[SR-142] mutating function in protocol extension erroneously requires `var` declaration of class variables

[groue]

Previous ID	SR-142
Radar	None
Original Reporter	@groue
Type	Bug
Status	Resolved
Resolution	Done

Environment

Version 7.1.1 (7B1005)

Additional Detail from JIRA

Votes	1
Component/s	Compiler
Labels	Bug, LanguageFeatureRequest
Assignee	None
Priority	Medium

md5: c88df0cab57ceac17b5038fd1dff4b54

is duplicated by:

• SR-2464 Reference types can't call mutating functions declared in a protocol extension.
• SR-2693 Cannot assign to property of an instance of a class declared in a protocol extension
• SR-4229 Cannot use mutating member on immutable value: 'self' is immutable
• SR-7636 Reference type has struct mutability rules applied to it
• SR-8805 Cannot set protocol extension property on adopting reference type which is immutable
• SR-4227 Can't call mutating protocol extension method in class init
• SR-6120 Bug with Class and Subtype existentials, when protocol doesn't have class requirement
• SR-10775 Mutating method in protocol extension always requires variable to be var

relates to:

• SR-2220 A protocol's mutating keyword is enforced on AnyObject when using generics
• SR-4541 Class-Only protocol inheriting from normal protocol require to be var to allow assign to their properties.

Issue Description:

Hello,

The following code won't compile, when it should:

protocol P { }

extension P {
    mutating func m() { }
}

class C : P { }

let c = C()
// error: cannot use mutating member on immutable value: 'c' is a 'let' constant
c.m()

This error should not exist: the type of variable c is C. Mutating methods can not be added to C, because of the 'mutating' isn't valid on methods in classes or class-bound protocols error. Hence the mutating qualifier of P.m should not apply to c, and c should be able to be declared as a let variable.

[groue]

The issue is also there in Apple Swift version 2.2-dev (LLVM 46be9ff861, Clang 4deb154edc, Swift 778f829)

[groue]

See groue/GRDB.swift#12 (comment) for a practical problem raised by this issue: some APIs have to expose a base class instead of protocol as soon as they want to let adopting user types "override" default implementations. (updated to invalidate)

[groue]

For what it's worth, there is a (verbose) workaround, which is to redeclare the mutating method in the adopting class:

protocol P { }

extension P {
    mutating func m() { }
}

class C : P {
    // redeclare m() without the mutating qualifier
    func m() {
        // call protocol's default implementation
        var p: P = self // mandatory temp since (self as P).m() won't compile
        p.m()
    }
}

// OK, no error
let c = C()
c.m()

[gregomni]

Looked into this briefly, and it's a lot harder than just changing the error checking, since m() has an implicit `inout P` first argument. Probably the solution is an automatic version of Gwendal's workaround: declaring a hidden `var p: P = c` and calling the method on p.

[groue]

Thanks for the investigation, Greg!

[lilyball]

Automatically declaring a hidden var like that would cause surprising behavior when the mutating method actually assigns to self:

protocol P {
    init(name: String)
    mutating func foo()
}

extension P {
    mutating func foo() {
        self = Self.init(name: "foo")
    }
}

class C : P {
    var name: String
    required init(name: String) {
        self.name = name
    }
}

let c = C(name: "a")
c.foo() // assuming automatic creation of hidden variable
print(c.name)

With the automatic hidden variable, the c after c.foo() must still be the c from before the call (because it is let), but that means the mutating method didn't actually do anything.

[groue]

Thanks Kevin. I overlooked self assignment in default implementation, which requires the implementation described by Greg.

So it isn't a bug, but actually a feature, and anyone bothered by this behavior should discuss it on the swift-evolution mailing list, isn't it?

[swift-ci]

Comment by Josh Avant (JIRA)

For posterity, here's a link to additional conversation about this on swift-evolution:
https://lists.swift.org/pipermail/swift-evolution/Week-of-Mon-20151207/001575.html

[slavapestov]

As Kevin explained this is not a bug, because the method implementation may assign to self.