f32, f64 missing from_le, from_be, to_le, to_be, swap_bytes implementation

[Timmmm]

For all the integer primitives there are functions for converting the endianness: from_le, to_le, from_be, to_be and swap_bytes (they probably should be a trait but they aren't for whatever reason).

The floating point types f32 and f64 don't have these methods defined.

[hanna-kruppe]

Can you explain why they should have those methods? Floating point types aren't a natural choice for bit twiddling. They don't even have more fundamental bit manipulation operations such as bitwise operators or shifts/rotates. The usual way to work with the bit representation of floats is to turn them into integers with the same number of bits, and manipulate that integer.

[est31]

There is also the issue of signaling NaN's. See more discussion on this in #39271.

[Timmmm]

I'm wrapping libsoundio and that allows audio devices to support little and big endian floating point samples. These functions would be useful for converting between those.

For example I have this in a macro, which works for all the integer types, but not floating point types:

	fn from_raw_le(ptr: *const u8) -> Self {
		Self::from_le( unsafe {
			*(ptr as *const _)
		} )
	}

See also wikipedia. Finally I feel like they should be there just for completeness.

[tbu-]

This particular function would be unsafe for floats btw, arbitrary bit patterns might be signaling NaNs for floats.

[Mark-Simulacrum]

I think we could have these, though it's not clear to me that we should. They'd certainly have to be unsafe, I think.

[dtolnay]

This seems reasonable to me. I would consider a PR to add these.

[frewsxcv]

For f32::swap_bytes, would it be sufficient to just transmute to u32 and call u32::swap_bytes?

[tbu-]

I think it would be better to not store floating points of nonnative endian in f32 or f64, creating a wrapper type for that sounds safer (as you won't forget to cast it).

[frewsxcv]

Just to clarify my last comment, i'm talking about the implementation of f32::swap_bytes, just wondering the easiest way to accomplish the byte swap

[steveklabnik]

Triage: no change

[tbu-]

@steveklabnik Since this issue was created, we have created alternative methods to convert floats from/to bytes in big/little endian: #60446.