Fix parsing for casting an empty return expression

[cramertj]

Partial fix for #28784.

[rust-highfive]

r? @nikomatsakis

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

[petrochenkov]

Hm, I think you need to

• add the same condition to NtIdent below
• actually list all possible ident that can start an expression.

The list should look like this if I'm not missing anything:

• all !is_any_keyword idents
• is_path_segment_keyword idents
• box, break, continue, false, for, if, loop, match, move, return, true, unsafe, while

[cramertj]

I agree. However, I didn't get a response when I asked about the more general solution in the issue, so I decided to keep the scope small to start (just `as`). If a more general (i.e. "correct") solution is wanted, I'd be happy to do that instead.

…

On Jan 25, 2017 2:18 PM, "Vadim Petrochenkov" ***@***.***> wrote: Hm, I think you need to - add the same condition to NtIdent below - actually list all possible ident that can start an expression. The list should look like this if I'm not missing anything: - all !is_any_keyword idents - is_path_segment_keyword idents - box, break, continue, false, for, if, loop, match, move, return, true, unsafe, while — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <#39303 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AFr9KRzjtZbpiJ5cqQRypcYzcCTeEXWBks5rV8ocgaJpZM4LuCZH> .

[petrochenkov]

@cramertj
Making can_begin_expr precise is the general solution for the return sub-problem of #28784 (it affects couple of other places as well), so it's wanted.

[cramertj]

@petrochenkov Great! I'll make that change then.

[cramertj]

@petrochenkov Just started working on this, and I wanted to clarify the desired behavior for the NtIdent case. Currently, is_keyword, is_any_keyword, is_strict_keyword, etc. return false for Interpolated(NtIdent(..)) (or rather, they return false for all non-Idents). Should that be changed, too?

[petrochenkov]

@cramertj
Actually, disregard my comment about NtIdent, it looks like macro machinery always unwraps it into Ident immediately, so it can never be encountered during parsing.

[cramertj]

@petrochenkov
Does this look right to you?

fn ident_can_begin_expr(ident: ast::Ident) -> bool {
    let ident_token: Token = Ident(ident);

    !ident_token.is_any_keyword() ||
    ident_token.is_path_segment_keyword() ||
    [
        keywords::Box.name(),
        keywords::Break.name(),
        keywords::Continue.name(),
        keywords::False.name(),
        keywords::For.name(),
        keywords::If.name(),
        keywords::Loop.name(),
        keywords::Match.name(),
        keywords::Move.name(),
        keywords::Return.name(),
        keywords::True.name(),
        keywords::Unsafe.name(),
        keywords::While.name(),
    ].contains(&ident.name)
}

[cramertj]

I don't think while and for can be expressions, can they? I'm guessing you added those because of the discussions around breaking with a value.

[petrochenkov]

@cramertj

let a: () = while true { break };

They are expressions, they are just always result in () at the moment.

[cramertj]

Huh. TIL.