romaji fails to build on beta,

[Mark-Simulacrum]

romaji, cc @uzimith

With a few minutes of time invested I wasn't able to narrow this down to a smaller example that still succeeded on stable.

! 
Nov 02 07:32:11.418 INFO kablam! error[E0282]: type annotations needed
Nov 02 07:32:11.418 INFO kablam!   --> src/romaji.rs:80:24
Nov 02 07:32:11.418 INFO kablam!    |
Nov 02 07:32:11.418 INFO kablam! 80 |                 x if x.is_youon() && buffer.is_katakana() => {
Nov 02 07:32:11.419 INFO kablam!    |                        ^^^^^^^^ cannot infer type for `T`
Nov 02 07:32:11.419 INFO kablam!    |
Nov 02 07:32:11.419 INFO kablam!    = note: type must be known at this point

cc @rust-lang/libs

[dtolnay]

Minimized:

struct S;

impl S {
    fn method(&self) -> bool {
        unimplemented!()
    }
}

fn get<T>() -> T {
    unimplemented!()
}

fn main() {
    match get() {
        x if x.method() => {}
        &S => {}
    }
}

[dtolnay]

@rust-lang/compiler because this doesn't look to be caused by a library change.

[pnkfelix]

Hmm I wonder if this is some sort of bad interaction between 1. the new way to do pat variables in guards (which is activated by NLL) and 2. Method lookup and/or autoref...

... but this test isn’t asking for the 2018 edition, right? (I think we are still building the old style MIR if you don’t request NLL. But maybe that changed ...)

[eddyb]

This can be caused by the typeck order. To make this work, all arms' patterns must be checked before any guards are.

[pnkfelix]

This was injected between rustc 1.31.0-nightly (46880f4 2018-10-15) and rustc 1.31.0-nightly (bef62cc 2018-10-16)

Log of bors commits in that time in the details block

% git log 46880f4..bef62cc --author bors --format=oneline
bef62cc Auto merge of #54486 - orium:obligation-forest-graphviz, r=nikomatsakis
8a7048b Auto merge of #54251 - varkor:silence-bad_style, r=Manishearth
99ab2f4 Auto merge of #54937 - alexcrichton:update-stdsimd, r=nikomatsakis
5ea8eb5 Auto merge of #55067 - ljedrz:generic_iterator_related_improvements, r=petrochenkov
df0d6ad Auto merge of #55023 - euclio:llvm-error-handler, r=cuviper

[pnkfelix]

Hmm this line in the description of #55067 set off some alarm bells in my head:

typeck: don't collect into a vector when unnecessary