I do know how to save data between config changes (using onSavedInstanceState and checking the Bundle in onCreate, for example, for not being null) and I also do know how to make data persistent (via SharedPreferences), but here I need an in-between-solution.
I have a resource-heavy fragment which gets killed (onDestroyView() and onDestroy() are being called) when I replace it with another resource-heavy fragment. When I return to the former, its state is gone (the Bundle is null), making it persistent would save state across "sessions" which contradicts state-saving in other fragments. Is there a way how have "session-scoped" persistence in Android? By "session-scoped" I mean a period longer than the Android life-cycle and shorter than "forever".
So, there are a bunch of ways of saving the fragment and activity states.
If you have several fragments and an activity, and you are simply replacing the fragment view with those fragments then you may use graph level ViewModel. Check out navigation architecture component.
If you just want to save the state of the fragment, just create ViewModel for fragment view and make its lifecycle as activity lifecycle. So the ViewModel does not get cleared or killed until the host activity for the fragment is destroyed.
You may check out view model here in Android Documentation: ViewModel
Here is the good comparison for managing saved states, which options to consider 1) ViewModel, 2)Saved states or 3) Persistence storage
Depending on the size of data and context you may choose one of the given options. I would fully recommend reading it. Here is the link
Saving UI States
Related
Android Architecture Components provide the LiveData and ViewModel classes which are more lifecycle-friendly and designed for a leaner Activity/Fragment. These classes handle storing data across configuration changes, but I'm confused about their use compared to the Activity framework APIs. Are onSaveInstanceState(Bundle) and onRestoreInstanceState(Bundle) still necessary or useful for preserving activity state?
onSaveInstanceState & onRestoreInstanceState is still useful.
ViewModel holds data only when process is alive.
But, onSaveInstanceState & onRestoreInstanceState can hold data even if process is killed.
ViewModel is easy to use and useful for preserving large data when screen orientation changes.
onSaveInstanceState & onRestoreInstanceState can preserve data when process is in background.(in background, app process can be killed by system at anytime.)
Assume a scenario :
user is in activity A , then navigates to activity B
but because of low memory Android OS destroys activity A , therefor the ViewModel connected to it also destroys. (You can emulate it by checking Don't keep activities in Developer options)
now user navigates back to activity A, Android OS try's to create new Acivity and ViewModel objects. therefor you loosed data in ViewModel.
But still values in savedInstanceState are there.
As well as the other answers which talk about the ViewModel's persistence beyond simply configuration changes, I think there are a couple more use cases:
Performance reasons
Sometimes you don't want to store all of the latest values of view attributes in the ViewModel for performance reasons. You may have greater need to save them when the view is being re-created. For example, user's scroll position on a view within your activity/fragment. You probably don't want to save the scroll position every time the user scrolls. But you might want to save that onSaveInstanceState so you can restore that when the view is recreated (onRestoreInstanceState).
Initialization to perform after restore
Some views may require initialization especially because of the restore, due to the complex design of those views not being able to save everything. For example, I had a WebView and if the user was in the middle of loading a page during a configuration change, I want the WebView to try to load the new page (rather than the old one). After restoring the state, the observers of LiveData will get the latest values but this doesn't help much with something like this (I only want the view to load a page from the ViewModel at the point of restore, not at other times). So we just do that initialization via the restore state.
Final word
With all this stuff I would advocate keeping your onSaveInstanceState and onRestoreInstanceState as simple as possible. Ideally just call a method on the ViewModel and that's it. Then we can extract all of the logic from the view into the ViewModel, and the view is just left with boilerplate code.
I faced with that problem when I started to use ViewPager. As every page is nested Fragment, I can't call setRetainInstance(true) for it. So, I need to store Fragment's state to a Bundle and cancel/recall remote API methods onViewAttached/Detached which I don't want to.
What I learn about this situation:
I can use RecyclerViewPager to avoid using nested Fragments, but I still can't keep an instance of Presenter. One of the ways is to use a static field, but the same thing can I do with the Fragment.
Create some kind of rootViewPager under MainActivity and use it in Fragments via setVisibility(GONE/VISIBLE) and replacing Adapter. So, every Fragment placed into this ViewPager will not have parent Fragment and I will solve my case. Suitable and elegant, but not the best solution as I think.
Any other variants?
In Mosby 3.0 Presenters can be retained even without setRetainInstance(true) ... I would suggest to wait until 3.0 release ...
So, I need to store Fragment's state to a Bundle and cancel/recall
remote API methods onViewAttached/Detached which I don't want to.
Mosby 2.0 does exactly that for you but you have to make your ViewState and your data implement Parcelable. In that case the presenter instance won't survive screen orientation changes, but presenter will "resume" on the same state / point (a new presenter instance will be created, async tasks etc. might be restarted too). See RestorableViewState (javadoc is slightly outdated, because it mentions that this is the only way to work with activities which since Mosby 2.0 is not true anymore)
Since orientation changes happen fairly quickly one would think that keeping a Fragment in memory during that time would be more efficient than recreating it again.
