Released Episode 1, discussion: Architecture of modern Android apps with Hannes Dorfmann

[artem-zinnatullin]

Yay! Feedback is really appreciated :smile_cat:

[yshrsmz]

thanks for the great podcast! could you share the repository which Hannes said is a good mvvm example?

[sockeqwe]

Here you go: https://github.com/rheinfabrik/android-mvvm-example

As already said, it's build with RxJava instead of android data-binding engine, but it should be possible to understand the "big picture" without having a good rx java knowledge.

Btw. I have sent a pull request with the missing links to the show note

[artem-zinnatullin]

Jeez, I forgot to add a bunch of links including links to Hannes profiles and projects (it was 7 am..), will do as soon as possible!

@yshrsmz thank you for the feedback :)

[atetc]

Good job, don`t stop.

[AdamCopperfield]

@artem-zinnatullin how do you keep background tasks execution during orientation changes without saving presenter?

[artem-zinnatullin]

@AdamCopperfield usually apps I write has almost full offline support so they have kind of persistant queue for operations that need to be done, even if they were started offline, + local db or files for storing content. So in case of orientation change I just need to save id of the operation or content I need to display.

But sometimes when you need to do request without using such queue I'd use replay() operator from RxJava and persist the Observable with or without Dagger via retained Fragment or Activity.getLastNonConfigurationInstance(), if you don't use Rx, you can apply similar technique for any kind of callback listeners.

@atetc thanks :)

[Dmitry-Borodin]

Great talk about modern topic, don't stop.

[fabioCollini]

Great podcast! I am working on an MVVM implementation and I think this pattern is great! You are right, using it you move all the logic from the Activity to the ViewModel but I don't agree that the ViewModel contains spaghetti code. It's pure Java code, so it's easy to write JUnit test and refactor it. And you can create custom attributes (thanks to BindingAdapter annotation) to avoid complex UI code in xml or in ViewModel. The framework I am working on is this: https://github.com/fabioCollini/mv2m, would be great to hear your feedback! :)

[panzerdev]

So after seeing that @artem-zinnatullin is using locals db or some persistence anyway in his apps, I can understand why things might work for him.

1. But if you introduce a persistent layer you have to be sure, that the data you are persisting is even possible to be persisted. So if you decide to just live in memory, because you can never be sure if the content you're asking for, is even valid anymore on the backend side, you have to save the state you gathered in onSaveInstanceState to not have to do everything all over again. Well technically you could think of all kind of ways to do that, but this a quite convinient way for multiple reasons.
2. Also this single Activity for all the things idea seems just wrong to me. Why? Well Android is based on Intents and you can't for example create an Activity by just calling the Constructor. So the Whole idea is to have components which can be started with a certain set of values of from certain events like Broadcasts. For this the Activity will be started by the system if needed. When you now have a complex Activity and you try to handle all the things in there is will get complicated really really quick no matter how many immuatable or statless objects you have. Because your user has a state in mind and excptects certain behaviours. But I think this topic is to big for one comment. I just think you can easily use multiple Activities in a very good way.
3. I liked your discussion overall and I'll continue to listen. But I felt like you really fast got into the realm of more code is better. More of this, more of that, more abstraction and more more more. I personally think that you should always try to keep the code as simple and as slim as you can. Complexety will be added by time to it anyway so goldplating it to beging with or introducing abstraction which you will never use is just wasted time and makes things way harder to maintain. The less code the better IMHO. Wrting code is easy, getting rid of it later is hard.

[Gi-lo]

First of all, thanks again for your mention and the kind words :) Let me quickly go through a few points both of you made:

• No event features on the language level: This actually was a huge problem we had compared to our iOS colleagues. Where they can use KVC, we have to do it manually. This leads to some boilerplate code as also seen in my example project.
• Code in XML: I would never ever aim to shift code, be it simple binding or formatting of strings, from java to xml. As you said this makes things more unpredictable and testable. That's also one reason I do not use DataBinding engines.
• Lifecycle in ViewModels: This should never be the case and defeats the view model pattern.

Now on to the most important aspect you mentioned:

MVVM does not scale well.

