Skip to content

CONCEPT Cited by 1 source

Opt-in marker interface

Definition

An opt-in marker interface is an interface (or protocol, or trait, depending on language) whose sole purpose is for an implementor to declare opt-in participation in some cross-cutting framework behaviour. Unlike rich interfaces that prescribe real methods, a marker interface's primary contract is "I implement this → the platform should treat me specially."

Pinterest's PerfImageView / PerfTextView / PerfVideoView are marker interfaces-with-readiness-methods (a hybrid: small method surface, but the primary semantics is the opt-in declaration). A product engineer marks their ImageView as a PerfImageView to tell the platform: "this image is content-critical; count it toward Visually Complete." (Source: sources/2026-04-08-pinterest-performance-for-everyone).

Why opt-in and not auto-detect

Pinterest's platform could, in principle, auto-detect every ImageView in the view tree. Instead it requires the product engineer to explicitly tag each view that should count. Opt-in rather than auto-detect is a deliberate design choice:

  • False-positive avoidance — not every ImageView on screen matters for Visually Complete (loading shimmers, decorative chrome, background imagery, collapsed UI elements). Auto-detection would block Visually Complete on every stray ImageView that happens to be slow, producing never-ready or late-ready signals.
  • Product intent is not observable — the platform can't tell from a view's type whether it carries content-critical meaning. Only the product engineer knows whether this particular hero image is the one the user is waiting for.
  • Tagging is cheap — implementing a marker interface is one import and a few lines; the cost of opt-in per view is minutes.
  • Discoverability — the marker makes the performance-relevant views grep-able and lintable.
  • Java / Kotlin marker interface (Serializable, Cloneable) — classical pure-marker use.
  • Go marker interface (empty interface, type MyMarker interface {}) — same idea.
  • Rust trait — traits serve both prescriptive (fn fmt(&self) -> ...) and marker (Send, Sync, Copy) roles.
  • Swift protocol conformance — protocols can be pure-marker or prescriptive.
  • C# attribute / Java annotation (@Instrumented, @Loggable) — non-interface way to achieve the same opt-in semantics without forcing a type relationship.

Pinterest's hybrid of marker with small readiness method sits between pure-marker (nothing prescribed) and rich-interface (full behaviour contract). The marker half carries the opt-in signal; the method half lets the platform consume per-instance state without reaching into private implementation.

The load-bearing correctness property: tagging accuracy

The whole scheme's correctness depends on product engineers tagging exactly the right views:

  • False negatives (critical view not tagged) → Visually Complete fires before the user-perceivable content is actually visible → metric under-reports latency.
  • False positives (irrelevant view tagged) → Visually Complete blocked on a view the user doesn't care about → metric over-reports latency or never fires.

Production-quality opt-in instrumentation typically adds guardrails:

  • Lint rules — flag common omissions ("this ImageView is at the top of a BaseSurface and is not a PerfImageView").
  • Code review conventions — performance-critical PRs get reviewed by the Perf team.
  • Baseline validation — A/B test Visually Complete timestamps against human-rater video review to catch drift.

Pinterest's post doesn't describe any of these guardrails — opt-in correctness is an open question left to the implementor.

  • Annotations (Java, Kotlin, Python decorators) — a non-type-system way to achieve opt-in; often paired with annotation processors or reflection at runtime.
  • Composition over inheritance — delegates or subscriber objects can carry the opt-in semantic without forcing a type relationship; trades off discoverability for flexibility.
  • Registration APIs — "call perfTracker.register(this) in onCreate" — the imperative analogue; trades static type-system guarantees for runtime flexibility.

Seen in

  • 2026-04-08 Pinterest — Performance for Everyone (sources/2026-04-08-pinterest-performance-for-everyone) — canonical; PerfImageView / PerfTextView / PerfVideoView as hybrid marker-plus-readiness interfaces; product-engineer-tagged opt-in; consumed by BaseSurface view-tree walk.
Last updated · 319 distilled / 1,201 read