Add (limited) CTFE

[nrc]

See #253 for a limited form of CTFE. We can use this issue to track a more complete form too.

[archshift]

Can we revisit this discussion now that the 1.0 launch craze has passed?

[sfackler]

This seems like it's covered by the const fn work?

[ticki]

@sfackler Yeah, but const fn needs to be able to perform non-trivial functionality. For this to be done, we need const fn methods in traits and impls of traits.

[ranma42]

Moreover the functions mentioned in #253 (size_of, min_align_of, pref_align_of), just like several other intrinsic-based functions, are not yet const.

[JinShil]

I was quite surprised that the following code does not compile:

const CONDITION: bool = true;
const RESULT: u8 = if CONDITION { 1 } else { 0 };

This seems like it's covered by the const fn work?

I thought so too, but this doesn't compile:

#![feature(const_fn)]
pub struct BitField (u8, u8);
impl BitField
{
    const fn get_ms_bit_index(&self) -> u8 {
        if self.0 > self.1 { self.0 } else { self.1 }
    }
}

[nagisa]

Its in progress work. See the “postponed” label on the right hand side.

[petrochenkov]

@JinShil
The workaround is to use an array of two elements:

const CONDITION: usize = 1; // true as usize
const RESULT: u8 = [0, 1][CONDITION];

[petrochenkov]

The lack of if in constant expressions is a pretty serious deficiency, its C analogue ?: works even in preprocessor constant expressions. And C code does rely on this property of :?, this creates problems when porting.

[Amanieu]

If we allow if in constant expressions then it would make sense to allow match as well.

[kevincox]

Both if and match sound reasonable to me.

[frewsxcv]

Does anyone know what the status of conditionals within const fns are? Is it blocked on something in particular? An RFC? Implementation?

[BenWiederhake]

Found this RFC after creating rust-lang/rust#42906 . What can I do to help? What's the next step?

[eddyb]

@BenWiederhake See https://github.com/solson/miri.

[petrochenkov]

MIRI is merged into rustc and is will be used for all const evaluation soon (rust-lang/rust#46882), and size_of is already constexpr, so I don't think there's value in keeping this issue open.