Allow let bindings and destructuring in constants and const fn

[oli-obk]

r? @eddyb

cc #48821

[eddyb]

Can you move this into local_qualif?

[eddyb]

This seems bad, because statement_like is used for a lot of errors - it should maybe be split into several functions?

[eddyb]

I think you want to replace the qualif always (e.g. let mut x = None; x = Some(TypeThatNeedsDrop);).
Also, if you have this special case, why do you need to do anything about statement_like?

[eddyb]

This is bad because it breaks promotion in const fn. Also, you want mutable borrows to locals outside of const fn too, right? I think we should orthogonally allow &mut foo in regular static, so only const needs a special-case at the outermost level (and promotion also needs to keep disallowing &mut NotZST).

[eddyb]

@oli-obk If you want to get this merged quickly, I'd suggest dropping the borrowing changes.

cc @nikomatsakis A bit of a conundrum: static X: &'static mut Foo = &mut Foo {...}; would be fine to allow because static providing immutable access means it's effectively only &Foo.
However, const Y: &'static mut Foo = &mut Foo {...}; can't be allowed (unless Foo is ZST) because the const semantics would result in copies of a &mut Foo (only valid for ZSTs).

But we do want to allow const Z: Bar = Bar::new(&mut Foo {...}); with a temporary lifetime for the &mut Foo reference, so the "top-level escaping mutable reference" is relevant here.
So I'm wondering if, for const specifically, we want to adjust the temporary lifetime rules, such that &mut expr expressions aren't special-cased (same as if they were written id(&mut expr)).
That means const ARRAY: &'static mut [T; 0] = &mut []; would still work, but only through promotion (which has the same rules, IIRC, as what we allow today at the-top level, except around drops, which shouldn't be an issue for [T; 0] types, the only ones we ever allowed).

[oli-obk]

I removed the borrowing changes

[shepmaster]

Ping from triage, @eddyb !

[oli-obk]

This is very low priority atm. We could close it until after the next beta cut.

[emilyalbini]

Ping from triage @eddyb! This PR needs your review.

[emilyalbini]

Ping from triage! Can @eddyb or someone else from @rust-lang/compiler review this?

[eddyb]

Isn't this condition flipped? EDIT: this appears to be the same for all allows_unstable.

[eddyb]

I would instead put an if around all the code below, just to make sure it's explicit.

[oli-obk]

Note to self. more tests:

// ok
const X: FakeNeedsDrop = { let x = FakeNeedsDrop; x };
// error
const Y: FakeNeedsDrop = { let mut x = FakeNeedsDrop; x = FakeNeedsDrop; x };
// error
const Z: () = { let mut x = None; x = Some(FakeNeedsDrop); }

[shepmaster]

Ping from triage @oli-obk ! Any time to work on this in the near future?

[emilyalbini]

Ping from triage @eddyb! This PR needs your review.

[oli-obk]

#50603 takes priority over this PR, because it fixes a soundness hole (there will be conflicts, so we're waiting for #50603 to get merged first)

[kennytm]

#50603 has been merged.

[oli-obk]

amazingly there are no conflicts whatsoever. @eddyb this is ready for another review

[eddyb]

Can you combine these two match arms?

[eddyb]

r=me with the nit fixed

[oli-obk]

@bors r=eddyb

[bors]

📌 Commit 2483c81 has been approved by eddyb

[bors]

⌛ Testing commit 2483c81 with merge 9f80ea3...

[bors]

☀️ Test successful - status-appveyor, status-travis
Approved by: eddyb
Pushing 9f80ea3 to master...

[emilyalbini]

This PR needs to be backported to beta in order to backport #50909.
Cause: #51037 (comment)

cc @rust-lang/compiler

[oli-obk]

i can create a version of #50909 that doesn't depend on this, it's just a debug statement that conflicts

[emilyalbini]

@oli-obk that would be awesome!