[copilot-finds] Bug: WhenAllTask crashes when children complete after fail-fast

[github-actions]

Bug: WhenAllTask crashes when children complete after fail-fast

Branch with fix: copilot-finds/bug/fix-whenall-child-completion-after-fail-fast

A PR could not be created automatically because the repository settings do not allow GitHub Actions to create pull requests. Please create a PR from the branch above.

Problem

WhenAllTask.onChildCompleted() has two bugs that can crash orchestrations using the fan-out/fan-in pattern:

Bug 1: Throw after fail-fast

When a WhenAllTask fails fast on the first child failure, it sets _isComplete = true. If any remaining child task later completes (in the same or subsequent event batch), onChildCompleted() throws "Task is already completed". This exception propagates through the executor and crashes the orchestration with an unexpected error — or worse, overwrites a correctly completed orchestration status when the orchestrator caught the WhenAll failure.

File: packages/durabletask-js/src/task/when-all-task.ts:29-31

Bug 2: Fall-through on last-task-fails

When the failing task happens to be the last child to complete, the fail-fast block sets _isComplete = true but falls through to the result-collection block (this._tasks.map(t => t.getResult())). Since getResult() throws when _exception is set, this crashes the orchestration through an unintended code path, bypassing the generator entirely.

File: packages/durabletask-js/src/task/when-all-task.ts:35-43

Root Cause

1. The onChildCompleted guard used throw instead of return for already-complete tasks. WhenAnyTask correctly uses if (!this.isComplete) — WhenAllTask was inconsistent.
2. Missing return after the fail-fast block allowed fall-through to the result-collection code that calls getResult() on all tasks, including the failed one.

Fix (implemented on branch)

WhenAllTask.onChildCompleted() (2 changes)

• Replace throw new Error("Task is already completed") with return to silently ignore subsequent child completions after fail-fast
• Add return after the fail-fast exception assignment to prevent fall-through to getResult()

RuntimeOrchestrationContext.resume() (1 change)

• Add early return guard when _isComplete is true, preventing attempts to resume a finished generator when subsequent task completion events trigger resume() on an already-completed context

Testing (3 new tests, all passing)

1. "should fail whenAll correctly when the failing task is the last to complete" — Verifies Bug 2
2. "should not crash when additional tasks complete after whenAll fails fast" — Verifies Bug 1
3. "should preserve orchestration result when whenAll failure is caught and other tasks complete" — Verifies the catch-and-handle scenario

All 780 tests pass. Lint clean.

Risk

Low risk. The fix only changes behavior for the already-complete case (which previously threw) and the fall-through case (which was always incorrect). WhenAnyTask already uses the return pattern.