I have a rather complicated List with nested RecyclerViews. I get it that nested RecyclerViews aren't the best solution, but in my case it is one of few solutions that create structured code and meet the requirements. I have attached an image of the structure. You can take telegram as an example to improve your understanding of the structure. Basically I have an outer RecyclerView RV-1 with Items RV-1-Item and an inner RecyclerView RV-2 with Items RV-2-Item. So far so good, my problem is that the outer RecyclerView recycles views as intended, but if one of the RV-1-Items comes into view, all ViewHolders of RV-2 are created (That means that sometimes more than 100 ViewHolders are created). To sum it all up my question is how to force the inner RecyclerView RV-2 to recycle ViewHolders as well.
I know that the inner RecyclerView RV-2 has to have a hight of wrap_content because it depends in the count of the inner items, also i cannot set setHasFixedHeigth(true) (and I don't know if it would help) because during runtime new RV-2-Items can be added into RV-2. I also tried to set setNestedScrollingEnabled(false) on RV-2 because I read a lot about it online but it didn't help me either.
So basically this is how I configure
RV-1
layoutManager = LinearLayoutManager(context)
isNestedScrollingEnabled = false
RV-2
setHasFixedSize(true)
layoutManager = LinearLayoutManager(context).apply {
reverseLayout = true
}
In addition to that I have some ItemDecorators but they only create the space between the items, so they shouldn't have to do anything with the problem.
To sum it all up:
The outer RV-1 recycles it ViewHolders as intended but the inner RV-2 creates all ViewHolders at once, even if they are not on screen. I assume that this is the case because RV-2 has a height of wrap_content and when the layout_height need to be measured it needs to create all views. THE QUESTION: Is there a way to force RV-2 to recycle its views?
EDIT:
Also I am using a shared RecycledViewPool between all RV-2 RecyclerViews but that isn't really related to the problem, because even if the ViewHolders are shared between the RecyclerViews, an RV-2 RecyclerView shouldn't create ViewHolders that aren't visible when it is initialised.
EDIT 2:
A lot of comments and related questions say that two vertical nested RecyclerViews isn't a possible thing in android, in case all visitors of this question think the same my question is: How would you implement such a structure. It is obvious that I could make a single view which has a IM (Round Image View) and RV-2-Item and just make the IM invisible when it isn't needed. In my opinion this somehow makes the structure more complicated. Furthermore a requirement is that the IM on the left side of RV-1-Item must have the ability to move up and down in RV-1-Item, which is obviously easier with my current structure.
EDIT 3: (My last one I promise)
The Problem I have shown can be solved by using the approche I explain in my EDIT 2, even if it isn't the best solution it would work. But the issue is that I have an even more complex screen where this approche wouldn't work anymore because I have three nested RecyclerViews. I could get that number down to two with the approche of EDIT 2 but I would still be left with two nested RecyclerViews and I cannot think of a workaround that could solve the problem of the remaining two nested RecyclerViews. I attached an image of the even more complex screen which contains a the interface of the app with marked sections to help you to understand the structure.
(Not quite an answer to your specific question in solving "how to not get the RecyclerView to create all items at once", but something that most likely will fix your specific problem by not using nested recyclerviews at all)
I would suggest (in a quite similar way as already suggested in this answer), to flatten your feed into one recyclerview
(No matter how much you tweak your nested recyclerview architecture, imho it will never be as performant than having just one recyclerview, and as you don't need nested scrolling (I guess), just one recycler view should be your best option).
I would propose to not think of your feed in the way your data is structured, but in a way you want to show it and how it can be split into smaller items which are "look alikes" / consist of the same things.
From your screenshot I would see for example the following items / view types for each chat item:
Chat header (the thing with the icon and the text "New Group")
the user badge (the picture with the text "Jürgen")
a message item (one bubble of text, so e.g. in your screenshot at the bottom there would actually be 3 of those items, one for each message)
The section with the date and the action/reply items.
Those items are way smaller than a whole chat item, and therefore can be faster created / recycled.
For each of those items, create a view-type and a view-holder, and treat them as seperate recycler-view items.
The recyclerview will, when the getItemViewType method is correctly used, create / prepare the correct type of view for the position you need.
For this to work, the adapter needs to add some logic, as your data most likely will be structured something like
a list of chats, and each chat has a name and some messages to display
and we need it as
the first 6 elements are for the first chat, where the first position
is the header, the second the user badge, the next 3 items are message
items and then we need an action item.
So you basically need to calculate how many recyclerview items you will need to show each single chat-item, which could be a calculation along the lines:
1 chat header item + 1 user badge item + 3 message items + 1 action/reply item = 6
This calculation needs to be performed for each chat item of your data list separately.
