[Concurrency] Correct enqueue behavior in with... functions by adopting nonisolated(nonsending) #80753

ktoso · 2025-04-11T07:24:15Z

The problem is that even with the #isolation parameter the non-Sendable
async closure operation still would potentially hop off the caller
isolation. We introduced this change because the plan was the
non-Sendable closure would run on the isolated parameter's isolation,
but that's not actually the case: an "re-enqueue" can still happens in those cases.

Instead, we can use the nonisolated(nonsending) on the function and closure,
in order to guarantee there is no hop between those at all, and
developers can trust that adding this cancellation handler will not
cause any unexpected isolation changes and hops.

The API was always documented to not hop as we execute the operation,
so this brings the correct and expected behavior.

This corrects the following APIs:

withCheckedContinuation
withUnsafeContinuation
withCheckedThrowingContinuation
withUnsafeThrowingContinuation
withTaskGroup
withThrowingTaskGroup
withDiscardingTaskGroup
withThrowingDiscardingTaskGroup
TaskLocal.withValue

as well as task group's next() and related ones...

This gets down the "re-enqueue" count of those methods by far, and makes all these more reliable and predictable.

resolves rdar://140110775
resolves rdar://162192512

ktoso · 2025-04-12T01:37:11Z

This is blocked by incorrect behavior of the caller execution on a closure rdar://149107104

ktoso · 2025-04-30T03:19:35Z

Now blocked on rdar://150017382

ktoso · 2025-05-15T07:23:10Z

@swift-ci please smoke test

stdlib/public/Concurrency/CMakeLists.txt

gottesmm · 2025-06-24T18:06:54Z

@swift-ci smoke test

gottesmm · 2025-07-07T18:53:35Z

@swift-ci smoke test

ktoso · 2025-07-08T07:58:49Z

Good news, this was now unblocked by #82858

stdlib/public/Concurrency/TaskCancellation.swift

ktoso · 2025-07-08T09:32:40Z

@swift-ci please smoke test

ktoso · 2025-07-08T11:17:37Z

This is almost unblocked, we need to fix throwing nonisolated nonsending (caller isolated) closures to not lose isolation -- rdar://155313349

And there seems to be a problem with noncopyable types and nonisolated nonsending (caller isolated) closures as well; It shows up in cancellation_handler_only_once.swift:20 as a error: copy of noncopyable typed value. -- rdar://155316655

Giving it a run to see if anything else

ktoso · 2025-07-08T11:44:11Z

@swift-ci please smoke test

ktoso · 2025-07-11T02:36:05Z

By now this is unlikely to make it into 6.2.0

ktoso · 2025-10-15T14:58:08Z

@swift-ci please smoke test

ktoso · 2025-10-15T14:58:16Z

Fixing all the ABI tests to follow.

ktoso · 2025-12-15T11:29:56Z

@swift-ci please smoke test

jamieQ

some further unsolicited feedback/questions for your consideration :)

stdlib/public/Concurrency/TaskGroup.swift

stdlib/public/Concurrency/TaskLocal.swift

jamieQ · 2025-12-15T14:20:45Z

test/Concurrency/Runtime/cancellation_handler_operation_does_not_hop.swift

+func globalTestFunc(isolation: isolated (any Actor)? = #isolation) async {
+  isolation!.assertIsolated("wat in \(#function)!")
+  await withTaskCancellationHandler {
+     isolation!.assertIsolated("wat in \(#function)!")


might be good to test that this works without a capture of the isolated parameter too, to ensure it's relying on the nonisolated(nonsending)-ness of the closure, and not the isolated capture logic.

ktoso · 2025-12-16T23:30:17Z

@swift-ci please smoke test

jamieQ · 2025-12-19T11:40:32Z

stdlib/public/Concurrency/TaskCancellation.swift

-#endif
-public func withTaskCancellationHandler<T>(
+@export(implementation)
+public nonisolated(nonsending) func withTaskCancellationHandler<T>(


another case that occurred to me recently that probably should get the same treatment (though probably is quite a bit less commonly used): withTaskPriorityEscalationHandler

ktoso · 2026-01-14T03:05:29Z

@swift-ci please smoke test

ktoso · 2026-01-14T11:10:45Z

Builds timed out heh

ktoso · 2026-01-14T11:11:27Z

A bunch of real failures to follow up on:

