Field access of cast expression: discrepancy between parser-lalr and rustc

[rprichard]

It's not clear to me whether this code is valid syntax. It looks wrong, so maybe it ought to be rejected. rustc rejects it, but parser-lalr accepts it:

struct S { f: i32 }
fn main() {
 let s = S { f: 42 };
 s as S.f;
}

It looks like it should parse as s as (S.f), and that wouldn't parse. parser-lalr parses it as (s as S).f

test.rs:4:11: 4:12 error: expected one of `!`, `(`, `::`, `;`, `<`, `}`, or an operator, found `.`
test.rs:4 s as S.f;
 ^

(To test the LALR grammar, build the check-grammar target, then run grammar/parser-lalr -v.)

[brson]

Current output:

rustc 1.16.0 (30cf806ef 2017-03-10)
error: expected one of `!`, `(`, `::`, `;`, `<`, or `}`, found `.`
 --> <anon>:4:11
 |
4 | s as S.f;
 | ^

error[E0308]: mismatched types
 --> <anon>:4:5
 |
4 | s as S.f;
 | ^^^^^^ expected (), found struct `S`
 |
 = note: expected type `()`
 found type `S`

error: aborting due to previous error

[nikomatsakis]

Seems clear that parser-lalr is wrong here.

[nikomatsakis]

However, I can see why it arises. foo.bar is not a valid type, after all.

[steveklabnik]

Triage; is parser-lalr still a thing? is this issue relevant?

[Mark-Simulacrum]

I don't think we currently maintain any separate grammar - see #64896.