Add implementation of Decode for Box<str> and Box<[T]>

[mversic]

Closes #564

I can't think of a way to implement Decode for all unsized types

[bkchr]

Can you check the already existing Box Decode implementation. It should be in the same way as this implementation.

[mversic]

Do you think that decoded String will have some excess capacity or will it have capacity == len?

[mversic]

I've taken a bit different route here

[mversic]

should I maybe use WrapperTypeDecode like this:

impl WrapperTypeDecode for Box<str> {
	type Wrapped = String;

	fn decode_wrapped<I: Input>(input: &mut I) -> Result<Self, Error> {
		// Guaranteed to create a Vec with capacity == len
		let vec = Vec::from(Box::<[u8]>::decode(input)?);
		// Guaranteed not to reallocate the vec, only transmute to String
		let str = String::from_utf8(vec).map_err(|_| "Invalid utf8 sequence")?;

		// At this point we have String with capacity == len,
		// therefore String::into_boxed_str will not reallocate
		Ok(str.into_boxed_str())
	}
}

[mversic]

can I get a response here?

[mversic]

hi, this PR should be ready for merge. Can someone approve it now?

[nazar-pc]

Would be great to get an impl for Arc<[T]> as well (see #633)

[mversic]

Would be great to get an impl for Arc<[T]> as well (see #633)

I don't want to complicate this PR further as it's not being reviewed as it is.
I'd be happy to implement Rc<[T]>/Arc<[T]> in a separate PR.

[bkchr]

@koute could you give this another look please.

[bkchr]

Suggested change

	assert_eq!(str.capacity(), str.len());
	debug_assert_eq!(str.capacity(), str.len());

[bkchr]

Maybe we could also have some sort of check that the underlying pointer did not change and there wasn't any reallocation.

[bkchr]

Btw the function is now stable.

[koute]

Sorry for the delay.

[koute]

As @bkchr mentioned, use Box::assume_init since it's stable now (MSRV is 1.82.0, which is old enough to use)

[mversic]

using this would require you to bump MSRV

current MSRV (Minimum Supported Rust Version) is 1.79.0 but this item is stable since 1.82.0
you may want to conditionally increase the MSRV considered by Clippy using the clippy::msrv attribute
for further information visit https://rust-lang.github.io/rust-clippy/master/index.html#incompatible_msrv
#[warn(clippy::incompatible_msrv)] on by default [incompatible_msrv]

[koute]

I think it's fine to bump the MSRV here; 1.82 should be old enough at this point.

[bkchr]

Yeah.

[koute]

This is stable already, so use it.

[mversic]

see this

[koute]

This is gonna be mega slow for primitive types. We should use the same trick that we use in e.g. Decode for [T; N] or decode_vec_with_len where we read the data in bulk if T is a primitive type, otherwise this will be a huge performance trap.