mem::size_of() is not a const function

[DemiMarie]

mem::size_of() is not a const function, even though it is always a compile-time constant. This is seemingly arbitrary and is annoying in practice.

[eefriedman]

I'm pretty sure fixing this would be a one-line patch: std::intrinsics::size_of is already treated as a compile-time constant.

[eefriedman]

Err, wait, nevermind, wasn't testing correctly. Anyway, seems reasonable (although it would probably be tricky to support in certain contexts, like array bounds).

[eddyb]

It's not arbitrary, just very hard to support in the current compilation, which is at odds with CTFE.
There are plans to make type-checking of constant expressions in any positions possible before everything else has been type-checked, which would allow using something like miri to evaluate a much, much, broader set of constants.

[mtanski]

This pretty annoying, you can't use it in side of trait const expressions now.

trait DataType {
   type DiskType;
   const DISK_SIZE: usize;
   // ...
}

struct UInt32;

impl DataType {
   const DiskType = u32;
   const DISK_SIZE: usize = std::mem::size_of<Self::DiskType>();
}

Clearly this doesn't work, but it seams like it should.

[burdges]

I think the active thread on this is rust-lang/rfcs#1144 but the associated constant one rust-lang/rfcs#1168 looked cool too.

[SimonSapin]

@eddyb tells me on IRC that implementation difficulties are now longer relevant now that type checking happens on demand.

[frewsxcv]

There's a PR open for this now: #42859