rustc: treat const bodies like fn bodies in middle::region.

[eddyb]

Allows T::ASSOC_CONST to be used without a T: 'static bound.

cc @rust-lang/compiler @rust-lang/lang

[rust-highfive]

r? @brson

(rust_highfive has picked a reviewer for you, use r? to override)

[eddyb]

r? @nikomatsakis

[arielb1]

Can't this drop WF constraints on types in associated constants? A: no, it's a regionck method. But because we want to allow free fn calls etc. in associated constants, isn't it better to teach regionck that the "outer scope" of associated constants is not in fact 'static but rather some kind of anonymous free region?

[eddyb]

I don't really understand what the point of checking substitutions is, can someone explain that?

[arielb1]

The idea is that computing <&'a T as Trait>::CONST requires that T: 'a at the use-site. This shouldn't be needed for soundness (because of "region truncation") but sounds like a good property.

We still want regionck to be smarter, tho (because of calls and the like).

[arielb1]

You need to fix "droplessness" - have temporary_scope of constants be None again.

[00:16:48] error: borrowed value does not live long enough
[00:16:48]    --> /checkout/src/libcore/char_private.rs:370:33
[00:16:48]     |
[00:16:48] 370 |   const NORMAL1: &'static [u8] = &[
[00:16:48]     |  _________________________________^
[00:16:48] 371 | |     0x5e, 0x22,
[00:16:48] 372 | |     0x7b, 0x05,
[00:16:48] 373 | |     0x03, 0x04,
[00:16:48] ...   |
[00:16:48] 521 | |     0x01, 0x86, 0x3f,
[00:16:48] 522 | | ];
[00:16:48]     | | -
[00:16:48]     | |_|
[00:16:48]     |   does not live long enough
[00:16:48]     |   temporary value only lives until here
[00:16:48]     |
[00:16:48]     = note: borrowed value must be valid for the static lifetime...
[00:16:48] 
[00:16:48] error: aborting due to 70 previous errors
[00:16:48] 
[00:16:48] error: Could not compile `core`.

[eddyb]

@arielb1 Yeah, I was telling @nikomatsakis that I have no clue how I would do this.

[nikomatsakis]

So I've got a rather meandering PR (that is now working, though) that refactors the region-maps stuff somewhat. (It converts them into queries keyed by the enclosing item-like, in particular.) It doesn't directly bear on this, I guess, though it certainly...interacts. It doesn't seem too hard to have the idea of a "static temporary scope", though.

Right now we have this "terminating scopes" set, I guess one way would be to make that into a map where the value is a Region, so that when we are searching for the innermost temporary scope, if we encounter a terminating scope, we use the value (which, for statics would be 'static, but otherwise would be a scope. Alternatively, we could skip the step of inserting into the terminating scopes set, and use the fact that we never find a terminating scope as a signal; this makes me mildly uncomfortable, but it could work.

One of the things I've been wondering about here:

For constant expressions like [foo; 3], I guess that the scopes involved with 3 ought to be distinct from the enclosing function, right?

[eddyb]

@nikomatsakis Yeah I was wondering if I should have a visit_ty method that isolates nested expressions like that. I'll try removing the terminating scope for now, I guess?

[nikomatsakis]

@eddyb i.e., specifically in this case? You'll have to make sure that there is no parent scope, as well, since otherwise we'll walk right into the enclosing function as we traverse the scope tree. But given that these are forced to be a 'constant context' that seems correct anyhow.

[arielb1]

@eddyb - are you still working on this?

[eddyb]

Yeah, I'm waiting for #41662 to land first.

[bors]

☔ The latest upstream changes (presumably #41702) made this pull request unmergeable. Please resolve the merge conflicts.

[eddyb]

Okay, turns out I was close, but @nikomatsakis had to remind me that I had to update ParameterEnvironment::for_item so it uses the correct outermost scope for constants.

[nikomatsakis]

@bors r+

[bors]

📌 Commit 90af729 has been approved by nikomatsakis

[bors]

⌛ Testing commit 90af729 with merge f1140a3...

[bors]

☀️ Test successful - status-appveyor, status-travis
Approved by: nikomatsakis
Pushing f1140a3 to master...