Add to_bytes and from_bytes to primitive integers

[SimonSapin]

Discussion issue turned tracking issue: #49792

[SimonSapin]

r? @sfackler

[rust-highfive]

r? @Mark-Simulacrum

(rust_highfive has picked a reviewer for you, use r? to override)

[oli-obk]

cc @BurntSushi

[BurntSushi]

This doesn't need to be a transmute right? Can you do something like this instead?

unsafe { *(&self as *const Self as *const u8 as *[u8; mem::size_of::<Self>()]) }

Maybe we don't care about decreasing the use of transmute?

Also, neat, I didn't realize mem::size_of could be used at the type level.

[SimonSapin]

It could be that, but that seems harder to read to me. Why would it be preferable? Regarding transmute I often prefer to fully qualify it with a turbofish, but in this case that felt unnecessary since it forms the entire body of a method that has an explicitly-written signature.

size_of became a const fn in #42859

[oli-obk]

I think transmute is fine here, because we're transmuting non-reference types. For reference/pointer types I agree that we should not be using transmute (this is what we suggest in clippy at least)

[BurntSushi]

Sounds good to me!

[SimonSapin]

@oli-obk I’m not opposing this principle, but could you say more about why?

[oli-obk]

I'm assuming the "why" refers to "transmute is fine for non-reference/pointer types":

let u = transmute::<T, U>(t); additionally checks whether the size of both sides of the conversion match. Going through raw pointer casts is essentially transmute_copy.

There's no increased source for errors when using transmute with all types specified (in this case specified in the function signature of the surrounding function). The raw pointer casts are very sigil heavy and need a few reads to figure out.

If you are transmuting references on the other hand, the size check doesn't gain you anything (because references to sized types are always one pointer large). In that case the raw pointer casts are also less sigil heavy because you don't need the all the dereference or reference expressions, but it's mostly a cast.

[BurntSushi]

I love it! This should help remove some unsafe from byteorder.

[sfackler]

Nit: These should probably mention from_be and from_le instead.

[SimonSapin]

I find it amusing that we have for example both to_be and from_be but their implementation is exactly the same. Anyway, fixed.

[sfackler]

r=me with the one doc thing fixed

[SimonSapin]

@bors: r=sfackler

[bors]

📌 Commit 4472991 has been approved by sfackler