[SR-9654] Swift 5: Compiler crashes on SILGen for subscript with autoclosure argument

[swift-ci]

Previous ID	SR-9654
Radar	rdar://problem/47254122
Original Reporter	broadway_lamb (JIRA User)
Type	Bug
Status	Resolved
Resolution	Done

Environment

5.0-DEVELOPMENT-SNAPSHOT-2019-01-13-a

Apple Swift version 5.0-dev (LLVM ed64f75263, Clang e5012f3421, Swift 5a827a9)

Target: x86_64-apple-darwin18.2.0

Additional Detail from JIRA

Votes	0
Component/s	Compiler
Labels	Bug, CodeCoverage, CompilerCrash
Assignee	@vedantk
Priority	Medium

md5: aaa569abe2db3ea87736204ac2ff957c

Issue Description:

This code crashes the compiler when code coverage is enabled:

struct S {
  subscript(i: Int, autoclosure: @autoclosure () ->  Int) -> Int {
    get {
      return 0
    }
    set {

    }
  }
}

I am using the latest (5.0-DEVELOPMENT-SNAPSHOT-2019-01-13-a) snapshot.

The SPM project to reproduce the issue: https://github.com/broadwaylamb/CompilerCrasherAutoclosureSubscript

Run:

$ swift build
$ swift test --enable-code-coverage

--enable-code-coverage is important.

[belkadan]

cc @vedantk, @jckarter

[vedantk]

When this program is built with the `-enable-testing` flag, the swift compiler synthesizes a top-level with an invalid source location. This confuses the coverage pass, which expects valid source locations on decls always.

It probably doesn’t make sense to fake a source location, so I added a defensive check in #21854

[swift-ci]

Comment by Sergej Jaskiewicz (JIRA)

@vedantk awesome, thank you!

[swift-ci]

Comment by Sergej Jaskiewicz (JIRA)

@vedantk I was wondering will this make it into Swift 5?

[vedantk]

At this point swift-5.0-branch isn’t changing very much. I’m not sure whether or not this change is a candidate for it.

[swift-ci]

Comment by Sergej Jaskiewicz (JIRA)

@vedantk that is unfortunate, since it is clearly a regression, the same code works just fine with Swift 4.

[belkadan]

Bob ended up pulling it in here: #22798