[SR-8756] NSDecimalNumber(string:) regression in Xcode 10 GM (Returns NaN).

[swift-ci]

Previous ID	SR-8756
Radar	None
Original Reporter	freak4pc (JIRA User)
Type	Bug

Environment

Xcode 10 GM
Xcode 9.4.1

Additional Detail from JIRA

Votes	0
Component/s	Compiler
Labels	Bug, 4.2Regression
Assignee	None
Priority	Medium

md5: 215ce993c53eac04883adc50b9b63c05

Issue Description:

When running our unit tests, spotted a weird regression in some conversion to `NSDecimalNumber` that's acting very odd, specifically in Xcode 10.

I reduced the issue to as few lines as possible.

Given the following, seemingly unrelated extension:

extension Dictionary where Key == String {
    func value<T>(for key: Key, or: T) -> T {
        return self[key] as? T ?? or
    }
}

The following code behaves irregularly (see Example 2). Seems related to inference but not entirely sure how, since the "String" return type is properly explicit; having this wasn't an issue in Xcode 9.

let json = ["x": "5", "b": "12"]

// ### Example 1 ###
// #################
// Returns 5, as expected, on all versions of Xcode
print(#line, NSDecimalNumber(string: json.value(for: "x", or: "0")))

// ### Example 2 ###
// #################
// Xcode 9.4.1 Swift 4.1: Prints 0.
// Xcode 10 GM Swift 4.2: Should be 0, Printing NaN.
// Xcode 10 GM Swift 4 mode: Should be 0, Printing NaN.
print(#line, NSDecimalNumber(string: json.value(for: "y", or: "0")))

// ### Example 3 ###
// #################
// Same, but explicitly casting, works correctly on all versions of Xcode
print(#line, NSDecimalNumber(string: json.value(for: "y", or: "0") as String))

// ### Example 4 ###
// #################
// Same, when extracting string separately.
// Works as expected on all versions of xcode
let value = json.value(for: "y", or: "0") // Returns "0" on all versions of Xcode
print(#line, NSDecimalNumber(string: value))

Appreciate your help!
Shai.

[belkadan]

NSDecimalNumber.init(string:) takes an optional String, so T is getting inferred to String? rather than String in Xcode 10. That matches the dictionary lookup, and then you pass nil to the initializer.

@rudkx, we have another bug for this behavior change, right?