17:00:49 Failed Tests (19):
17:00:49 Swift(macosx-x86_64) :: Concurrency/Reflection/reflect_task.swift
17:00:49 Swift(macosx-x86_64) :: Concurrency/Runtime/actor_counters.swift
17:00:49 Swift(macosx-x86_64) :: Concurrency/Runtime/actor_dynamic_subclass.swift
17:00:49 Swift(macosx-x86_64) :: Concurrency/Runtime/async_let_fibonacci.swift
17:00:49 Swift(macosx-x86_64) :: Concurrency/Runtime/async_task_executor_nsobject.swift
17:00:49 Swift(macosx-x86_64) :: Concurrency/Runtime/async_task_sleep.swift
17:00:49 Swift(macosx-x86_64) :: Concurrency/Runtime/async_task_sleep_cancel.swift
17:00:49 Swift(macosx-x86_64) :: Concurrency/Runtime/async_taskgroup_cancellation_race.swift
17:00:49 Swift(macosx-x86_64) :: Concurrency/Runtime/async_taskgroup_discarding_dontLeak.swift
17:00:49 Swift(macosx-x86_64) :: Concurrency/Runtime/async_taskgroup_discarding_neverConsumingTasks.swift
17:00:49 Swift(macosx-x86_64) :: Concurrency/Runtime/async_taskgroup_next_on_completed.swift
17:00:49 Swift(macosx-x86_64) :: Concurrency/Runtime/async_taskgroup_next_on_pending.swift
17:00:49 Swift(macosx-x86_64) :: Concurrency/Runtime/async_taskgroup_throw_rethrow.swift
17:00:49 Swift(macosx-x86_64) :: Concurrency/Runtime/cancellation_handler_concurrent.swift
17:00:49 Swift(macosx-x86_64) :: Concurrency/Runtime/sleep_executor.swift
17:00:49 Swift(macosx-x86_64) :: Concurrency/Runtime/startImmediately.swift
17:00:49 Swift(macosx-x86_64) :: Concurrency/Runtime/startImmediately_order.swift
17:00:49 Swift(macosx-x86_64) :: Concurrency/ShrinkWrap.swift
17:00:49 Swift(macosx-x86_64) :: Sanitizers/tsan/norace-task-group-cancellation.swift

Though at least we're past not compiling into runtime issues now, that's progress.

This drastically lessens the number of unexpected task enqueues and hops that these APIs cause, and therefore improves performance and even correctness in a few cases (like the withContinuation APIs)

ktoso · 2026-01-18T11:23:44Z

rebased

ktoso · 2026-01-18T11:23:50Z

@swift-ci please smoke test

xedin · 2026-01-18T17:04:34Z

@ktoso there is a linux failure about attributes:

 /home/build-user/swift/stdlib/public/Concurrency/TaskCancellation.swift:99:2: error: cannot use '@_silgen_name' and '@abi' on the same global function because they serve the same purpose

ktoso · 2026-01-20T05:30:34Z

@swift-ci please smoke test

ktoso · 2026-01-20T10:31:27Z

Need to figure out how to resolve this in swift-testing:

15:59:08  /home/build-user/swift-testing/Sources/Testing/Running/Runner.RuntimeState.swift:211:22: error: sending 'body' risks causing data races [#SendingRisksDataRace]
15:59:08  209 |     runtimeState.testCase = nil
15:59:08  210 |     return try await Runner.RuntimeState.$current.withValue(runtimeState) {
15:59:08  211 |       try await test.withCancellationHandling(body)
15:59:08      |                      |- error: sending 'body' risks causing data races [#SendingRisksDataRace]
15:59:08      |                      `- note: sending nonisolated(nonsending) task-isolated 'body' to nonisolated instance method 'withCancellationHandling' risks causing data races between nonisolated and nonisolated(nonsending) task-isolated uses
15:59:08  212 |     }
15:59:08  213 |   }
15:59:08  
15:59:08  /home/build-user/swift-testing/Sources/Testing/Running/Runner.RuntimeState.swift:246:26: error: sending 'body' risks causing data races [#SendingRisksDataRace]
15:59:08  244 |     runtimeState.testCase = testCase
15:59:08  245 |     return try await Runner.RuntimeState.$current.withValue(runtimeState) {
15:59:08  246 |       try await testCase.withCancellationHandling(body)
15:59:08      |                          |- error: sending 'body' risks causing data races [#SendingRisksDataRace]
15:59:08      |                          `- note: sending nonisolated(nonsending) task-isolated 'body' to nonisolated instance method 'withCancellationHandling' risks causing data races between nonisolated and nonisolated(nonsending) task-isolated uses
15:59:08  247 |     }
15:59:08  248 |   }
15:59:08  
15:59:08  /home/build-user/swift-testing/Sources/Testing/Test+Discovery.swift:93:36: error: sending 'taskGroup' risks causing data races [#SendingRisksDataRace]
15:59:08   91 |             }
15:59:08   92 |           }
15:59:08   93 |           result = await taskGroup.reduce(into: result) { $0.insert($1) }
15:59:08      |                                    |- error: sending 'taskGroup' risks causing data races [#SendingRisksDataRace]
15:59:08      |                                    `- note: sending nonisolated(nonsending) task-isolated 'taskGroup' to nonisolated instance method 'reduce(into:_:)' risks causing data races between nonisolated and nonisolated(nonsending) task-isolated uses
15:59:08   94 |         }
15:59:08   95 |       }
15:59:08  
15:59:08  /home/build-user/swift-testing/Sources/Testing/Test+Discovery.swift:107:36: error: sending 'taskGroup' risks causing data races [#SendingRisksDataRace]
15:59:08  105 |             }
15:59:08  106 |           }
15:59:08  107 |           result = await taskGroup.reduce(into: result) { $0.insert($1) }
15:59:08      |                                    |- error: sending 'taskGroup' risks causing data races [#SendingRisksDataRace]
15:59:08      |                                    `- note: sending nonisolated(nonsending) task-isolated 'taskGroup' to nonisolated instance method 'reduce(into:_:)' risks causing data races between nonisolated and nonisolated(nonsending) task-isolated uses
15:59:08  108 |         }
15:59:08  109 |       }

