Precedence of type macros is confusing

[jendrikw]

When calling an associated function on a struct, type macros behave very strange and produce weird error messages.

Example on playground

struct S<T>(PhantomData<T>);

impl<T> S<T> {
    fn associated_function() {}
}

macro_rules! TypeMacro {
    ($x:ty) => { S<$x> };
}

fn main() {
    // compare the error mesages:
    
    // error: expected one of `.`, `;`, `?`, or an operator, found `::`
    let _ = TypeMacro!(u8)::associated_function();
    
    // error: chained comparison operators require parentheses
    let _ = S<u8>::associated_function();
    
    // conclusion:
    // TypeMacro!(u8) is not the same S<u8>
    
    
    // this works
    let _ = <S<u8>>::associated_function();
    let _ = S::<u8>::associated_function();
    let _ = <TypeMacro!(u8)>::associated_function();
}

Naturally, one would expect that TypeMacro!(u8) is the same as S<u8> because it is defined that way.

The result is the same on all rust versions currently in playground

• 1.27.1
• 1.28.0-beta.9 2018-07-09 718e759
• 1.29.0-nightly 2018-07-10 e5f6498

[petrochenkov]

The syntax is PATH::assoc_item, not TYPE::assoc_item, for non-path types angle brackets have to be used - <[T; N]>::assoc_item.

[T; N]::assoc_item/(A, B)::asoc_item/etc even have a special error recovery and help message for this case:

error: missing angle brackets in associated item path
 --> src/main.rs:4:13
  |
4 |     let x = [a; b]::clone;
  |             ^^^^^^^^^^^^^ help: try: `<[a; b]>::clone`

Supposedly mac!()::assoc_item should have this help as well, but it doesn't for some reason.

[jendrikw]

I've got a few questions about this.
What is a path? I don't think this is documented in the book.
Is S::<u8> the same as <S<u8>>? If so, why are there two ways to write it?

[petrochenkov]

@jendrikw

What is a path?

Something like this a::<b>::c<d>::e<f>, including single-segment cases like a<b> or a.
At least the syntax appendix in the book show it - https://doc.rust-lang.org/book/second-edition/appendix-02-operators.html#appendix-b-operators-and-symbols.

Is S::<u8> the same as <S<u8>>?

No, the <TYPE>::assoc_item is a way to write associated items for "complex" types that are not paths like *const T or [T; 10].
It's not necessary for a type like S<u8> because it's a path syntactically.

[jendrikw]

Well, I didn't know the syntax for a path because noone told me I needed a path. Even now, I don't know what it's needed for.

But do you agree that the error messages should me the same?

[petrochenkov]

Yes, the error message should ideally look like one for [T; N], e.g.

error: missing angle brackets in associated item path
 --> src/main.rs:4:13
  |
4 |     let x = mac!()::clone;
  |             ^^^^^^^^^^^^^ help: try: `<mac!()>::clone`

[durka]

The syntax is PATH::assoc_item, not TYPE::assoc_item, for non-path types angle brackets have to be used - <[T; N]>::assoc_item.

Quick note: path doesn't work either:

error: expected one of `!`, `.`, `;`, `?`, `{`, `}`, or an operator, found `::`
  --> expr.rs:18:49
   |
18 |     match {macro_rules! call { ($t:path) => { $t::foo() } } call!(Foo);} {
   |                                                 ^^          ----------- in this macro invocation
   |                                                 |
   |                                                 expected one of 7 possible tokens here

[petrochenkov]

Fixed in #59058