[5.0] DefiniteInitialization: Storing back to the 'self' box in a class init is OK.

[jckarter]

Explanation: Fixes a regression where spurious "self used before super.init" errors would occur when an argument to a self.init or super.init delegation threw.

Scope: Regression from 4.0

Issue: rdar://problem/37007554

Risk: Low, small bug fix

Testing: Swift CI, test case from Radar

[jckarter]

@swift-ci Please test

[jckarter]

@gottesmm Does this look OK for the 5.0 branch?

[swift-ci]

Build failed
Swift Test Linux Platform
Git Sha - 453c439

[jckarter]

@phausler You know about this issue on the 5.0 Linux bot?

Foundation/Data.swift:998:15: error: type 'Data' does not conform to protocol 'Collection'
public struct Data : ReferenceConvertible, Equatable, Hashable, RandomAccessCollection, MutableCollection, RangeReplaceableCollection {
              ^
Foundation/Data.swift:1008:22: note: possibly intended match 'Data.Indices' (aka 'Range<Int>') does not conform to 'Collection'
    public typealias Indices = Range<Int>
                     ^

[phausler]

This is a regression, perhaps related to the changes made for ranges recently?
Cc @airspeedswift

[airspeedswift]

Ah, so the problem seems to be that we didn't create a 5.0 branch for corelibs-foundation, so if you take advantage of a new feature on master, it'll break the swift 5.0 branch tests.

Let me see if I can check in a workaround for now.