So if you only have this single chat item in your list of data to display, you actually need to tell the adapter to create 6 items (by returning in this case 6 at getItemViewCount()).
Then, you need to tell the adapter using the getItemViewType(position: Int) function, at which position of the recyclerview which type of view the adapter needs to prepare.
So there you again need some logic to say that e.g. on position 0 the chat header for the first chat item should be, at position 1 the user badge for the first chat item, at position 2-4 message items should be, on position 5 the action item and at position 6 the chat header for the second chat should be and so on
(again, the logic then needs to be in place for all chat items, and it can get really messy / complicated, as to calculate each chat items view types for a position, e.g. all prior chat element view counts need to be recalculated, too (in order to know at which recycler-view position your current chat item starts)).
As this tends to blow your adapter up, I would suggest (if you don't already do so), to get some manager / delegate architecture in there.
So e.g. have a delegate for each view type, and a manager which calculates the number of recyclerview items / view types needed for each chat item.
Just for reference:
Some time ago we had a situation similar to yours
(a recycler-view with a design similar to a social media feed, which should show the first n comments in the feed and we displayed the comments for each feed item (which was a recyclerview item) with another recyclerview in the item) and also after some troubles with performance which we could not manage to resolve just flattened the recyclerview, and never had performance troubles again.
A lot of comments and related questions say that two vertical nested RecyclerViews isn't a possible thing in android
This is not true; whoever says this is not a thing has not done it and thinks it's not possible. It is possible, albeit with complications, side-effects, and most likely, the annoyance of your users when they tap around trying to scroll up/down and the wrong touch interceptor wins.
Why is this a problem?
On iOS, when you try to do something that the platform devs didn't think it was good, most people and other devs scream at you: don't fight the framework!!!.
On Android, we see the craziest Java (and now Kotlin) implementations of things that makes you wonder what are we -developers- learning at school and what are we teaching?! and yet nobody says anything (for the most part) :p
The truth is, you're trying to design a complicated user interaction and data transformation, and yet, your attempt is biased by trying to use the data "as you have it" (which implies dealing with these two different RV/Adapters), as opposed to do what one should do: transform the data for presentation.
This leads me to the next question:
How would you implement such a structure.
Well, for starters, I don't know how your data looks like, nor where it's coming from; I don't know what your users can do with your data, outside of the obvious scrolling.
I also don't know how your data wants to be presented, aside from your mock up.
But I do know the situation very well. A list of things, which also contain their own list of things.
Case: The List of List
It is doable; you can have a list and inside said list, have another list. I've done it. I've seen it done by others. I've used it. I also never liked the idea of having this "small" scrollable thing, fighting to see who scrolls first when I tap "the wrong place".
I would not do this. If the inner-list is big (say more than 3 items per outer item), I would not present it as scrollable content.
What I would do (considering the things I do not know about your problem) is to have a single list displaying all the content properly flattened.
This has a issue with your content:
What if the inner-lists are super long, wouldn't this cause them all to be displayed? YES, and that's why I wouldn't do it this way if the data (as you described) can have 100 items. An options is to display the 3 first items with a "more" link to now open the inner-list "full screen"; this is 10 times better than the nested list from a user's PoV and from the technical aspect of it.
Another alternative, is to keep this single long list (RV-1) and let users "expand" the list to launch another full-screen list depicting the contents of RV-2, in a separate window. This is even better.
The time you'll spend implementing this and getting rid of the mess of code you probably have right now, will make you wonder why didn't you suggest this in the first place.
If this is something you absolutely cannot do, then I cannot offer you much more advice, for now you're tied to unknown to me business/product rules. Ultimately, the price will be paid by the users of your app, when they have to scroll that nightmare :)
Take a Step Back
Let me be clear, I am not criticizing you or your solution; I'm merely pointing out that, in my experience, this "pattern" you have here is not a good user experience.
Format your data for presentation, not the other way around. Your data should be properly shaped so it can be properly presented with the tools you have.
You're fighting against the tools Android is giving you; you're giving a RecyclerView (and its adapter) a lot of new problems to deal with when it already has a lot going on.
Think about it: RecyclerViews have to do a lot of things; Adapters must also conform to a few interfaces, ensure things are dispatched as soon as possible, calculate Diffs (if using a ListAdapter<T,V>), etc. Activities/Fragments? They have a lot on their plates dealing with ... well "Android"; now you're asking all these components to also handle a complicated scenario of scrolling content, touch recognition, event handling, view inflation, etc.
All this, while expecting each view to take 16ms or less (to stay above 60 FPS scrolling speed, your view/viewHolder should not take more than 16ms to do all it needs.
