Implement type ascription.

[Ryman]

Rebased #21836

cc @eddyb (didn't see any reply on irc), @nrc

[rust-highfive]

r? @nikomatsakis

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

[nrc]

@Ryman I'm not sure if you have already, but I think we need some changes so that we do not attempt to coerce in a type ascription expression when the expression is in reference (lvalue) context. This is required for soundness. We also need tests for that. (See RFC 987 for more details).

[nrc]

And thanks for doing the rebase!

[Ryman]

@nrc Only a quick rebase and cfail/rpass run, hadn't noticed the requirements change other than the loss of 'ascriptions in patterns'. Hope it's still useful for @eddyb!

[eddyb]

@Ryman thanks!
@nrc I don't even know how to tell whether something is an lvalue or not, this early during typeck.
There is a function in middle::ty which can return a few kinds of lvalue/rvalue kinds, used from trans, not sure if it's applicable here.

[nrc]

@eddyb: Using those fns in middle::ty would be my first bet. Otherwise, it is a strictly syntactic property - you just need to know the context of the expression. If none of that works out, then you could check later in type checking (mem categorization, I guess) that if you find an ascription expression, and it's an lvalue, then check to see if there is an adjustment on that expression, if there is, error out at that stage.

[eddyb]

The issue was related to lifetime invariance in mutable lvalues, wasn't it?

I do wonder what we do right now for (&mut &mut lvalue): T and (&mut *&mut &mut lvalue): T.
I expect that we either propagate the expected type and coerce the inner &mut in both case, or we don't in either.
If we do, then we either have a pre-existing soundness hole to fix or it works due to regionck and borrowck.
If we don't, then we can reuse the same decision for type ascription.

I haven't had the chance to experiment with this fully.

Oh and I didn't think of unsizing coercions on a pointer-type lvalue - but that would consume the lvalue, so it shouldn't have bad interactions with ascription.

[eddyb]

Some quick playpen experimentation suggests we do propagate expected types down, but we don't attempt coercions on lvalues:
http://is.gd/Hsa3ow
http://is.gd/Gu75Ca
http://is.gd/9fXuIt

[nikomatsakis]

@eddyb that sounds right. In particular, the & and &mut operators don't do coercions.

[bors]

☔ The latest upstream changes (presumably #23857) made this pull request unmergeable. Please resolve the merge conflicts.

[nikomatsakis]

After some discussion with @nrc conclusion was that I should take over this PR too.

[alexcrichton]

Closing in favor of an upcoming rebase by @nikomatsakis!

[cristicbz]

I recently remembered that type ascription was supposed to land "any minute now" some time ago, is this still going to happen?

PS: Also may I suggest not closing issues in favour of upcoming rebases until the latter are actually opened (just so as not to have things fall through cracks). Not trying to be passive aggresive, just a genuine suggestion!

[nrc]

ping @nikomatsakis who was going to do the rebase

[pnkfelix]

@alexcrichton @cristicbz

I can see why we would close a PR if e.g. its functionality is going to be provided in a different manner (so that people do not waste their time reviewing code that isn't even close to what is planned to be added).

But closing a PR based solely on an anticipated future rebase that has not been opened does seem misguided. After all, what if someone else were willing to do the rebase in the meantime?

(In other words, I think that closing this PR may have been an error.)

[alexcrichton]

I'm fine reopening, I'm just trying to keep the queue clear!

[apasel422]

Is someone working on the rebase of this?

[nikomatsakis]

I was supposed to, but I haven't had the time. I'd be happy to mentor an
attempted rebase, though.

On Wed, Oct 7, 2015 at 3:42 PM, Andrew Paseltiner notifications@github.com
wrote:

Is someone working on the rebase of this?

—
Reply to this email directly or view it on GitHub
#23773 (comment).

[strega-nil]

@nikomatsakis I'd like to attempt a rebase. Would you like to mentor
me? :P

[nikomatsakis]

I would be happy to, but there has been someone doing the rebase that I've
been talking to on IRC. I haven't heard from them lately though, I should
check in.

On Sun, Oct 25, 2015 at 7:44 AM, Nicholas notifications@github.com wrote:

@nikomatsakis https://github.com/nikomatsakis I'd like to attempt a
rebase. Would you like to mentor
me? :P

—
Reply to this email directly or view it on GitHub
#23773 (comment).

[apasel422]

I had been working on the rebase, but haven't been able to work on it recently, so @GBGamer is welcome to take it over.

[nikomatsakis]

Ah, OK. Did you have an intermediate branch that might be worth looing at?
@GBGamer, feel free to ping me on IRC to discuss.

On Mon, Oct 26, 2015 at 2:11 PM, Andrew Paseltiner <notifications@github.com

wrote:

I had been working on the rebase, but haven't been able to work on it
recently, so @GBGamer https://github.com/GBGamer is welcome to take it
over.

—
Reply to this email directly or view it on GitHub
#23773 (comment).