I can prove you wrong here :D When used correctly ViewModels can be used for big to huge applications. An example can be seen in BVB and trivago. Both application heavily use the Model View ViewModel pattern. The main thing to keep in mind is, that ViewModels are not the Silver Bullet to everything. They won't be fixing rotation issues, your app's communication flow etc. They simply are an additional layer between your view and some data. With that in mind you can build small, composable ViewModels for your application.

[sockeqwe]

@fabioCollini That's a nice library. I just have some nitpicking feedback: What I've meant with spaghetti code is the lack of separation of concerns. I think you can see a good example here:

public class NoteViewModel extends ViewModel<String, NoteModel> {

    private final Executor backgroundExecutor;
    private final Executor uiExecutor;
    private NoteLoader noteLoader;
    private NoteSaver noteSaver;
    private MessageManager messageManager;
    public final ObservableBoolean loading = new ObservableBoolean();
    public final ObservableBoolean sending = new ObservableBoolean();

    public NoteViewModel(Executor backgroundExecutor, Executor uiExecutor, NoteLoader noteLoader, NoteSaver noteSaver, MessageManager messageManager) {
        this.backgroundExecutor = backgroundExecutor;
        this.uiExecutor = uiExecutor;
        this.noteLoader = noteLoader;
        this.noteSaver = noteSaver;
        this.messageManager = messageManager;
    }

    @NonNull @Override public NoteModel createModel() {
        return new NoteModel();
    }

    @Override public void resume() {
        if (!loading.get() && !model.isLoaded() && getArgument() != null) {
            reloadData();
        }
    }

    public void reloadData() {
        loading.set(true);
        backgroundExecutor.execute(new Runnable() {
            @Override public void run() {
                executeServerCall();
            }
        });
    }

    private void executeServerCall() {
        try {
            final Note note = noteLoader.load(getArgument());
            uiExecutor.execute(new Runnable() {
                @Override public void run() {
                    model.update(note);
                    loading.set(false);
                }
            });
        } catch (Exception e) {
            uiExecutor.execute(new Runnable() {
                @Override public void run() {
                    model.getError().set(true);
                    loading.set(false);
                }
            });
        }
    }

    public void save() {
        boolean titleValid = checkMandatory(model.getTitle(), model.getTitleError());
        boolean textValid = checkMandatory(model.getText(), model.getTextError());
        if (titleValid && textValid) {
            sending.set(true);
            backgroundExecutor.execute(new Runnable() {
                @Override public void run() {
                    try {
                        Note note = new Note(null, model.getTitle().get(), model.getText().get());
                        String noteId = model.getNoteId();
                        if (noteId == null) {
                            noteId = noteSaver.createNewNote(note).getObjectId();
                            model.setNoteId(noteId);
                        } else {
                            noteSaver.save(noteId, note);
                        }
                        hideSendProgressAndShoMessage(R.string.note_saved);
                    } catch (RetrofitError e) {
                        hideSendProgressAndShoMessage(R.string.error_saving_note);
                    }
                }
            });
        }
    }

    private void hideSendProgressAndShoMessage(final int message) {
        uiExecutor.execute(new Runnable() {
            @Override public void run() {
                messageManager.showMessage(activityHolder, message);
                sending.set(false);
            }
        });
    }

    private boolean checkMandatory(ObservableString bindableString, ObservableInt error) {
        boolean empty = bindableString.isEmpty();
        error.set(empty ? R.string.mandatory_field : 0);
        return !empty;
    }

    @Override public ActivityResult onBackPressed() {
        return new ActivityResult(true, new Note(model.getNoteId(), model.getTitle().get(), model.getText().get()));
    }
}

This NoteViewModelis huge and has a lot of responsibilities: Licecycle, Navigation (onBackPressed()), than it interacts with more than one "business logic" like MessageManager, NoteLoader, NoteServer. You also change the state That's what I mean when I'm talking about shifting spaghetti code from Activity to ViewModel. You might could split this responsibilities into more decoupled components (like commands, controller, etc.). Keep up your good work!