ktoso · 2026-01-21T06:21:52Z

swiftlang/swift-testing#1493
@swift-ci please smoke test

ktoso · 2026-01-21T11:37:37Z

More new compat issues uncovered


16:58:20  /home/build-user/swift-tools-protocols/Sources/SKLogging/LoggingScope.swift:92:71: error: cannot convert '@concurrent @Sendable () async throws -> Result' to 'nonisolated(nonsending) () async throws -> Result' because crossing of an isolation boundary requires parameter and result types to conform to 'Sendable' protocol
16:58:20   90 | ) async rethrows -> Result {
16:58:20   91 |   return try await LoggingScope.$_subsystem.withValue(subsystem) {
16:58:20   92 |     return try await LoggingScope.$_scope.withValue(scope, operation: operation)
16:58:20      |                                                                       |- error: cannot convert '@concurrent @Sendable () async throws -> Result' to 'nonisolated(nonsending) () async throws -> Result' because crossing of an isolation boundary requires parameter and result types to conform to 'Sendable' protocol
16:58:20      |                                                                       `- note: type 'Result' does not conform to 'Sendable' protocol
16:58:20   93 |   }
16:58:20   94 | }

ktoso · 2026-01-21T13:28:58Z

swiftlang/swift-tools-protocols#35
swiftlang/swift-testing#1493
@swift-ci please smoke test

ktoso requested a review from a team as a code owner April 11, 2025 07:24

ktoso force-pushed the wip-improve-withTaskCancellationHandler-with-caller-execution branch 2 times, most recently from de601db to 1303d99 Compare April 11, 2025 07:28

ktoso added the concurrency Feature: umbrella label for concurrency language features label Apr 17, 2025

ktoso mentioned this pull request Apr 25, 2025

[TypeChecker] Allow closures to assume nonisolated(nonsending) #80872

Merged

ktoso force-pushed the wip-improve-withTaskCancellationHandler-with-caller-execution branch 2 times, most recently from a583630 to 9ea3293 Compare April 30, 2025 02:13

ktoso commented May 15, 2025

View reviewed changes

stdlib/public/Concurrency/CMakeLists.txt Outdated Show resolved Hide resolved

ktoso force-pushed the wip-improve-withTaskCancellationHandler-with-caller-execution branch 3 times, most recently from 25a963e to be2b2ad Compare July 8, 2025 09:28

ktoso commented Jul 8, 2025

View reviewed changes

stdlib/public/Concurrency/TaskCancellation.swift Outdated Show resolved Hide resolved

ktoso commented Jul 8, 2025

View reviewed changes

stdlib/public/Concurrency/TaskCancellation.swift Show resolved Hide resolved

jamieQ mentioned this pull request Aug 3, 2025

Share algorithm apple/swift-async-algorithms#357

Merged

ktoso force-pushed the wip-improve-withTaskCancellationHandler-with-caller-execution branch from 9f1e063 to 0e5a262 Compare October 14, 2025 09:23

