Clearing cache and reacting to state disposal

So far, we've seen how to create/update some state. But we have yet to talk about when state destruction.

Riverpod offers various ways to interact with state disposal. This ranges from delaying the disposal of state to reacting to destruction.

When is state destroyed and how to change this?

When using code-generation, by default, the state is destroyed when the provider stops being listened.
This happens when a listener has no active listener for a full frame. When that happens, the state is destroyed.

This behavior can be opted out by using keepAlive: true.
Doing so will prevent the state from getting destroyed when all listeners are removed.

// We can specify "keepAlive" in the annotation to disable
// the automatic state destruction
@Riverpod(keepAlive: true)
int example(ExampleRef ref) {
  return 0;
}

remarque

Enabling/disabling automatic disposal has no impact on whether or not the state is destroyed when the provider is recomputed.
The state will always be destroyed when the provider is recomputed.

attention

When providers receive parameters, it is recommended to enable the automatic disposal. That is because otherwise, one state per parameter combination will be created, which can lead to memory leaks.

Reacting to state disposal

In Riverpod, there are a few built-in ways for state to be destroyed:

• The provider is no longer used and is in "auto dispose" mode (more on that later). In this case, all associated state with the provider is destroyed.
• The provider is recomputed, such as with ref.watch. In that case, the previous state is disposed, and a new state is created.

In both cases. you may want to execute some logic when that happens.
This can be achieved with ref.onDispose. This methods enables registering a listener to whenever the state is destroyed.

For example, you may want use it to close any active StreamController:

@riverpod
Stream<int> example(ExampleRef ref) {
  final controller = StreamController<int>();

  // When the state is destroyed, we close the StreamController.
  ref.onDispose(controller.close);

  // TO-DO: Push some values in the StreamController
  return controller.stream;
}

attention

The callback of ref.onDispose must not trigger side-effects. Modifying providers inside onDispose could lead to unexpected behavior.

info

There are other useful life-cycles such as:

• ref.onCancel which is called when the last listener of a provider is removed.
• ref.onResume which is called when a new listener is added after onCancel was invoked.

info

You can call ref.onDispose as many times as you wish. Feel free to call it once per disposable object in your provider. This practice makes it easier to spot when we forgot to dispose of something.

Manually forcing the destruction of a provider, using ref.invalidate

Sometimes, you may want to force the destruction of a provider. This can be done by using ref.invalidate, which can be called from another provider or from a widget.

Using ref.invalidate will destroy the current provider state. There are then two possible outcomes:

• If the provider is listened, a new state will be created.
• If the provider is not listened, the provider will be fully destroyed.

class MyWidget extends ConsumerWidget {
  @override
  Widget build(BuildContext context, WidgetRef ref) {
    return ElevatedButton(
      onPressed: () {
        // On click, destroy the provider
        ref.invalidate(someProvider);
      },
      child: const Text('dispose a provider'),
    );
  }
}

info

It is possible for providers to invalidate themselves by using ref.invalidateSelf. Although in this case, this will always result in a new state being created.

Fine-tuned disposal with ref.keepAlive

As mentioned above, when automatic disposal is enabled, the state is destroyed when the provider has no listeners for a full frame.

But you may want to have more control over this behavior. For instance, you may want to keep the state of successful network requests, but not cache failed requests.

This can be achieved with ref.keepAlive, after enabling automatic disposal. Using it, you can decide when the state stops being automatically disposed.

@riverpod
Future<String> example(ExampleRef ref) async {
  final response = await http.get(Uri.parse('https://example.com'));
  // We keep the provider alive only after the request has successfully completed.
  // If the request failed (and threw), then when the provider stops being
  // listened, the state will be destroyed.
  ref.keepAlive();

  // We can use the `link` to restore the auto-dispose behavior with:
  // link.close();

  return response.body;
}

remarque

If the provider is recomputed, automatic disposal will be re-enabled.

It is also possible to use the return value of ref.keepAlive to revert to automatic disposal.

Example: keeping state alive for a specific amount of time

Currently, Riverpod does not offer a built-in way to keep state alive for a specific amount of time.
But implementing such a feature is easy and reusable with the tools we've seen so far.

By using a Timer + ref.keepAlive, we can keep the state alive for a specific amount of time. To make this logic reusable, we could implement it in an extension method:

extension CacheForExtension on AutoDisposeRef<Object?> {
  /// Keeps the provider alive for [duration].
  void cacheFor(Duration duration) {
    // Immediately prevent the state from getting destroyed.
    final link = keepAlive();
    // After duration has elapsed, we re-enable automatic disposal.
    final timer = Timer(duration, link.close);

    // Optional: when the provider is recomputed (such as with ref.watch),
    // we cancel the pending timer.
    onDispose(timer.cancel);
  }
}

Then, we can use it like so:

@riverpod
Future<Object> example(ExampleRef ref) async {
  /// Keeps the state alive for 5 minutes
  ref.cacheFor(const Duration(minutes: 5));

  return http.get(Uri.https('example.com'));
}

This logic can be tweaked to fit your needs. For instance.
For example you could use ref.onCancel/ref.onResume to destroy the state only if a provider hasn't been listened for a specific amount of time.