Regression from stable: "Unexpected tail in unsized_info_ty"

[withoutboats]

There's been a regression in a recent nightly regarding dispatch on unsized types:

trait Foo {
    fn foo() { }
}

struct Bar;

trait Baz {
}

impl Foo for (Bar, Baz) { }


fn main() {
    <(Bar, Baz) as Foo>::foo()
}

https://is.gd/NjyQaO

[withoutboats]

It looks like this function (type_of::unsized_info_ty) just doesn't handle unsized tuples. Can't tell how it started getting called but it did.

[est31]

Using the bisect-rust tool I've bisected the regression to commit 9fa25a7, PR #42015

cc @nikomatsakis

[nikomatsakis]

Huh, interesting. I sort of expect this program to be an error, because (IIRC) in tuples we do not permit unsized types, but maybe that's not true or I'm misremembering.

[withoutboats]

We permit the final member of a tuple to be unsized: rust-lang/rfcs/pull/1592

[nikomatsakis]

ok, that makes sense

[nikomatsakis]

triage: P-high

I'll investigate and try to stay on top of this. Oddly, I'm not reproducing on my local build, though it includes a few further changes.

[Mark-Simulacrum]

I am unable to reproduce as well in the same nightly as playpen uses on both macOS and Ubuntu, so there is something interesting going on here for sure I think.

[est31]

Confirmed. The ICE is reproducible in rustc 1.19.0-nightly (28fd1e519 2017-05-27) but not reproducible any more in rustc 1.19.0-nightly (d47cf08d5 2017-05-28).

Should we add a test and close this, or should we investigate?

[arielb1]

Niko fixed this in #42251 but put the wrong issue on the fixes line. Closing.

[Mark-Simulacrum]

Huh, I wasn't able to reproduce with the 05-27 nightly, but that may have been a fluke of my environment; not sure. I'm fine with closing of course, though.