[SR-41] Covariance does not work for functions with variadic arguments

[swift-ci]

Previous ID	SR-41
Radar	rdar://23011851
Original Reporter	nikita (JIRA User)
Type	Bug

Attachment: Download

Environment

Xcode 7.1.1 (7B1005)

Additional Detail from JIRA

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

md5: 1344eed40e3fc0135966b4277a47f007

relates to:

• SR-27 Passing variadic function as a function parameter crashes compiler with SIGSEGV

Issue Description:

Swift 2.2 covariance unable to resolve functions covariance with variadic arguments.

Steps to Reproduce:
Attached a playground that shows the problem, as well as screenshot with errors. Here is a copy & paste of the code for your convenience:

func giveMeVoid(callback: (Void)->Void) {
    
}

func giveMeInt(callback: (Int)->Void) {
    
}

func any(values:Any?...) {
    
}

any()
any(5)

giveMeVoid(any) //This line will fail
giveMeInt(any) //This line will fail

Expected Results:
It is expected that `any` will be accepted by both giveMeVoid and giveMeInt functions, as they can work as `Void` and `Int` functions when they called directly.

Actual Results:
Non of functions accept 'any'. Following errors are thrown:

error: cannot convert value of type '(Any?...) -> ()' to expected argument type '(Void) -> Void'
error: cannot convert value of type '(Any?...) -> ()' to expected argument type '(Int) -> Void'

[CodaFi]

This is not a variance issue, this is an implicit coercion that you're asking for. We would need to add additional thunks to help with the wrapping here. @rjmccall how feasible is something like this?

[rjmccall]

We're not really set up for it; the subtype-conversion logic really wants to translate specific arguments, not completely rework them structurally. It could be fixed, though.

This is an interesting idea, but I'm not sure we want to generalize function conversions to apply to literally anything where you could forward the arguments this way.