[twic]

The planned const generics syntax is going to look pretty weird, because it builds on type generics. If Rust had chosen a different syntax for type generics, const generics might look less weird.

The futures design was built with epoll in mind, and now people are trying to wrap it round io_uring, they are feeling some pain. Would a different design have worked better, without massive drawbacks?

[masklinn]

> The planned const generics syntax is going to look pretty weird, because it builds on type generics. If Rust had chosen a different syntax for type generics, const generics might look less weird.

Could you provide an example? Because I don't really see it, except by requiring that non-const generics also be explicitly annotated?

[twic]

I'm thinking of the need to have blocks wrapping expressions in types [1]:

  const X: usize = 7;
  
  let x: RectangularArray<i32, 2, 4>;
  let y: RectangularArray<i32, X, {2 * 2}>;

Simple cases are fine, but i suspect a lot of real-world use of const generics are going to require blocks.

AIUI, this is only because <> was chosen as the container for type arguments, and > is also a legal operator in expressions. So, for example, if Rust had used Scala-esque [] for type arguments, this would not be a problem.

[1] https://rust-lang.github.io/rfcs/2000-const-generics.html#ap...

[Uther]

While I would prefer brackets, that would not be enought to solve all the problems since, there would be conflict with array syntax too.

[twic]

But the square brackets used in array access nest, so the parsing is never ambiguous.

The reason > needs braces is that without them, if you're parsing and so far you've seen these characters:

  let y: RectangularArray<i32, X, 2>

You don't know if that final > is ending the argument list, or is a greater-than operator, with the rest of an expression coming after it.

But if it was square brackets, then a closing square bracket on its own can only ever be the end of the list:

  let y: RectangularArray[i32, X, 2]

Because if it was closing an array access, there would have to have been an opening square bracket:

  let y: RectangularArray[i32, X, a[1]

[twic]

In case anyone ever stumbles across this thread, a couple of posts about this:

https://keleshev.com/parsing-ambiguity-type-argument-v-less-...

https://soc.me/languages/stop-using-for-generics

[masklinn]

That's what I originally thought about, but the thing is `]` in and of itself is not an operator while `>` is, so

    foo<bar, qux>corge>

"is" valid syntax while

    foo[bar, qux]corge]

is not. The former requires some sort of disambiguation, while the latter is not. Put an other way, in terms of syntax `[]` always parses the same way (with `[` is infix and `]` terminates it), there can be ambiguity between indexing and generics but it doesn't really matter for the original parsing.

For `<>` however, the parsing itself can be ambiguous as `>` could either be the terminator of an earlier `<` or it could be the infix `>` operator. That is where the issue lies, you can't know how to build the AST without either explicit disambiguation, or infinite lookaheads.