Migrate into bloc

[chimon2000]

I really love what you are doing here, as I have been exploring how to use bloc w/ Riverpod. I am wondering if it makes sense for this code to live inside of the bloc repository eventually. Tagging @felangel for his thoughts.

[kranfix]

Thanks for appreciate my work. I would like this code to live within the bloc repository (at least the useBloc hook function).

By the moment, flutter_hooks_bloc has the goal to reduce the nested widgets in the tree and flutter_hooks was a suitable tool for this.

I am also sending some PR to flutter_hooks to have a better information in the widget inspector. For example, with flutter_bloc, the widget tree looks like this:

but with the flutter_hooks_bloc it looks shorter:

This modification to the widget tree in the widget inspector will be released tomorrow (v0. 9.0). By the moment, this reduction of the widget tree is available for MultiBlocListener, but I'm working in MultiBlocProvider and MultiRepositoryProvider.

I hope @felangel is interested in these changes that I am making.

[felangel]

Hey all, I think there is definitely some interesting work being done but I'm still not entirely convinced that hooks is the direction we should be moving in. It's something that is being actively discussed and I will continue to monitor and experiment with the APIs as well but I don't think it makes sense to try to introduce hooks into the flutter_bloc package at this time.

The problem that hooks are solving in this case is providing an alternative to the builder pattern which is very prevalent throughout all of flutter. I would love to discuss some concrete use-cases for which hooks are better suited than builders in the context of the bloc library because generally speaking, there should be one bloc per feature so there should be very few cases where you need to nest BlocBuilders and BlocListeners (features should generally not depend on other features). In my experience having many nested BlocBuilders is usually a symptom of blocs that are data-driven or resource-driven and not feature-driven.

Would love to hear everyone's thoughts! Keep up the great work @kranfix and thanks @chimon2000 for starting this conversation! 😄

[chimon2000]

@felangel I agree that hooks are a divisive topic, and I am actually more interested in the non-hooks version that @kranfix is working on vs. the hooks version (check out #12).

I think that the problems that riverpod by itself solves (type safety, simplified state composition, simplified testing) are extremely beneficial.

[felangel]

@chimon2000 yeah I definitely agree and it's something I'm happy to work towards but I am also hesitant about introducing a dependency on riverpod in it's current, experimental state. Eventually once it stabilizes and if the community begins to shift away from provider I think it would make sense but currently provider is still the most stable and widely used solution and I think forcing developers to use riverpod or having multiple DI solutions would be less beneficial. I'm happy to continue this conversation and continue experimenting in the meantime so that we are better prepared to make a shift as needed when everything has stabilized 👍

[chimon2000]

Totally fair @felangel. I expect this will be an open conversation dependent on when riverpod hits 1.0. I know Remi has mentioned the experimental tag will be dropped soon 🤞

[rrousselGit]

I already removed the experimental status actually.
That doesn't mean it will have a 1.0, as that's a different topic. But besides minor changes for future features, it can be considered ""stable""
An article should come at some point during September.

Anyway, I'll be blunt: Do we really need BLoC with Riverpod? Genuinely wondering.
As Cubit == StateNotifier, and there's an equivalent to all the other features (if not, make an issue ☺️)

[chimon2000]

@rrousselGit I think that's a fair question. There are many similarities between the two. I am thinking specifically of users who may already be heavily invested in the bloc ecosystem, but want some of the convenience that riverpod provides.

Even in that case they may find, migrating from provider to riverpod not worth the hassle. I suppose the outcome of felangel/bloc#1429 may affect this as well.

[kranfix]

@rrousselGit cubit has Changes and Bloc has Transitions (and also Changes because it extends Cubit). They allows to monitor the app easily. Outside of that, I don't see any significant difference with StateNotifier.

[rrousselGit]

Transitions are built directly inside Riverpod with ProviderScope.observers

[kranfix]

@rrousselGit The transitions with ProviderScope.observers are no enough useful as Transitions and Changes from bloc library. In bloc, you can have a global or a custom Change/Transition for each cubit/bloc.

Outside of that, I agree with StateNotifier == Cubit, but despite I like riverpod with StateNotifierProvider, I really need Change/Transition and that's the reason why I am working in riverbloc.

[rrousselGit]

In bloc, you can have a global or a custom Change/Transition for each cubit/bloc.

I'd appreciate if you could explain why ProviderObserver is not enough, with a concrete example

New features can be added to Riverpod if there is a need.

[kranfix]

Currently, each observer in ProviderScope.observers has the same execution. To make custom observers, each observer must ask all the time for the provider's runtimeType in order to customize the information. In addition, both Transition and Change give information about a current and previous state, while riverpod gives only the current provider's value.

Then I was thinking in opening an issue/PR with a pair of mixins from StateNotifier:
- mixin StateNotifierTransition<T> on StateNotifier<T>
- mixin StateNotifierChange<T> on StateNotifier<T>

And adding a static StateNotifier.observer similar to Bloc.observer.

[rrousselGit]

Currently, each observer in ProviderScope.observers has the same execution. To make custom observers, each observer must ask all the time for the provider's runtimeType in order to customize the information.

I don't understand what this means.

The previous value could be added to the update event, but I really don't understand what you are trying to do specifically.
Some examples would be helpful.

[tlvenn]

Just stumble on this issue thanks to google, one thing that seems to be missing with the observer on riverpod side as outlined by @kranfix is the current state. There are many cases where it is super useful to have both the current and the new state (time travel, undo - redo, replay, etc...).