[SR-9975] Protocol Extension Bugs

[swift-ci]

Previous ID	SR-9975
Radar	rdar://problem/48315560
Original Reporter	jesands (JIRA User)
Type	Bug

Environment

Xcode Version 10.1 (10B61)
Apple Swift version 4.2.1

All examples run in a Swift Playground.

Additional Detail from JIRA

Votes	0
Component/s	Compiler
Labels	Bug
Assignee	None
Priority	Medium

md5: d4d208ed9f32df731ab2823a18e0708c

Issue Description:

There are several issues surrounding protocols with default implementations in extensions. The first two here are definitely bugs, the third is arguably a bug.

1. Incorrect behavior for statically defined constants

protocol A {
    var a: Int { get }
}

extension A {
    var a: Int { 
        return 1
    }
}

class Implementation: A {
    static let a: Int = 2
    let a = Implementation.a
}

let implementation = Implementation()
print(implementation.a)
print(implementation.a as Int)
print((implementation as A).a as Int)

Expected output: "2 2 2"
Actual output: "2 2 1".
Note: Moving the static variable declaration to another class produces the expected output. Declaring the `let a` as `let a: Int` also produces the expected output.

2. Declaring uninitialized variables not caught by the compiler:

class Z {}

protocol A {
    var z: Z { get }
}

extension A {
    var z: Z {
        return Implementation().z
    }
}

struct Implementation: A {
    let z = globalZ
}

let globalZ = Implementation().z

Expectation: compiler error. Circular definition.
Actual error: "Execution was interrupted, reason: EXC_BAD_ACCESS (code=1, address=0x0)."

3. Type based variable overloading

protocol A {
    var a: Float { get }
}

extension A {
    var a: Float {
        return 1
    }
}

class Implementation: A {
 let a = 2
}

let implementation = Implementation()
print(implementation.a)
print(implementation.a as Float)
print((implementation as A).a)

Expected output: Compiler error or "2.0 2.0 2.0".
Actual output: "2 1.0 1.0"

[belkadan]

#2 and #3 are tracked elsewhere (I'll find them in a minute), but #1 seems valid!

[belkadan]

Okay, #2 is rdar://problem/35633741 but doesn't seem to have an SR. #3 is SR-6689.

Let's use this report to track #1.