Attachment: Download

Issue Description:

We encountered a weird reproducible crash after we submitted a minor bug fix version of our app with latest Xcode. The only change may affect seems to be Xcode and Swift version.

So I tried to make this minimal project to reproduce the issue:

https://github.com/allenhsu/SwiftStructOverflowDemo

The main code is all in ViewController.swift

class ViewController: UIViewController {
    private struct Curve {
        var points = [CGPoint]()
    }
    
//// Solution 1
//    private class Curve {
//        var points = [CGPoint]()
//    }
    
    private var curve: Curve?

    override func viewDidLoad() {
        super.viewDidLoad()
        step1()
        step2()
    }

    func step1() {
        curve = Curve()
        curve?.points = [CGPoint(x: 0, y: 0), CGPoint(x: 1, y: 1)]
        
//// Solution 2
//        var curve = Curve()
//        curve.points = [CGPoint(x: 0, y: 0), CGPoint(x: 1, y: 1)]
//        self.curve = curve
    }
    
    func step2() {
        if let curve = curve {
            print("total points in dashpoints: \(curve.points.count)")
            print("points: \(curve.points)")
        }
    }
}

I think it's straightforward and self-explaining, the issue is that in the `step2` method, the curve.points become invalid, so curve.points.count become an arbitrary large number, and if you iterate over it, it'll crash.

Although I know how to quick fix / work around this issue, I think this may be a issue introduced by the compiler, because:

1. If I turn "Optimization Level" of "Swift Compiler - Code Generation" to "No Optimization", the output is correct. That's why the issue didn't happen in development environment. Maybe that's also the direction to look into, the optimization part.

2. The issue can also be resolved if I change the step1 method into:

    func step1() {
        var curve = Curve()
        curve.points = [CGPoint(x: 0, y: 0), CGPoint(x: 1, y: 1)]
        self.curve = curve
    }

Also, if I set break point in step1 method, the curve and its points is valid. So the second guess is it may be also related to live cycle and scope of variables when being optimized.

All that said, I'm not 100% percent sure if it's a but of the compiler, or there's anything I'm doing wrong.