Tuple struct patterns can match unit structs

[petrochenkov]

This compiles on stable and nightly, but it shouldn't:

struct S;

fn main() {
    let S(..) = S;
}

[steveklabnik]

/cc @rust-lang/lang

This kind of makes sense because .. is "doesn't matter", but it's not very useful. I'm not sure if we want to disallow it or not.

[petrochenkov]

The problem is that "kinds" of structs are different, ()-structs and {}-structs never mix, normally.
I'm working on a fix, it's a direct continuation of #28336

[tbu-]

I think there was a recent RFC touching this.

[arielb1]

@petrochenkov

You can't match { } structs/variants because they don't have a value namespace entry, only unit structs/variants.

[durka]

To be honest, I don't see why S(..) can't match a {}-struct as well.

[arielb1]

@durka

You will get an unresolved enum variant, struct or const error.

[durka]

@arielb1 I know it doesn't work, but I don't see why it couldn't be made to work.

[pnkfelix]

@durka see rust-lang/rfcs#218 ; in particular https://github.com/rust-lang/rfcs/blob/master/text/0218-empty-struct-with-braces.md#empty-tuple-structs ; we are allowing unit and braced structs to mix in ways that we do not want to allow tuple structs to mix as well.

[nikomatsakis]

Regarding the ability of S(..) to matching against a "unit struct", I've always seen this as a feature -- basically a way to enumerate a variant without saying anything about how much data is in it -- but I guess it is a bit fishy, particularly given the distinction between brace and tuple structs. I'm not really sure what my opinion is on this matter yet.

[petrochenkov]

@nikomatsakis
If we follow the route mentioned in rust-lang/rfcs#218 (comment), then S{..} will naturally be a "match anything" pattern for structs/variants.