illegal moves permitted in statics and constants

[nikomatsakis]

This code compiles on nightly today but it should not, because Foo is not Copy:

#![feature(const_fn)]

struct Foo(usize);

const fn get(x: Foo) -> usize {
    x.0
}

const X: Foo = Foo(22);
static Y: usize = get(*&X); // also a problem for `const Y`

fn main() {
}

Note that the get(*&X) in a function body properly results in an error.

[nikomatsakis]

triage: P-medium

[nikomatsakis]

This is likely a problem of how borrowck kind of special-cases the handling of statics a bit. Should be relatively easy to fix. Could be fixed "en passante" by MIR borrowck; I'll also try to add some mentoring tips.

[cengiz-io]

Hello @nikomatsakis. I'd like to take a shot at this if I'm allowed.

[cengiz-io]

My first observation is that this isn't related to const_fn. Am I right?

I'm reading the RFC now and doesn't seem to require Copy on const_fns.

EDIT: I somehow missed the static Y: usize = get(*&X); part. Now it's starting to make sense 👍

[cengiz-io]

Hello again @nikomatsakis

I'm working on this right now.

[nikomatsakis]

@cengizio ok -- how are you making out? Sorry I didn't get around to leaving any notes on how to proceed. Let me know if you still want me to go digging around -- or privmsg me on IRC (nmatsakis).

[nikomatsakis]

@cengizio so looking at the code a bit, it seems clear that the problem is that the borrowck code has this kind of custom and woefully insufficient bit of code dedicated to statics and constants. It really ought to be using the same pathways as the rest of the code. (In particular, the ExprUseVisitor, which is a bit of code that interprets the HIR and gives callbacks corresponding to moves, borrows, etc). Unfortunately, those pathways are kind of setup more for regular expression bodies; I'm not 100% sure how easy they will be to apply.

I feel like this bug fix is "just" a refactoring, but the refactoring itself may not be all that simple. Doing the borrowck on the MIR we generate for constants would fix it, but it's not clear when MIR borrowck will land.

[cengiz-io]

@nikomatsakis as we talked in IRC, I think this should be discussed widely before we start implementing anything.