PureComputed or Computed with deferEvaluation: Choosing the Right Approach for Automatic Disposal in Complex Dependency Chains

Goal

Determine the optimal strategy for derived values that participate in deep dependency graphs while ensuring that subscriptions are cleaned up automatically when no longer needed.

Constraints & Uncertainty

• PureComputed (introduced in Knockout 3.5) automatically disposes when all subscriptions are removed, but the documentation offers no clear guidance on how this behaves when the pureComputed is an intermediate node in a chain of computed observables.
• Computed with deferEvaluation requires explicit disposal and may persist, potentially leaking memory in long‑running applications.
• In nested scenarios, disposing an intermediate pureComputed could silently break downstream computed values, yet a manual disposal of the downstream chain may also be necessary.

Unresolved Decision

When building a view model with multiple layers of derived observables, should developers default to pureComputed for its automatic disposal, or prefer computed with deferEvaluation and manual cleanup to avoid unintended chain termination?

Specific Questions

• Does pureComputed automatically dispose intermediate nodes in a nested chain without breaking downstream dependencies?
• What are the observable side effects on subscription counts when a pureComputed is removed from a chain of computed observables?
• In practice, which approach yields better performance and lower memory usage in large view models with thousands of observable changes?