maybe your version has better error messages

[aldanor]

It's not perfect, but quite readable. Perhaps could be improved by something like

impl<const LEN: usize, const CAP: usize> Assert<LEN, CAP> {
    pub const OK: usize = CAP - LEN - 1; /* insufficient array capacity */
}

This yields:

error: any use of this value will cause an error
 --> src/main.rs:4:27
  |
4 |     pub const OK: usize = CAP - LEN - 1; /* insufficient array capacity */
  |     ----------------------^^^^^^^^^^^^^-
  |                           |
  |                           attempt to compute `0_usize - 1_usize`, which would overflow

[bluss]

It seems like the length - coming from s.len() can't be used there in current Rust, unfortunately.

[aldanor]

It seems like the length - coming from s.len() can't be used there in current Rust, unfortunately.

Ah, yea, indeed, it would only work with "true consts"...

So what's the plan then, wait till MSRV >= 1.57 bump for the crate eventually? (assuming passing s.len() to const assert works there)

[bluss]

It won't work there either (I'm developing on nightly locally, so that's the first behaviour I see). I'm not sure what the missing feature would be called.

But I'll elaborate - on Rust 1.57 we can panic in constants, so that's better than the current solutions. Not a reason to bump on its own, though (?)

[aldanor]

Sorry for bringing the old issue back from the dead :) Wonder if this would be feasible, given we can panic in constants?

Const constructors are still a big missing piece...

[utkarshgupta137]

Bump

[jon-zu]

Is there a reason to use slices instead of an Array [T; M] with an assertion that M <= CAP (assert works fine now in const fns). I've tried this successfully, the only downside is Copy is still required, and we have to pass the Array as reference else the compiler complains that It can't determine the destructor.

I would argue this const constructor for ArrayString and one for Arrays covers almost all cases.

[GummyGun]

Is there any reason this is not currently in yet.
Is the crate dead/stalled?

[ghost]

Is this still Work in Progress? Or are we just waiting for a feature to stabilize?

[Ad1n]

Rebased this onto current master as #327 (#327) ArrayString half only . Your commit authorship preserved : it cherry-picked cleanly with no conflicts.

The main thing I added is evidence on the MSRV question: I built with rustc 1.51.0 in both std and no_std and confirmed from_str_const needs no bump - a valid const compiles an over-long one is a compile error at 1.51. So the 1.57 concern from #294/#296 doesn't apply to this half. Also fixed len as u32 - > len as LenUint which would otherwise break 16-bit targets since #234. All CI green, including miri.

Happy to close #327 if youd rather land yours directly - i mainly wanted to show it still applies.