[SR-12108] Array.init in flatMap crash: Unrecognized selector

[swift-ci]

Previous ID	SR-12108
Radar	rdar://problem/58999154
Original Reporter	Banannzza (JIRA User)
Type	Bug

Attachment: Download

Environment

XCode 11.3 11.5

Additional Detail from JIRA

| | | |------------------|-----------------| |Votes | 0 | |Component/s | Compiler | |Labels | Bug, RunTimeCrash | |Assignee | None | |Priority | Medium | md5: c3a241e803036ffde4a8401eae0b9c38

Issue Description:

The runtime error Unrecognized selector -[__lldb_expr_1.SomeClass count] when trying to call Sequence protocol methods

Reproduce only with Class (otherwise compiler time error)

class SomeClass {}

let value: SomeClass? = SomeClass()
let array: [SomeClass] = value.flatMap(Array.init) ?? []
print(array.count)

[theblixguy]

Hmm, reproduces on 5.2 too. This doesn't crash though:

let array: [SomeClass] = SomeClass?.none.flatMap(Array.init) ?? []
print(array.count)

[beccadax]

@swift-ci create

[beccadax]

-dump-ast shows that Array.init selects an underscored initializer used for array bridging:

(declref_expr type='(Array<SomeClass>.Type) -> (AnyObject) -> Array<SomeClass>' location=<stdin>:4:46 range=[<stdin>:4:46 - line:4:46] decl=Swift.(file).Array extension.init(_immutableCocoaArray:) [with (substitution_map generic_signature=<Element> (substitution Element -> SomeClass))] function_ref=unapplied)

This is a result of our lax rules for matching unlabeled parameters.

Banannzza (JIRA User), I'm not sure which initializer you were hoping Array.init would match, but in any case you're not getting the one you wanted. You should probably write this like value.flatMap { [$0] } ?? [].

[swift-ci]

Comment by Aleksey Ostapenko (JIRA)

brentdax (JIRA User), so is everything ok that the compiler allows me to compile it?

[beccadax]

I wouldn't quite say it's "okay"—it's an undesirable behavior—but it's expected given our language rules. The language allows Array.init to match against any set of argument labels; I and several others consider this to be a mis-feature, but it is a feature designed into the language and we couldn't change it without breaking source compatibility.

It's something we could perhaps fix in Swift 6 (or whatever the next version is that allows breaking changes to the language), but it will require an evolution proposal. Such a proposal has previously been discussed in this thread: https://forums.swift.org/t/require-parameter-names-when-referencing-to-functions/27048

[beccadax]

(Sorry for the chain-reply…)

Thinking about it a little more, we could at least warn when this construct matches an initializer whose first argument label starts with an _. The compiler knows this is a convention for "no user-serviceable parts inside", and warnings don't break source compatibility.

[swift-ci]

Comment by Aleksey Ostapenko (JIRA)

brentdax (JIRA User) Thanks!

[theblixguy]

Would there be a fix-it attached to it? If so, what would it do?

[swift-ci]

Comment by aaron crespo (JIRA)

Thankfully we caught this bug today in Dev before it went out to Prod. Xcode 11.5, Swift 5.2.4.