@Gi-lo Regarding scaling, you are right MVVM can scale and you have proved that. I should have said that I, personally, am not able to scale it when working with XML bindings. It didn't worked good for me on Windows Phone and also with android data binding engine it's a little bit ugly. IMHO, it works for quickly building bring some two way data binding small UI things together like EditText. That is my definition of a MVVM (just for a small UI element). But having one ViewModel managing i.e. RecyclerView and ProgressBar is from my point of view not ideal. MVVMC makes that slightly better and allows sharing "model" between ViewModels. But overall, YES, I have to agree MVVM can scale

@panzerdev Yes, usually I prefer having multiple activities. As you have already said, Intents are fundamental in Android especially for deep linking. Actually, that is one things that I wish fragments would support somehow out of the box. However, we can build also a decoupled component that is responsible for navigation and deep link resolving. The nice thing of one activity is that you can animate your views much smoother as the user navigates through your app from "screen" to "screen", i.e. TransitionManager works like a charm with almost zero code. Sure we could do that with ActivityTransitions but you will notice (especially on low end device) a delay when starting the new Activity. Nevertheless, usually i also prefer having multiple Activities over one single Activity with lot of fragments.

[artem-zinnatullin]

@panzerdev

1. For sure if you don't have persistence you probably need to use savedInstanceState
2. One Activity significantly simplifies animations between screens. Another reason is that managing complex back stack of Activities is no easier than custom back stack of Views, etc.
3. Well, yeah, I see your points, but we didn't say that adding more code is good, we said that dividing code to simple components with single responsibility simplifies whole project :)

Thank you for feedback!

// Interesting discussions, btw, same story as @sockeqwe, not sure I'll be able to scale MVVM project on Android.

[fabioCollini]

@sockeqwe you are right, that class is a mess :( But one reason is that the demo project doesn't use RxJava to keep it simple. Using RxJava the same class can be written in a better way:

public class NoteViewModel extends RxViewModel<String, NoteModel> {

    private NoteLoader noteLoader;

    private NoteSaver noteSaver;

    private MessageManager messageManager;

    public final ObservableBoolean loading = new ObservableBoolean();

    public final ObservableBoolean sending = new ObservableBoolean();

    public NoteViewModel(SchedulerManager schedulerManager, NoteLoader noteLoader, NoteSaver noteSaver, MessageManager messageManager) {
        super(schedulerManager);
        this.noteLoader = noteLoader;
        this.noteSaver = noteSaver;
        this.messageManager = messageManager;
    }

    @NonNull @Override public NoteModel createModel() {
        return new NoteModel();
    }

    @Override public void resume() {
        if (!loading.get() && !model.isLoaded() && getArgument() != null) {
            reloadData();
        }
    }

    public void reloadData() {
        subscribe(
                loading::set,
                noteLoader.load(getArgument()),
                model::update,
                t -> model.getError().set(true)
        );
    }

    public void save() {
        boolean titleValid = ValidationUtils.checkMandatory(model.getTitle(), model.getTitleError());
        boolean textValid = ValidationUtils.checkMandatory(model.getText(), model.getTextError());
        if (titleValid && textValid) {
            subscribe(
                    sending::set,
                    noteSaver.save(model.getNoteId(), model.getTitle().get(), model.getText().get()),
                    note -> messageManager.showMessage(activityHolder, R.string.note_saved),
                    t -> messageManager.showMessage(activityHolder, R.string.error_saving_note)
            );
        }
    }

    @Override public ActivityResult onBackPressed() {
        return new ActivityResult(true, new Note(model.getNoteId(), model.getTitle().get(), model.getText().get()));
    }
}

Maybe it continues to have too many responsibilities but it's not difficult to refactor it to delegate to other objects (for example something like a Use Case in clean architecture). How do you solve this problem in MVP? I think that MVP with passive view and this version of MVVM (I am not sure this is a real MVVM implementation) are very similar; the main difference is that in a presenter you call a method on the view to update it while in the View Model you update the model and then the data binding framework will accordingly update the view.

[Dmitry-Borodin]

Please, turn off skipping silence in next records. Subsecond gaps makes it difficult to understand.

[artem-zinnatullin]

Yeah, I understand, actually episode was about 1h 20m and we discussed some other things but decided to make it shorter and my audio editing skills are not good enough yet :)

But, at the same time episode was very informative, I hope, haha :)

On Tue, 9 Feb 2016, 23:16 Dmitry notifications@github.com wrote:

Please, turn off skipping silence in next records. Subsecond gaps makes it difficult to understand.

— Reply to this email directly or view it on GitHub https://github.com/artem-zinnatullin/TheContext-Podcast/issues/1#issuecomment-182044263 .

