Add a trait for checked indexing

[tbu-]

No description provided.

[tbu-]

I would like to bikeshed the name a bit, checked_index and checked_index_mut are quite long, I'd find shorter names preferable.

[bluss]

The usual name is get and get_mut, like slices do it. This was mentioned on irc today by @Kimundi (not w.r.t arrayvec).

[tbu-]

These names are already taken though. Maybe get_ and get_mut_?

[bluss]

Maybe checked index is good then. If we extend this to &str by the way, the has_index check needs to include character boundary checks (is_acceptable_index in odds).

I don't think has_index method should have any panicking case?

[tbu-]

has_index currently has the panicking case that start > end, but I consider that acceptable as that indicates a programmer error.

I have a small error there, it will currently panic if you use start.. but it is out of range. In that case it panics.

It would need to include boundary checks for &str, yeah. Can we do that in a follow-up PR?

[tbu-]

Fixed the mistake.

Thanks for your fast response times by the way, it's wonderful!

[bluss]

Good catch.

I know I said I would easily accept this functionality, just imagined it as a simple function or so. I don't know.

I think there should be no possiblity of panic -- I use checked versions in places where you want to avoid panics, a possible panic in the optimizer's analysis may throw it off and miss some other optimizations.

[llogiq]

Why not slice and slice_mut?

[bluss]

It's going to cover both indexing and slicing, just like the Index trait.

[bluss]

Maybe we can use just something like this?

I'd like to not reuse the Index impls since we want a no-panic codepath anyway.

trait CheckedIndex<Idx> {
    type Output;
    fn checked_index(&self, i: Idx) -> Option<&Self::Output>;
}

[bluss]

Optionally: Even add len() into the trait. That would put us close to a "Random Access (Iterator)" kind of interface.

[tbu-]

The optimizer has no problem with seeing the double bounds check, it will eliminate the panic anyway – I don't think not re-using Index is a good idea, as it just produces code duplication.

[bluss]

That's a good point, it's true most of the time.

[bluss]

I'm inclined to close this, since I don't agree with the approach. The crate space is free — it doesn't have to be in odds in particular.

[tbu-]

That's fine with me!