[SR-75] Referencing a protocol function crashes the compiler about swift HOT 8 CLOSED

Comment by Arthur Sabintsev (JIRA)

You're not supposed to call functions directly on a protocol. A class conforms to the protocol and implements the method.
Depending on if the function is static function or an instance method (in your example, it's the latter), you can call the function on the class or instance of the class.

I think your point was that the compiler is crashing due to this, and rather an error should be thrown, which is the case if you do something like:

protocol MyProtocol {
    static func MyFunc() -> String
}

let f = MyProtocol.MyFunc // error: static member 'MyFunc' cannot be used on instance of type 'MyProtocol.Protocol'

from swift.

Comment by Fabio Ritrovato (JIRA)

That's exactly my point 🙂

from swift.

Comment by Arthur Sabintsev (JIRA)

Got it. Nice find! 🙂

from swift.

We plan on making MyProtocol.someInstanceMethod work. You can already do this for classes, eg,

class Foo {
func f() { ... }
}

let f: Foo -> () -> () = Foo.f

It should have the same behavior for protocols:

protocol Foo {
func f()
}

let f: Foo -> () -> () = Foo.f

I plan on addressing this later.

from swift.

Comment by Jeremy Lew (JIRA)

@slavapestov This has been hanging around for a while, do you know if it's on any roadmap?

from swift.

Comment by Justin Guedes (JIRA)

@slavapestov Any update on this bug? 🙂

from swift.

Comment by Greg Omelaenko (JIRA)

Just wondering, how is

let f = MyProtocol.myFunc

different from

let f = { (x: MyProtocol) in x.myFunc }

? I understand this won't work if MyProtocol has an associatedtype, but it seems like this kind of expression can't reasonably work for those cases anyway. Am I missing something?

from swift.

grego (JIRA User) in fact you've hit the nail on the head; the two forms are equivalent, and the compiler should be able to handle the first just as it can handle the second.

The problem here is that partially-applied method references were emitted using a different code path than normal closures. I'm working on a change to convert these forms into ordinary closures in the type checker, which fixes various odd cases that were not supported (such as this one) and allows simplifying a lot of code in other parts of the compiler.

I've got a WIP PR up: #28698

from swift.