Syntax

[nrc]

Previous discussion

There is an open question around what syntax (if any) should be required to cause the deref function to be considered in a pattern. From this Zulip discussion the currently preferred option is 'no syntax'. I.e., we extend the match ergonomics rules to apply in any situation where there is a Deref impl, not just in the &T case (to be more precise, where the scrutinee (or a sub-part) has type &T and there is Deref impl for T, then we can match a pattern (or sub-pattern) with type <T as Deref>::Target.

So, the following examples should type check:

fn proposed(a: Option<String>, b: Rc<Box<&Vec<i32>>>) {
    match a {
        Some("hello") => {}
        _ => {}
    }

    match b {
        [1, 2, 3] => {}
        _ => {}
    }
}

[AlbertMarashi]

Would it be possible to have a kind of "path" based pure de-referencing system, so it wouldn't involve arbitrary code execution?

Eg

struct Box<T> {
   pointer: *const T for deref
}
// OR
struct Box<T> {
  pointer: *const T for pattern
}
// OR (this involves no new keywords) & fully backwards compatible
struct Box<T> {
  pointer: *const T for match
}

// and then...

enum Foo {
  A,
  B,
}
let item = Box::new(Foo::A);
match a_box {
  Foo::A => ...,
  Foo::B => ...
}

Can anyone foresee any issues with this kind of syntax?