[SR-8787] type(of:) not returning dynamically-created subclasses

[swift-ci]

Previous ID	SR-8787
Radar	None
Original Reporter	johnestropia (JIRA User)
Type	Bug
Status	Resolved
Resolution	Won't Do

Additional Detail from JIRA

Votes	0
Component/s	Standard Library
Labels	Bug
Assignee	None
Priority	Medium

md5: c0e5a08424340faee038a65df080116f

Issue Description:

I have a Swift class "CoreStoreManagedObject" which subclasses NSManagedObject. More subclasses of this type are created with "objc_allocateClassPair()", and then some NSObject methods are overrided dynamically using "class_addMethod()", in my particular case, "keyPathsForValuesAffectingValueForKey:".

Until Swift 4.1,

type(of: object).keyPathsForValuesAffectingValueForKey("someKeyPath")

always worked as expected, returning values funneled through the created subclasses. In Swift 4.2, "type(of🙂" seems to return the compile-time class (which returns nothing). I am not sure if this is by design or if I hit a rare edge case.

For now I work around this by writing

func cs_dynamicType<T: AnyObject>(of instance: T) -> T.Type {
    
    return object_getClass(instance) as! T.Type
}

and using it in place of "type(of🙂"

I reproduced this with the Unit tests in the library https://github.com/JohnEstropia/CoreStore

Commit: 40f458a09c0177d15ceab1561acaede2799c4f2d (post-fix)

Replace "cs_dynamicType(of🙂" usage in "DynamicModelTests.swift" back to "type(of🙂" to reproduce Unit Test failure.

[belkadan]

This was a deliberate change in Swift 4.2 to bring type(of🙂 in alignment with -class, which also fixed some important issues where the Swift runtime would look for information that was associated with the original user-defined class rather than the dynamic subclass. Your cs_dynamicType seems reasonable and correct to me for when you actually need the dynamic class type value.

@jckarter, @mikeash, anything to add?

[mikeash]

I think you've pretty much covered it. Just for some additional context, one major reason for this was to make sure that Bundle(for: type(of: object)) produced the expected result, which it doesn't when type(of: object) returns the magic runtime-allocated subclass when object is the target of KVO.

[swift-ci]

Comment by John Estropia (JIRA)

Thanks for the clarification. I think the the Bundle(for🙂 usage sounds reasonable.