ktoso requested a review from phausler as a code owner October 14, 2025 09:23

ktoso changed the title ~~[Concurrency] improve cancellation handler to not hop and use caller execution context~~ [Concurrency] Correct enqueue behavior in with... functions by adopting nonisolated(nonsending) Oct 14, 2025

ktoso force-pushed the wip-improve-withTaskCancellationHandler-with-caller-execution branch from fcdc60b to 01e0397 Compare October 15, 2025 15:17

alessionossa mentioned this pull request Dec 13, 2025

FIFO not respected in AsyncChannel apple/swift-async-algorithms#356

Open

ktoso force-pushed the wip-improve-withTaskCancellationHandler-with-caller-execution branch from 9a2d287 to 3ceae73 Compare December 15, 2025 07:17

jamieQ reviewed Dec 16, 2025

View reviewed changes

ktoso force-pushed the wip-improve-withTaskCancellationHandler-with-caller-execution branch 4 times, most recently from aa0fb44 to 13e6b40 Compare December 16, 2025 23:15

jamieQ reviewed Dec 19, 2025

View reviewed changes

ktoso force-pushed the wip-improve-withTaskCancellationHandler-with-caller-execution branch from 13e6b40 to 6a221b5 Compare January 14, 2026 01:26

ktoso mentioned this pull request Jan 14, 2026

[Concurrency] Adopt typed throws and nonisolated(nonsending) in withTaskCancellationHandler #85518

Merged

ktoso added 3 commits January 18, 2026 20:22

[Concurrency] Adopt noniso(nonsend) in many concurrency lib funcs

71337dd

This drastically lessens the number of unexpected task enqueues and hops that these APIs cause, and therefore improves performance and even correctness in a few cases (like the withContinuation APIs)

use @export(implementation) instead of AEIC

986f328

avoid @export and use _alwaysEmitIntoClient instead

72ac89e

ktoso force-pushed the wip-improve-withTaskCancellationHandler-with-caller-execution branch from 76e13aa to 72ac89e Compare January 18, 2026 11:23

rebase fixes

782c0e0

ktoso mentioned this pull request Jan 21, 2026

Prepare for potential nonisolated(nonsending) source break in stdlib swiftlang/swift-testing#1493

Closed

ktoso mentioned this pull request Jan 21, 2026

[DNM] Checking what changes may be necessary for stdlib nonisolatednonsending adoption swiftlang/swift-tools-protocols#35

Closed

[Concurrency] Correct enqueue behavior in with... functions by adopting nonisolated(nonsending) #80753

Are you sure you want to change the base?

[Concurrency] Correct enqueue behavior in with... functions by adopting nonisolated(nonsending) #80753

Uh oh!

Conversation

ktoso commented Apr 11, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

ktoso commented Apr 12, 2025

Uh oh!

ktoso commented Apr 30, 2025

Uh oh!

ktoso commented May 15, 2025

Uh oh!

Uh oh!

gottesmm commented Jun 24, 2025

Uh oh!

gottesmm commented Jul 7, 2025

Uh oh!

ktoso commented Jul 8, 2025

Uh oh!

Uh oh!

Uh oh!

ktoso commented Jul 8, 2025

Uh oh!

ktoso commented Jul 8, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

ktoso commented Jul 8, 2025

Uh oh!

ktoso commented Jul 11, 2025

Uh oh!

ktoso commented Oct 15, 2025

Uh oh!

ktoso commented Oct 15, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

ktoso commented Dec 15, 2025

Uh oh!

jamieQ left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

jamieQ Dec 15, 2025

Choose a reason for hiding this comment

Uh oh!

ktoso commented Dec 16, 2025

Uh oh!

jamieQ Dec 19, 2025

Choose a reason for hiding this comment

Uh oh!

ktoso commented Jan 14, 2026

Uh oh!

ktoso commented Jan 14, 2026

Uh oh!

ktoso commented Jan 14, 2026

Uh oh!

ktoso commented Jan 18, 2026

Uh oh!

ktoso commented Jan 18, 2026

Uh oh!

xedin commented Jan 18, 2026

Uh oh!

ktoso commented Jan 20, 2026

Uh oh!

ktoso commented Jan 20, 2026

Uh oh!

ktoso commented Jan 21, 2026

Uh oh!

ktoso commented Jan 21, 2026

Uh oh!

ktoso commented Jan 21, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

ktoso commented Apr 11, 2025 •

edited

Loading

ktoso commented Jul 8, 2025 •

edited

Loading

ktoso commented Oct 15, 2025 •

edited

Loading