test with CheckedContinuation loses testing context and crashes

[bernoulliconvergenator]

Description

Consider an Observable model that interacts with an actor model:

actor ActorModel {
   func doMyThing() -> Int {
      return 1
   }
}

@Observable final class ObservableModel: @unchecked Sendable {
   private(set) var actorValue: Int?
   private let actorModel = ActorModel()

   @MainActor func buttonPressed() {
      actorValue = nil // reset

      Task.detached {
         let newActorValue = await self.actorModel.doMyThing()
         #expect(newActorValue == 1)
         Task { @MainActor [weak self] in
            self?.actorValue = 2 // *** !! OOPS ***
         }
      }
   }
}

To test that buttonPressed() resets actorValue and then asynchronously assigns actorValue after calling actorModel.doMyThing(), we can construct a test with observation tracking inside a checked continuation:

@Test @MainActor func shouldFail() async throws {
   let observableModel = ObservableModel()
   try #require(observableModel.actorValue == nil)

   try await withCheckedThrowingContinuation { (cont: CheckedContinuation<Void, Error>) -> Void in
      @Sendable func observeActorValue(count: Int) {
         withObservationTracking {
            print("observing observableModel.actorValue=\(String(describing: observableModel.actorValue)) count=\(count)")
         } onChange: {
            Task { @MainActor in
               print("onChange observableModel.actorValue=\(String(describing: observableModel.actorValue))")
               switch count {
               case 0:
                  #expect(observableModel.actorValue == nil)
               case 1:
                  #expect(observableModel.actorValue == 1)
                  cont.resume()
               default:
                  Issue.record("recursed on observeActorModel (\(count) times, expected once")
               }
               observeActorValue(count: count + 1) // recurse
            }
         }
      }
      observeActorValue(count: 0)

      observableModel.buttonPressed()

      do {
         try #require(observableModel.actorValue == nil)
      } catch {
         cont.resume(throwing: error)
      }
   }
}

Because of the OOPS, the #expect in case 1: fails. But we also get a crash:

◇ Test shouldFail() started.
observing observableModel.actorValue=nil count=0
onChange observableModel.actorValue=nil
observing observableModel.actorValue=nil count=1
onChange observableModel.actorValue=Optional(2)
✘ Test «unknown» recorded an issue at CheckedContinuationMultipleEventsShouldFail.swift:24:19: Expectation failed: (observableModel.actorValue → 2) == 1
XCTest/HarnessEventHandler.swift:168: Fatal error: Issue recorded event did not contain a test

The test passes if we set the correct value in buttonPressed() (ie no OOPS), but are we certain the passing #expect is realized by the testing context?

Expected behavior

The test should record an issue and not crash.

Actual behavior

The test crashed.

Steps to reproduce

BugSwiftTestingCheckedContinuation.zip

swift-testing version/commit hash

Version 16.1 (16B40)

Swift & OS version (output of swift --version ; uname -a)

swift-driver version: 1.115 Apple Swift version 6.0.2 (swiftlang-6.0.2.1.2 clang-1600.0.26.4)
Target: arm64-apple-macosx15.0
Darwin BernouliConvergenator.local 24.1.0 Darwin Kernel Version 24.1.0: Thu Oct 10 21:00:32 PDT 2024; root:xnu-11215.41.3~2/RELEASE_ARM64_T6030 arm64

[grynspan]

Duplicate of #480

[bernoulliconvergenator]

Workaround option: https://github.com/bernoulliconvergenator/BCTest