@artem_zin

[Gi-lo]

@artem-zinnatullin For the first Podcast it was quite good and understandable. Another thing: I think you should open an issue for topic requests.

[artem-zinnatullin]

@Gi-lo definitely will create an issue for suggestions for the next episode soon!

[yshrsmz]

thanks for the updates!(sorry to be late

I have some questions. You said sometimes you have several presenters in one activity:

1. how do you communicate between these presenters? through activity/fragment and treat as usual view's callback?
2. how do you avoid calling same api request from different presenters at the same time? Say, HeaderPresenter and ListPresenter have to obtain data from same api at its startup time, and if each presenters call api separately, two exactly same requests are sent to a remote server. how do you solve this situation?

[sockeqwe]

You said sometimes you have several presenters in one activity

Yes, take gmail for example:

1. What kind of communication do you have? My presenters doesn't communicate with each other (and I don't know what they should say to each other :smile: ) . I guess you mean something like when a mail from MailView has been marked as favorite the MailPresenter have to inform the InboxPresenter so that the InboxView would display the mail as "favorite", right? In that case, it doesn't is up to the presenters to talk to each other. If you have look on the "model layer" of MailPresenter and InboxPresenter you will see that both have the same "model" MailProvider. So both presenter should observe the MailProvider for changes. If MailPresenter updates a mail (mark as favorite) thenMailPresenter will call MailProvider.markMailAsFavorite(mail). Since InboxPresenter is observing the MailProvider as well, MailProvider have to notify all its observers incl. InboxPresenter and then the InboxPresenter will update the InboxView. So there is no direct communication between MailPresenter and InboxPresenter: The "communication" is

   MailPresenter --markAsStarred()--> MailProvider --updates--> InboxPresenter

   So basically this is the observer pattern. People tend to use an EventBus to communicate such "mail marked as starred" Events between MailPresenter and InboxPresenter, but this is just because they are too lazy to implement the observer pattern on MailProvider correctly.

   Is that what you mean with "presenter communication"?

2. That depends a little bit on your implementation i.e. many http clients like are smart enough to detect that an api call to the same url are going to be executed twice an only execute it once but inform both "callbacks" afterwards (batched http calls). Also, if there is a significant delay between both request so that the requests cannot be batched together, then usually the later request will hit http client's cache. If this is not enough, then you have to implement something both can observe, i.e. both HeaderPresenter and ListPresenter talk to the same proxy instance as "model layer". Then it's the responsibility of the proxy to ensure that both get the same result at once (similar to http client batched requests). With RxJava you could use share() operator. Both Presenter subscribe to this "shared" observable. By doing so its ensured that both get the same result at the same time in their onNext().

Hope that answers your question.

[sockeqwe]

btw. as some people asked me about PresentationModel in combination with RecyclerView (I have mentioned that at the end where we have talked about testing), I have written those things down in a blog post: http://hannesdorfmann.com/android/adapter-commands

[Dmitry-Borodin]

@sockeqwe with gmail, if MailView button causing change a fragment, which should be done from activity or something. How will you do this operation? Activity scoped Navigator? What's your opinion in general: activity scoped presenters vs singleton scoped (easier to maintain view state in case of orientation change)

[sockeqwe]

Damn, I had navigation on list of topics to talk about during the podcast, but I forgot about that and we ran out of time anyway @artem-zinnatullin

I don't think that Navigation is the responsibility of the Presenter. I think Navigation should be part of the "View" layer. That doesn't mean that we hardcode and couple everything in the View i.e. Fragment. What I typically have a Navigator interface / class.

So let's say there is the GmailActivity that displays InboxFragment (InboxView) and MailFragment (MailView). So the Activity is just a "container" and therefore doesn't need a presenter. So now let's say that clicking on a "Mail Item" from InboxFragment should display the "mail details" in MailFragment (or replace the old one if there is already one).

