Still fresh in new config

Xcode Version 7.2 (7C68)
OSX 10.11.2 (15C50)
The Swift that ships with this Xcode

Michael, who'd be a good candidate to look into this?

In general you should send these to Nadav. But it is fine to park them on me in the short term.

To the reporter, we are going to be using the other bug tracker field to relate to our internal bug tracking system (I am not sure if in JIRA we can do multiple, but we should look into this).

So I need to relate this to our internal radar number. Here is the issue you posted: https://code.sealedabstract.com/drewcrawford/CaveJohnson2/issues/10.

I am looking at this now.

I have reproduced a crash with optimization given this compiler:

Apple Swift version 2.1.1 (swiftlang-700.1.101.13 clang-700.1.81)
Target: x86_64-apple-darwin15.0.0

I am trying with ToT.

This is my single file test case.

import Foundation

public final class ServerConfig { }

public final class XCSAPI {
private static func buildRequest(path: String, method: String = "GET", serverConfig: ServerConfig) throws -> [String: AnyObject] {
print("path (path)")
let url = NSURL(string: "https://:20343(path)")!
print("Data ("test".dataUsingEncoding(NSUTF8StringEncoding)!)")
return [:]
}
public static func integrateNow(botID: String, serverConfig: ServerConfig) throws -> Int {
//print(botID)
let response = try buildRequest("/api/bots/(botID)/integrations", method: "POST", serverConfig: serverConfig)
return 0
}

}

print("Hello, World!")

try! XCSAPI.integrateNow("test", serverConfig: ServerConfig())

I am reproducing with ToT btw. I am bisecting now.

It looks like the pass that exposes the issue is ARCSequenceOpts. It may just be exposing other bad behavior though.

Ok I figured out the problem. It is an issue with ARC but it is relatively benign in the sense that I already know the solution and it shouldn't be too bad. The issue is how we are handling the use of a try_apply insertion point. I thought due to a quirk in the way we compute insertion points this would not come up, but I was incorrect. So now I need to make the appropriate change to handle this correctly.

Turns out a similar problem can occur with loops if our pre-header insertion fails. I fixed this in:

To git@github.com:apple/swift.git
61e2b30..b235064 master -> master

I have a fix, I am just writing more tests/checking for any perf issues.

I have everything waiting to go. I am just waiting for Dmitri to fix the perf test suite so I can measure before/after performance just in case.

To git@github.com:apple/swift.git
86e6b81..4f69bfc master -> master