Since it's kept for a short time only, there seem to be little impact on memory.
What then shall be good reasons NOT to use setRetainInstance(true) for each and every Fragment?
What then shall be good reasons NOT to use setRetainInstance(true) for each and every Fragment?
Google's primary concern is that you'll screw up and have data members in the fragment that refer to the old activity that you do not clean up in post-configuration lifecycle method calls (e.g., onCreateView()). For example, you might hold onto a widget in a data member, where you do not immediately null out or repopulate that data member on a configuration change. If your fragment has a reference back to the old activity, the old activity (and everything it holds onto) cannot be garbage-collected until your fragment gets destroyed. This is one of the reasons why Google does not recommend retaining any fragment with a UI.
Firstly maybe check this link: http://android-er.blogspot.com/2013/05/how-setretaininstancetrue-affect.html
Basically you shouldn't always retain the instance of the fragment because fragments are attached to activities. Hence when an activity is re-created (i.e. on configuration change), the fragment needs to be re-associated to the new activity (which leads to extra coding and some extra problems). If you just unnecessarily setRetainInstance(true), you are giving yourself more error checking and coding to do for no reason. By setting the setRetainInstance(true), you will need to deal with a different fragment lifecycle as well because certain methods in the lifecyle are now skipped (i.e. the onCreate() is no longer called after configuration changes). As far as I understand, setRetainInstance(true), won't make it more efficient because you could use onSaveInstance to save any data that you would want to use in the recreation of the same kind of fragment.
I hope this helps.
When starting a new application for a client, I am asking myself again the same question about who should be responsible for loading data: activities or fragments. I have taken both options for various apps and I was wondering which pattern is best according to you in terms of:
limiting the code complexity.
handling edge cases (like screen rotation, screen going power save, loss of connectivity, etc.)
Option 1 - Activity loads data & fragment only displays it
This allows to have fragments that are just fed a bunch of objects to display. They know nothing about loading data and how we load that.
On the other side, the activity loads data using whichever method is required (for instance initially the latest 50 entries and on a search, loads the search result). It then passes it to the fragment which displays it. Method to load the data could be anything (from service, from DB, ... fragments only know about POJOs)
It's kind of a MVC architecture where the activity is the controller and fragments are the view.
Option 2 - Activity arranges fragments & fragments are responsible to fetch the data
In this pattern, fragments are autonomous pieces of application. They know how to load the data they are displaying and how to show it to the user.
Activities are simply a way to arrange fragments on screen and to coordinate transitions between application activities.
In theory you can do whatever you want, if it works.
Actually, the fragments and activities display data and deal with their own life cycles.
Since fragments belongs to activity so you have to use both in conjunction to better handle all the data but mostly it depends on your needs.
If you keep in mind the idea that the Fragment should provide the UI and the Activity should provide the processing then you have a good division of concerns and code which should allow the Fragment or the Activity to be reused.
If you know about the MVC - Model View Controller - design pattern then you can think of the Fragment as the View and the Activity as the Model.
Things get much more interesting when you build an application with multiple Fragments.
Some key points as a decide factor -
The idea of a Fragment is that it is a wrapped up chunk of UI that
can be used by any Activity that needs it. On this basis you have to
ask yourself if the event that has to be handled is the same for
every Activity or unique to each Activity. If it is the same then the
event handler is better written within the Fragment.
The Fragment doesn't have a UI of its own - it is displayed by an
Activity that the Fragment is associated with. The events are
generated by objects in the View hierarchy, which is owned by the
Activity. If you try to use Android Studio to add an event handler,
for example, it will add it to the Activity and not to the Fragment.
You can define the EventListener that you want to handle the event
in the Fragment and then hook it up to the View object in the
Activity in which you want to generate the event.
A Fragment is a class that implements the onCreateView method to
supply a View hierarchy that can be displayed by an Activity.
To use a Fragment in an Activity you have to add it using a
FragmentManager and a FragmentTransaction. You can add the Fragment
using the add method but nothing happens until you call the commit
method.
After the method that used the commit, usually the Activity's
onCreate, terminates the CreateView event runs the Fragment's
onCreateView and the Fragments View hierarchy is added to the
Activity's content.
You have to write code to save and restore any additional state the
Fragment may have.
If a task is common to all instances of the Fragment then its code
should live in the Fragment.
In particular the code to handle events can be defined within the
Fragment.
The Activity should be used to host code that processes the data
provided by the UI.
Attaching Activity event handlers to the Fragment's UI or is
difficult to do correctly.
From scenarios make decision what your app will be. Is it service,
activity, widget , even a content provider or a complex system,
including some different components. Test your decision against
scenarios.
All of these have to work after the Fragment has been destroyed and
recreated.
(1) Initialization of the Fragment, (2) Saving and restoring the Fragment's
state and (3) Implementing something like an event mechanism so the Fragment
can get the Activity's attention
The hardest part is implementing something like an event mechanism.