I have some kind of Navigation component (let's call this Navigator) in InboxFragment (i.e. injected by dagger and scoped to GmaiActivity). So when clicking on a Mail Item and want to display this mail then I would call in Fragment (not in Presenter) Navigator.showMail(mailID) (please note that i don't name it showMailFragment(mailId), we don't want to specify some implementation details). So Navigator.showMail(mailId) is now responsible to bring up MailFragment. So Navigator could internally have knowledge of FragmentManager and would start and commit the fragment transaction (Navigator will be Activity scoped with dagger).

i.e.

class TabletNavigator implements Navigator {

   private Activity activity;

   public TabletNavigator(Activity activity){
       this.activity = activity;
   }

    @Override
     public void showMail(int mailId){
         MailFragment fragment = ...;
         activity.getSupportFragmentManager()
                    .beginTransaction()
                   .replace(R.id.detailsContainer, fragment)
                   .commit();
     }
    ...
}

and then I could also have a PhoneNavigator that starts an Intent to bring up an activity that is just a container for MailFragment:

class TabletNavigator implements Navigator {

   private Activity activity;

   public TabletNavigator(Activity activity){
       this.activity = activity;
   }

    @Override
     public void showMail(int mailId){
        Intent i = new Intent(activity, MailDetailsActivity.class);
        activity.startActivity(i);
     }
    ...
}

And then in the InboxFragment I would just inject the Navigator via dagger and depending if the user uses a Tablet or Phone:

class InboxFragment extends Fragment {

   @Inject Navigator navigator;

     void onMailClicked(Mail mail){
          navigator.showMail(mail.getId());
    } 

I also think that this concept of Navigator plays quite good with shared Element transitions. We could have a method like Navigator.showMail(int mailId, View sharedElement) which then will setup the shared FragmentTransition or ActivityTransition.

Probably @artem-zinnatullin is doing that entirely different or have found a better way.

activity scoped presenters vs singleton scoped

Not sure what exactly do you mean with scoped presenters ?

Scoping depends on the complexity of your app. i.e. if I have one single Retrofit object in my whole app than I do singleton scoped. I would say that I start singleton scoped and most of my apps are fine with that. If during development I found out that some components are not always needed or lifecycle bounded (i.e. Navigator will be activity scoped, because it has a hard reference to the activity) then I will begin to create smaller scoped dagger components. Typically I have businesslogic singleton scoped (app wide) and Navigation activity scoped.

[Dmitry-Borodin]

Thank you for clarification navigation. About presenter's scope. Mosby used per activity presenter, instead of singleton scoped presenter. Why? It is easier to maintain view state in singleton, and there is no problems, if singleton doesn't know anything of view internals.

[sockeqwe]

First, please note that Mosby is just some kind of scaffold and you can plug in custom behaviour very easily by providing a custom delegate (see http://hannesdorfmann.com/mosby/viewstate/ scroll down until "Delegation chapter"), so you could add your custom "singleton presenter delegate" very easily.

So Mosby's "default behaviour" is to have presenters that live as long as the view lives (and I don't see any reason why the presenter should live longer than the corresponding view, presenter will be garbage collected when the view gets garbage collected). However, this "default behaviour" is smart enough to determine if the view will be destroyed permanently\ (i.e. user has pressed back button) or has been destroyed temporarly because of a screen orientation changes and will be recreated afterwards. With the later one, the presenter can survive screen orientation changes and the "landscape view" gets simply attached / detached from the same presenter instance as "portrait view".

I believe that this "default behaviour" makes sense for 95% percent of an apps use case.

So what about "singletons presenters". Well, when will they garbage collected? Never. What if you have a view that is uses multiple times, i.e. You have Fragment (View) and use this Fragment in different Activities. Will they all share the same singleton presenter instance? Does that makes sense? Or what if you reuse the same Fragment in a ViewPager?

I don't see any advantage of implementing a presenter as "singleton" for this reasons. Have I missed something? I would like to hear what the advantages of "singleton" presenters are compared to "Mosby's" default behaviour.

[b1uebyte]

Great start, guys!

How do you organize package tree with mvp?

[sockeqwe]

@b1uebyte usually I group my classes by feature like here

Basically I group every screen (or subscreen like a fragment or a custom view) in is own package containing all the related classes except model classes which live in is own package like this (MVP example):

• com.example
  • mainmenu
  • MainMenuView (MVP View interface)
  • MainMenuFragment (implements MainMenuView)
  • MainMenuPresenter
  • MainMenuAdapter (if recyclerview is used)
  • model
  • ... (everything from "model layer in MVP" like pojo classes, services etc.)

