[SR-1999] Inconsistency when casting Optional to Any and downcasting to non-Optional

[swift-ci]

Previous ID	SR-1999
Radar	None
Original Reporter	hbr (JIRA User)
Type	Bug
Status	Resolved
Resolution	Done

Environment

XCode 7.3
Swift 2.2
OSX 10.11.5
(also crashes with Swift 3)

Additional Detail from JIRA

Votes	0
Component/s	Compiler
Labels	Bug, Runtime
Assignee	@tbkka
Priority	Medium

md5: a4bc4c2e8aa88d6efc7c6a9cfcd04be5

relates to:

• SR-6279 Failing cast from Any to Optional

Issue Description:

Description

Consider a concrete type T extending a protocol P (class T : protocol P)
Cast an instance of T? (optional) to Any, and then downcast it :

• it seems valid to downcast to T (i.e : it works)

• on the contrary, downcasting to P yields a runtime exception :

  Could not cast value of type 'Swift.Optional<T>' to 'P'

Is there any reason why downcasting to a class and downcasting to a protocol implemented by the same class should yield different results ?
I think one should either get an exception in both cases, or no exception at all.

Rmq : this issue is not purely theoretical ; I bumped into it when sending optional values to a Swift pod making heavy use of generics (and heavy use of downcasts...)

Test case

Paste this in the main.swift of an XCode 7.3 OSX Command Line Tool project :

protocol FooProtocol {
    func doSomething()
}

class Foo: FooProtocol {
    func doSomething() {
        print("hello")
    }
}

let optionalFoo : Foo? = Foo()
let anyValue: Any = optionalFoo
    
let foo1 = anyValue as! Foo
foo1.doSomething()                  // prints 'hello', ok

let foo2 = anyValue as! FooProtocol // Could not cast value of type 'Swift.Optional<Foo>' to 'FooProtocol'
foo2.doSomething()

[belkadan]

@jckarter, @slavapestov, we've seen this before, yeah?

[slavapestov]

Yeah, there are definitely some funny edge-cases around optionals. I would stay away from casts that inject or eliminate optionality altogether, and use other constructs, such as pattern matching, instead.

Having said that, we should definitely address these issues, it is just that at this point it is not even completely clear what the correct behavior should be.

[swift-ci]

Comment by Hervé Bérenger (JIRA)

Thanks a lot for your quick feedback.
I guess it's related to other issues such as SR-912.
I'll try to workaround this using pattern matching as you suggest.

[tbkka]

#33561 should fix this for the non-optimized case (where the cast is being handled by the runtime). There may still be issues with optimized casts, though.

[swift-ci]

Comment by Hervé Bérenger (JIRA)

@tbkka : thanks a lot for the follow-up !

[tbkka]

After further testing, I believe this is fully fixed now.