In the case of the complex system, distribute functionalities and
data entities among application components. Make a list of components
and what they are (activities or smth else).
Make the list of UI components with description what they do (not HOW
yet) These will be widgets and activities or fragments or layouts
later.
Often you will want one Fragment to communicate with another, for example
to change the content based on a user event. All Fragment-to-Fragment
communication is done through the associated Activity. Two Fragments
should never communicate directly.
When your app is perfectly modular, fragments don't know about each
other. You can add a fragment, remove a fragment, replace a fragment,
and they should all work fine, because they are all independent, and
the activity has full control over the configuration.
You can't do anything with a Fragment unless you start a transaction.
Within the transaction you can set up what you want to happen,
usually add the Fragment to the current layout, but nothing happens
until you use the commit method.
Efficient handling of data with Screen Orientation -
When screen orientation changes, Android restarts the running Activity (onDestroy() is called, followed by onCreate()).
To properly handle a restart, it is important that your activity restores its previous state through the normal Activity lifecycle, in which Android calls onSaveInstanceState() before it destroys your activity so that you can save data about the application state. You can then restore the state during onCreate() or onRestoreInstanceState().
However, you might encounter a situation in which restarting your application and restoring significant amounts of data can be costly and create a poor user experience. In such a situation, you have two other options:
1) Retain an object during a configuration change
Allow your activity to restart when a configuration changes, but carry a stateful Object to the new instance of your activity.
2) Handle the configuration change yourself
Prevent the system from restarting your activity during certain configuration changes, but receive a callback when the configurations do change, so that you can manually update your activity as necessary.
What I would do is manage all data flow (bluetooth, database storage, etc)
in the Activity and use Fragments only for UI display or handling user input.
This way is easier to handle configuration changes/ screen rotations.
Also, if data flow things are heavy to be on UI thread, consider using a Service with a background thread.
If it is a "one shot" thing, you can use an IntentService,
otherwise you can implement a Bind Service and request a bind from anywhere you have Context.
For more read - fragment-and-activity-working-together.
Ideally neither Activity nor Fragment with UI should contain any "model" logic - these classes should be lightweight and responsible only for UI logic. But when you decide to make a separate model object you have a dilemma to choose where to initialise and store this object and how to deal with configuration changes. And here comes some handy trick:
You can create a model Fragment without UI, make it retain instance to deal with configuration changes (it's AFAIK the simplest way to save data across config. changes without troubles) and retrieve it anywhere you need via findFragmentById(). You make all expensive operations inside it once (using background thread, of course), store your data and you're done.
For more info, see Adding a fragment without a UI section.
UPD: There's now a better way to deal with configuration changes: ViewModel from Google's Architecture Components. Here's a good example.
I prefer and always implemented Option-2 over Option-1.
Reasons for not selecting Option-1:
We should have listeners for events triggered in Fragments and pass it back to activity to load data, process it and push it back to fragment, which makes work more complex.
An Activty can load any number of Fragments, Typically you end up questioning these questions to yourself in a scenario where your app is highly scalable and is already huge. Writing all the events in an activity and passing it over to fragment will be an complex altogether.
As #Ved Prakash mentioned, Handling screen orientation becomes complex if orientation is handled by Activty.
I have an example:
your application have 2 features A and B. the 2 features are independent each other. and each feature has a lot of screen.
you should create 2 activities A and B because when Activity A is used, Activity B should be released to reduce memory of app. And the same when B is used, A should be released. The memory of Context A and B are independent, if you want to send data from Activity A to B you must use intent or use global variable in Application Context. intent is managed by OS, not by application. When A send intent to B, if A is destroy is no problem with intent send to B. Activity is App module, it is can call by other applications (fragment is impossible).
for example: feature A has a lot of screen (ex: Fragment1, Fragment2). they use the same data and depend on each other. each screen should be a fragment. and when process with data you can get reference to data by calling function getActivity() and then get reference to variable of Activity context (or Activity memory) to write or read it. Fragment1 and Fragment2 are belong to Activity A Context.it means that Fragment 1 and Fragment 2 can transfer data with each other via variable of Activity context, it is easy to do . noticed that Intent is managed by OS,it is so expensive when send data via Intent.
For restoring the state of the activity after it is recreated (for instance after the screen-orientation change) I implemented onSaveInstanceState() and onRestoreInstanceState(). It is simple to save/restore simple information like int, double etc. But what about saving/restoring objects like Timer?
You cannot store "live" objects (like db connections) in the Activity arguments or saved instance data. Those mechanisms are designed so that the application can be completely stopped, so it only works with things that can be "serialized" and later restored.
What you can do is use fragments. If you add a fragment without UI (check here, look for
“Adding a fragment without a UI”) and call on it setRetainInstance(true) the fragment will get reattached to the activity, surviving any configuration change.
Hope it helps. (Remember you can use fragments with old Android versions by using the support package)