[copilot-finds] Bug: resume() while-loop silently swallows exceptions from immediately-failed tasks

[github-actions]

Problem

In RuntimeOrchestrationContext.resume() (packages/durabletask-js/src/worker/runtime-orchestration-context.ts, lines 169-197), the isComplete while-loop does not check whether a yielded task is failed before continuing the loop. When a task is already failed (e.g., a WhenAllTask with a pre-failed child), the loop calls generator.next(this._previousTask._result) instead of generator.throw(this._previousTask._exception). Since _result is undefined on failed tasks, the generator receives undefined instead of the exception, silently losing the error.

Root Cause

The isComplete while-loop (line 170) only checks !this._previousTask.isComplete to decide whether to break. Since failed tasks have isComplete = true, the loop continues iterating and sends the (undefined) result to the generator. The isFailed check exists at the top of resume() (line 148) but is never reached inside the while-loop because the loop handles all isComplete tasks uniformly.

Reproduction Scenario

1. Orchestrator schedules an activity (task1) and a timer
2. Orchestrator yields the timer (not the activity)
3. The activity fails while the timer is pending — task1 is now isComplete=true, isFailed=true
4. The timer fires, and the generator resumes
5. Generator creates whenAll([task1]) — WhenAllTask constructor calls onChildCompleted() for the pre-failed child, making the WhenAllTask immediately failed
6. Generator yields this already-failed WhenAllTask
7. Bug: The while-loop sees isComplete=true, does NOT break, and calls generator.next(undefined) — the failure exception is lost

Proposed Fix

Add a isFailed check inside the while-loop. When an immediately-failed task is detected, break out and delegate to the existing isFailed handling branch via recursive resume() call.

Impact

Severity: High. This bug causes silent loss of activity/sub-orchestration failure exceptions in orchestrators that use patterns like fire-and-forget activity scheduling followed by whenAll. The orchestrator receives undefined instead of the expected exception, leading to incorrect orchestration behavior and potential data loss.