Effects revisited in v19

Root vs View effects.

Matthieu Riegler -

Unlike other signals APIs, the effect hasn't stabilized yet and is still in developer preview. This has a reason as scheduling and reactivity context need some fine tuning based on developer feedback.

Effects are a construct which executes a reactive function whenever the signals it reads are updated. They're declared via the effect() function:

Effect definition

function effect(fn: (cleanupFn: CleanupFn) => void, options?: EffectOptions): EffectRef;

interface EffectOptions {
  injector?: Injector;
  forceRoot?: true;
  manualCleanup?: boolean;
}

Context dependent effects

Effects can be classified into two types: root effects and view effects. The type created is determined by the context in which the effect() function is called. If called within components, a view effect is generated; otherwise, a root effect is generated.

Note that, Angular handles for you which type of effect is creating depending on the calling context. There is no need to explicitly choose between view and root effects. This behavior can be overriden with the forceRoot flag that enforces the creation of a root effect.

Root effects

Root effects are top-level effects within an application and belong to no hierachy and independent from component updates. Effects created in root services for example are root effects

They are useful for operations like:

  • Propagating state changes to other signals (when computeds are not an option).
  • Synchronizing state with the backend or some local storage
  • Rendering not tied to a component (e.g. integration w/ other frameworks)
  • Logging/Debugging

Root effects are scheduled via a macrotask, on each ApplicationRef.tick (and while there are dirty root effects). They are queued in FIFO order: effects that become dirty first will execute first.

One particular consequence of dirty root effects running until the queue is empty is that you might see effects running kind-of synchronously.

  sig = signal(0);

  #myRootEffect = effect(
    () => {
      if (this.sig() < 5) {
        console.log(this.sig());
        this.sig.update((s) => s + 1);
      }
    },
  );

This sample code will log 5 times before running any Change Detection.

Concerning unit test, you can flush the root effects with TestBed.flushEffects().

View effects

View effects are effects within the component hierarchy and execute as part of change detection cycles. Because of this timing, view effects can be used to respond to input signal changes, or to update state used in child components (including creating and destroying child views).

The need for view effects is driven by two major use cases / concerns which are not served by root effects:

  1. Signal inputs.

Input signals are set as part of the change detection process. This timing is important for effects that monitor them. More importantly, required signal inputs can't be read until they received their first value. This means that effects must not be scheduled until the component's inputs have been set.

Crucially, input.required signals are not allowed to be read until they have received their initial value. This means that effects must not be scheduled until the component's inputs have been set.

  1. Effects that affect the component's state or the state of child components.

A major reason to create effects in components is to react to input changes and update component state, either by deriving new values for the component or its children, or by creating or destroying embedded views. For example, the following effect implements a reactive version of @if control flow:

Effect that needs to run ahead of CD

const show = computed(() => !!cond());
let view;
effect(() => {
  if (show()) {
    view = this.vcr.createEmbeddedView(this.childView);
  } else {
    view.destroy();
  }
});

Now imagine this effect runs after change detection. This would have 2 major consequences :

  • If a child view is created, it would have to schedule a new Change detection cycle, which would be inefficient
  • If the cond become false (and consequently show also), the embedded view would get change detected before being destroyed. This will result in a violation of the invariant that the effect is trying to enforce, and the embedded view will be change detected with a null value it was not written to handle.

In other words, effects that affect the rendering of their children must run before those children undergo change detection. Failing to do so can lead to broken invariants and crashes, as well as inefficiency. View effects ensure this guarantee. For this reasons view effect will run during change detection at the begining of each component check.

See this stackblitz example of this issue with v18 effect timing.

Scheduling

View effects are associated with a given node within a template. For example, given the template:

<div tooltipDirective>...</div>
<child-cmp childDirective />

The effects for tooltipDirective would execute when the update pass reached the <div> node, and the effects for <child-cmp> and childDirective would execute when the update pass reached the <child-cmp> node. This is the same mechanism that directive lifecycle hooks and host bindings use today.

View effects should run before host bindings, as they might update state which is read in host bindings. This is a similar timing to ngOnChanges / ngDoCheck. Note that this requires the parent view of the associated node to be refreshed in order to trigger the effect. This is usually the case anyway (in the common case, an input signal from that view was what triggered the effect) but is sub-optimal if the signal triggering the effect had nothing to do with the parent view. We could resolve this inefficiency with some investment, but it may not be worth the complexity.

AfterRenderEffect

With view effects triggering before the components updates & change detection, we still need a reactive primitive that would be triggered once the application is fully rendered.

This is what AfterRenderEffect is built for and it will execute registered effects during specific phases.

  • earlyRead
    • Use this phase to read from the DOM before a subsequent write callback, for example to perform custom layout that the browser doesn't natively support. Prefer the read phase if reading can wait until after the write phase. Never write to the DOM in this phase.
  • write
    • Use this phase to write to the DOM. Never read from the DOM in this phase.
  • mixedReadWrite
    • Use this phase to read from and write to the DOM simultaneously. Never use this phase if it is possible to divide the work among the other phases instead.
  • read
    • Use this phase to read from the DOM. Never write to the DOM in this phase.

afterViewEffect

afterRenderEffect({
  earlyRead: () => ...,
  write: () => ...,
  mixedReadWrite: . ..,
  read: () => ...,
});

Suggestions