Typically I have a java gradle module for the "model" layer. So I move the "model package" from above into is own java gradle module.

[artem-zinnatullin]

Probably @artem-zinnatullin is doing that entirely different or have found a better way.

Router is still kind of "idk" thing, I tried several approaches and hadn't found clear solution yet, last one was just a class that has methods like openFeedbackScreen() and only presenter knows about it.

How do you organize package tree with mvp?

Here I'm opposite to @sockeqwe and prefer packaging by components

• com.app
  • models
  • ui
  • presenters
  • views

[PrathameshKesarkar]

This is a very cool Podcast keep them coming found you guys on Android Dev Digest

@sockeqwe can we apply similar pattern like Navigator for NavigationDrawer i.e. when user selects different item from the drawer .

[PrathameshKesarkar]

@sockeqwe also do you have any good example of Repository Pattern

[sockeqwe]

So I started to answer here on github your question about the Repository Pattern. It turned out that my answer is getting too long, so I decided to move my answer in a dedicated blog post: http://hannesdorfmann.com/android/evolution-of-the-repository-pattern

A good example (also mentioned in the blog post) can be found here: https://github.com/android10/Android-CleanArchitecture

[PrathameshKesarkar]

@sockeqwe do you use navigator pattern like thing for Navigation Drawer

[sockeqwe]

I don't think so because I don't even know what the Navigator Pattern is 😄

I mean, I do have classes called Navigator, but I'm not sure if this is the Navigator Pattern you are talking about.

I.e. I have a interface Navigator for a imaginary News Magazine app:

interface Navigator {
  void showNewsList();
  void showNewsDetails(int newsId);
}

The I would have a PhoneNavigator:

class PhoneNavigator implements Navigator {
     private Activity activity;

     @Override
      void showNewsDetails(int newsId){
          Intent intent = new Intent( activity, NewsDetailsActivity.class);
          intent.putExtra("NewsId", newsId);
          activity.startActivity(intent);
      }

    ...
}

NewsDetailsActivity basically hosts a NewsDetailsFragment.

Then I also have a TabletNavigator

class TabletNavigator implements Navigator {
     private Activity activity;

     @Override
      void showNewsDetails(int newsId){
          activity.getFragmentManager()
                      .beginTransaction()
                      .replace(R.id.container, NewsDetailsFragment.newInstance(newsId))
                      .commit();
      }

    ...
}

The difference of both navigators is how they display "navigate" in the UI. PhoneNavigator starts a new Activity to display a list of news while TabletNavigator switches a Fragment in a Master / Detail view in the same Activity.

And then in my activities / fragments I instantiate either a PhoneNavigator or TabletNavigator (or inject one via dagger) depending on if the device is a Phone or a Tablet. By clicking on a view somewhere in the UI I will then call navigator.showNewsDetails(newsId) and I don't care about the implementation details (like starting Activity or replacing a Fragment). It's only about hiding implementation details and separation of concerns / single responsibility.

Is that what you call the "Navigator Pattern"?

[PrathameshKesarkar]

Thanks that the same thing I was talking about @sockeqwe . I have one more question for you in the podcast you said that we can have more than one Presenter in our Activity for example in this image the Inbox View has two presenter the Fragment presenter and SearchViewPresenter so my biggest concern is how will I attach View to the Presenter cause in my app I have a function setView(ViewInterface interface) which attach my view to presenter but how can I do same with some other View(Widget)

[sockeqwe]

Use View.onAttachedToWindow() and View.onDetachFromWindow()

[nomisRev]

@sockeqwe I've a question concerning your comment of using AdapterCommands or DiffUtil in the presenter. (I love the idea but I'm having an issue implementing it. Doing it in the adapter on the main thread is causing issues ~10ms calculation) I've read your blogposts about MVI, and I absolutely loved them!

I'm currently implementing it in a hobby project and I am using a BehaviorRelay to cache the latest "ViewModelState".

My questions is if I reduced the state of my previous List<Data> and my new List<Data> with a resulting List<AdapterCommands> I don't want that List<AdapterCommands> on a rotation change because in case I'm using a Fragment/Controller my adapter would be retained (same lifecycle as Fragment/Controller). The adapter commands would make my adapter out of sync.

Do you have a good solution for this? I can currently only think of dirty onces.

