Dart Concurrency

Make ownership, ordering, cancellation, and error propagation explicit.

Diagnose first

Identify the producer, consumer, lifecycle owner, ordering requirement, and desired behavior when work becomes obsolete. Reproduce the race or leak before adding synchronization-like machinery.

Decision rules

Use Future for one result and Stream for a sequence over time.
Avoid unawaited work unless independence and error handling are intentional.
Cancel subscriptions and controllers according to the owning object's lifecycle.
Guard UI updates after async gaps with the lifecycle mechanism appropriate to the project.
Prevent stale requests from overwriting newer state through cancellation, tokens, sequencing, or latest-wins logic.
Use isolates for measured CPU-bound work or isolation needs, not ordinary network I/O.
Send transferable values across isolate boundaries and surface failures to the caller.
Preserve stack traces when translating errors.

Do not hide async ownership inside global helpers. Keep retry and timeout policy at a boundary that understands idempotency and user intent.

Verification

Test success, failure, cancellation/disposal, and out-of-order completion. Run static analysis; add a focused stress or fake-time test when timing is central to the bug.

References

Read isolate communication and worker pools when handling heavy CPU computations, port messaging, or background worker threads.
Read stream cancellation and debouncing when managing subscription lifecycles, race conditions, debounce/throttle, or latest-wins semantics.