CABI: tighten subtask.cancel behavior again, add more tests - #726
Merged
Merged
Conversation
lukewagner
force-pushed
the
tweak-cancel
branch
from
September 21, 2026 17:36
c52323b to
d634a4a
Compare
dicej
added a commit
to dicej/wasmtime
that referenced
this pull request
Sep 22, 2026
WebAssembly/component-model#726 refined the behavior of `subtask.cancel` to more closely match how async-lowered calls behave, meaning the runtime must resume the cancelling thread specifically as soon as the cancellee suspends or exits. This required some refactoring, and I believe it's a net simplification of how the runtime decides which thread to switch to and when. - Both guest->guest calls and `subtask.cancel` use a new `SuspendReason::YieldingToSubtask` variant which replaces `WaitingForGuestSubtask` and means "I'm yielding specifically to the specified (via `ConcurrentState::switch_item`) subtask and expect to be resumed when that subtask suspends or exits." - I've removed `GuestTask::switch_item` in favor of a new `ConcurrentState::next_switch_item` field. This addresses a case exercised by the new `cancel-resumed-callback-switch.wast` test where a subtask, while being cancelled, explicitly resumes a thread belonging to a different task, which then suspends, in which case we need to resume the cancelling thread. This new `next_switch_item` field is saved and restored as needed so that a stack of cancels and guest->guest calls can each yield to their respective subtasks and return to their respective callers/cancellers without clobbering each other. - The new `cancel-targeted-resume.wast` test revealed that Wasmtime was out of compliance regarding `CALLBACK_CODE_YIELD`; when a task is cancelled while it is stacklessly yielding, we must promote it so the `cancel` event can be delivered deterministically. I've addressed this by defining a new `WakeOnCancel` enum type which can represent a thread which is waiting cancellably, yielding cancellably, or neither. Since the aforementioned spec PR has not yet been merged as of this writing, I haven't updated the `tests/component-model` submodule yet, but I've verified locally that the new tests pass.
dicej
added a commit
to dicej/wasmtime
that referenced
this pull request
Sep 22, 2026
WebAssembly/component-model#726 refined the behavior of `subtask.cancel` to more closely match how async-lowered calls behave, meaning the runtime must resume the cancelling thread specifically as soon as the cancellee suspends or exits. This required some refactoring, and I believe it's a net simplification of how the runtime decides which thread to switch to and when. - Both guest->guest calls and `subtask.cancel` use a new `SuspendReason::YieldingToSubtask` variant which replaces `WaitingForGuestSubtask` and means "I'm yielding specifically to the specified (via `ConcurrentState::switch_item`) subtask and expect to be resumed when that subtask suspends or exits." - I've removed `GuestTask::switch_item` in favor of a new `ConcurrentState::next_switch_item` field. This addresses a case exercised by the new `cancel-resumed-callback-switch.wast` test where a subtask, while being cancelled, explicitly resumes a thread belonging to a different task, which then suspends, in which case we need to resume the cancelling thread. This new `next_switch_item` field is saved and restored as needed so that a stack of cancels and guest->guest calls can each yield to their respective subtasks and return to their respective callers/cancellers without clobbering each other. - The new `cancel-targeted-resume.wast` test revealed that Wasmtime was out of compliance regarding `CALLBACK_CODE_YIELD`; when a task is cancelled while it is stacklessly yielding, we must promote it so the `cancel` event can be delivered deterministically. I've addressed this by defining a new `WakeOnCancel` enum type which can represent a thread which is waiting cancellably, yielding cancellably, or neither. Since the aforementioned spec PR has not yet been merged as of this writing, I haven't updated the `tests/component-model` submodule yet, but I've verified locally that the new tests pass.
dicej
approved these changes
Sep 23, 2026
pull Bot
pushed a commit
to langyo/wasmtime
that referenced
this pull request
Sep 23, 2026
…tecodealliance#14382) WebAssembly/component-model#726 refined the behavior of `subtask.cancel` to more closely match how async-lowered calls behave, meaning the runtime must resume the cancelling thread specifically as soon as the cancellee suspends or exits. This required some refactoring, and I believe it's a net simplification of how the runtime decides which thread to switch to and when. - Both guest->guest calls and `subtask.cancel` use a new `SuspendReason::YieldingToSubtask` variant which replaces `WaitingForGuestSubtask` and means "I'm yielding specifically to the specified (via `ConcurrentState::switch_item`) subtask and expect to be resumed when that subtask suspends or exits." - I've removed `GuestTask::switch_item` in favor of a new `ConcurrentState::next_switch_item` field. This addresses a case exercised by the new `cancel-resumed-callback-switch.wast` test where a subtask, while being cancelled, explicitly resumes a thread belonging to a different task, which then suspends, in which case we need to resume the cancelling thread. This new `next_switch_item` field is saved and restored as needed so that a stack of cancels and guest->guest calls can each yield to their respective subtasks and return to their respective callers/cancellers without clobbering each other. - The new `cancel-targeted-resume.wast` test revealed that Wasmtime was out of compliance regarding `CALLBACK_CODE_YIELD`; when a task is cancelled while it is stacklessly yielding, we must promote it so the `cancel` event can be delivered deterministically. I've addressed this by defining a new `WakeOnCancel` enum type which can represent a thread which is waiting cancellably, yielding cancellably, or neither. Since the aforementioned spec PR has not yet been merged as of this writing, I haven't updated the `tests/component-model` submodule yet, but I've verified locally that the new tests pass.
lukewagner
force-pushed
the
tweak-cancel
branch
from
September 23, 2026 15:02
d634a4a to
4aaac9d
Compare
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
This is a further refinement of #723 based on some discussion with @dicej. To keep things simple, testable and symmetric with how async lower works,
subtask.cancelis returned to it's pre-#716 behavior where the only thread resumed duringsubtask.cancelis an implicitcallbackthread that's waiting in its event loop (which is the only way to receiveTASK_CANCELLEDatm) and then control flow deterministically transfers back as soon as thecallbackexits or blocks. This PR also adds a bunch of tests to confirm that nothing else is run instead of or after, and that, while running, coop-thread-switching is allowed.