Awesome podcast btw 👍

[sockeqwe]

The "clean" way would be to let the information that you have applied List<AdapterCommands> or DiffResult flow down to the state reducer. So basically its just yet another intent. The State Reducer then would create a new "ViewModelState" with an empty list in case of AdapterCommands or an empty DiffResult since "nothing" has changed in the data list. Then render(newState) will be called again, but since List<AdapterCommands> / DiffResult is empty no ViewHolder will be updated, so no UI widget will be updated at all, but you are in a clean new state. Then after a screen orientation change you would apply exactly the previous model with the empty List<AdapterCommands> / DiffResult

Other workarounds that may or may not work in your use case:

• Only apply List<AdapterCommands> / DiffResult if RecyclerView's) Adapter's dataset != null. That works for screen orientation changes and most likely for back stack navigation (so it works whenever the view is destroyed).
  • Mosby MVI has a helper MviFragment.isRestoringViewState() that can be used inside your view.render(ViewModelState) to determine whether or not the render() method is called because the view gets reattached (i.e. screen orientation change, back stack navigation etc.) or not. You could implement something like that for your app too.

Go the clean way or use work around? It depends how complex your app / state is and if you are a "purist" or not :)

hope that helps!

[nomisRev]

@sockeqwe thanks for the quick response. I took a look at Mosby MVI and Mosby Conductor MVI but I have yet to give it a try.

I thought of "clearing" the List<AdapterCommands> when detach the view from the presenter, in a similar way as you described just at a different point in time. My presenters are retained on config change, and are singletons for each view. So no conflict can occur in cached view states in the behavior relay.

dataset != null that strategy won't work for me since my adapter is also retained on config change, I'm using Conductor and it scoped to the life of the Controller with Dagger. Either way I'm skeptical about last 2 workarounds since.

Thanks for the response! It was more of a general wonderment I had when listening to the podcast. I'm currently doing it similarly to this https://github.com/pakoito/FunctionalAndroidReference/blob/80dd1a6d26647d3023ae234f0b7b76422eecd9b5/app/src/main/java/com/pacoworks/dereference/widgets/BaseRecyclerAdapter.java but I want to move it away from the adapter / ui thread.

[sockeqwe]

Just a reminder: Retaining the Adapter on orientation changes is a memory leak since ViewHolders and their views has been instantiated with the layout inflater from the very first activity. So that activity cant be garbage collected.

Simon Vergauwen notifications@github.com schrieb am Di., 28. März 2017, 19:13:

@sockeqwe https://github.com/sockeqwe thanks for the quick response. I took a look at Mosby MVI and Mosby Conductor MVI but I have yet to give it a try.

I thought of "clearing" the List when detach the view from the presenter, in a similar way as you described just at a different point in time. My presenters are retained on config change, and are singletons for each view. So no conflict can occur in cached view states in the behavior relay.

dataset != null that strategy won't work for me since my adapter is also retained on config change, I'm using Conductor and it scoped to the life of the Controller with Dagger. Either way I'm skeptical about last 2 workarounds since.

Thanks for the response! It was more of a general wonderment I had when listening to the podcast. I'm currently doing it similarly to this https://github.com/pakoito/FunctionalAndroidReference/blob/80dd1a6d26647d3023ae234f0b7b76422eecd9b5/app/src/main/java/com/pacoworks/dereference/widgets/BaseRecyclerAdapter.java but I want to move it away from the adapter / ui thread.

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/artem-zinnatullin/TheContext-Podcast/issues/1#issuecomment-289839795, or mute the thread https://github.com/notifications/unsubscribe-auth/AAjnrrqyxfaiNGlhtMq6biFPnaRpiIGtks5rqT-qgaJpZM4HUABV .

[nomisRev]

@sockeqwe thanks for the heads up. I was unaware that an Adapter kept actual references to the viewholders. I thought it was more like a sort of factory/populator.

[Harin-Kaklotar]

Hey @artem-zinnatullin @sockeqwe that is awesome podcast episode. and It clear so many things about MVP. It's really helpful if you add some good android MVP demo link ?

[artem-zinnatullin]

It's 2.5 years old episode tho 😸

Afaik all hosts of the podcast moved past MVP since then, we want to record an episode about MVI / Reactive Redux sometime in future, but no ETA