[SR-7125] Swift 4.1 Xcode 9.3b4 regression

[keith]

Previous ID	SR-7125
Radar	rdar://problem/38159133
Original Reporter	@keith
Type	Bug
Status	Resolved
Resolution	Done

Environment

Xcode 9.3b4

Additional Detail from JIRA

Votes	0
Component/s	Compiler
Labels	Bug, 4.1Regression, TypeChecker
Assignee	@xedin
Priority	Medium

md5: c85e0d6950ba1dfcfa34d061adc53271

Issue Description:

This code compiles successfully in Xcode 9.2 with Swift 4:

protocol Bar {}
class Superclass {}

class SubclassA: Superclass, Bar {}
class SubclassB: Superclass, Bar {}

class Foo {
    var a: SubclassA? { return nil }
    var b: SubclassB? { return nil }

    func bar() {
        let bars: [Bar] = [
            self.a,
            self.b,
        ].flatMap { $0 }
    }
}

With Swift 4.1 this fails with this error:

foo.swift:15:11: error: 'flatMap' produces '[String]', not the expected contextual result type '[Bar]'
        ].flatMap { $0 }

If I change the `flatMap` to the new `compactMap` I get a slightly different error:

foo.swift:15:27: error: cannot convert value of type 'Superclass?' to closure result type 'Bar?'
        ].compactMap { $0 }

Changing it to this works as expected:

let bars: [Bar] = [
    self.a,
    self.b,
].compactMap { $0 as? Bar }

This is new code in our codebase since Xcode 9.3b3, so I'm not sure if this is new in beta 4 or not.

[jckarter]

cc @rudkx, @xedin. It seems to me this has a legitimate reason for compiling previously, and still should be able to compile using `compactMap`. Working backward from the contextual result type `[Bar]`, we should be able to infer `[Bar?]` for the array literal type, which is a supertype of all of the elements of the array literal.

[jckarter]

@swift-ci create

[jckarter]

Does it also work if you use $0 as Bar? to give explicit type information to the closure body?

[keith]

It does compile with that in the closure body, that's the last example above. It also compiles with this:

let bars: [Bar] = [
    self.a as Bar?,
    self.b as Bar?,
].compactMap { $0 }

[jckarter]

Sorry, I read that you had as? Bar in the closure body in your last example, which is slightly different from as Bar?.

[keith]

Ah sorry I misread that! No that fails with this error:

bar.swift:15:27: error: 'Superclass?' is not convertible to 'Bar?'; did you mean to use 'as!' to force downcast?
        ].compactMap { $0 as Bar? }
                       ~~~^~~~~~~
                          as!

[xedin]

@rudkx It seems like we do something very wrong here since with `$0 as? Bar` the result of the closure is inferred to be `Superclass` and `$0` to be `Superclass?` instead of `Bar` and `Bar?`. I'm going to take a look, it might be an easy fix.

[xedin]

@swift-ci create

[xedin]

The problem should be resolved by #15009 @keith, thanks for reporting and you can verify the fix by using next available nightly snapshot.

[keith]

@xedin I tested the master snapshot from 3/8 and it does compile now. Thanks! What's the process for getting that fixed in the 4.1 branch?