Instead, I'm asking you to take a step back, grab the data you have, compose it, transform it, map it, and create the data structure that can better serve the components you have (a RV + Adapter + a simple View).
Good luck :)
I'm currently researching why some of my RecyclerViews do not perform very well. The UI lags especially when the whole content of the view needs to be drawn.
Using systrace I've come across a large frame in which the RecyclerView layouts its views:
As you can see the same views are drawn repeatedly. Inside the systrace I've found the following description on the RecyclerView frame:
OnLayout has been called by the View system. If this shows up too many times in Systrace, make sure the children of RecyclerView do not update themselves directly. This will cause a full re-layout but when it happens via the Adapter notifyItemChanged, RecyclerView can avoid full layout calculation.
The views of my RecyclerView are using google's data binding with observable values.
Meaning to update a TextView I don't use textView.setText(titleTxt) but title.set(titleTxt) where title would be of the type ObservableField<String>.
So evidently I'm not calling notifyItemChanged but only update observable fields on objects bound to the layout.
Is it wrong to use data binding in this case performance wise? If so, whats the point of data binding then? Mobile UI consists largely of lists.
Unfortunately Google left data binding behind with the arrival of Kotlin, so if you think it is slowing things down like it did with me, just stop using it or move to Kotlin all together. Kinda sucks for us to try and work with things that are constantly breaking and being left behind, but that's just what Google decided to become.
So our android app uses recyclerview with a cursor adaptor and then we are using https://github.com/codepath/android_guides/wiki/Endless-Scrolling-with-AdapterViews for endless scrolling with a loader reset event (that will on the next create loader increase the limit for the amount of data we load) that will then update the view.
But the issue is, every time this happens, the recyclerview is at the bottom. So to solve that we are recording the last visible item with findFirstVisibleItemPosition and then using scrollToPosition when the view is ready.
But we have found that if the user scrolls past that point, they will get this jump / jitter as the load finishes because it loads them to that exact item when in fact they are sometimes partly thru it. We try to use scrollTo with exact X and Y (that we record beforehand) but that is not working with recyclerviews as I guess they won't know the exact dimensions of the items.
I know that we can for example only trigger this at the end of the scroll (IE have the buffer size to be 0 or 1) so we know we are at the exact end of the list and can then begin there, but we would like this to be as smooth as possible.
If there is any suggestion I would love to know.
This is not a code problem, I interpret the guidelines as that being OK.
I've been researching a way of building an infinitely scrolling calendar-like view in Android, but I've reached an impasse.
Right now my dilemma is that most of the similar views available have their children placed relative each other in a recurring style. With this I mean:
item 4 comes after item 3, which comes after item 2, and there is constant padding/margin between all items.
What I need is a way to produce an infinitely long scrollable view that may, or may not, contain items. The items should be placed at variable positions within the view. The best way I can describe a similar looking view is a one-day calendar-like view that is infinitely scrollable.
So far my best two bets are using the new RecyclerView with a custom LayoutManager (this seems very complex and still not perfectly documented by Google though). I like this approach because, among other things, it is optimized for displaying large sets in a limited view.
My other solution would be to build a completely custom View. However, with that solution I loose the adapter unless I build a container view (which is probably more complex than building a layout manager).
How would you go about solving such a problem? Tips are appreciated, I don't need code examples, just ideas which path is the best to solve this problem.
Thanks.
Apologies if I've misunderstood the guidelines
Edit: How I resolved this problem
My first solution to use RecyclerView with a special Decorator seemed promising, but it remained a "hack" so we decided not to go for that solution since we were afraid of the complications that it would create down the line.
To solve the problem I went with a SurfaceView instead of an Adapter, this means having to rewrite all the adapter-functionality for my SurfaceView but it seemed to be the best way of solving this issue of very custom drawing and layout managing for my use-case.
It still would be nice to build a custom Viewgroup that can handle this kind of layout problems.
ListView and ListAdapter are based on a fixed list, so the current infinite-scrollers just keep adding more and more data to the end of the list.
But what you want is scroller similar to Google's Calendar app which has a bi-directional infinite scroller. The problem with using ListView and ListAdapter in this case is that if you add data to the front of the list, the index of any one item changes so that the list jumps.
If you really start thinking about this from the MVC perspective, you realize that ListAdapter does not provide a model that fits this need.
Instead of having absolute indexing (i.e. 1, 2, 3, 4, etc), what you really want is relative indexing, so instead of saying "Give me the item at index 42" you want to say "here's an item, give me the five items before it". Or you have something like a calendar date which is absolute; yet — unlike your device's memory — it has effectively no beginning or end, so what you really want here is a "window" into a section of that data.
A better data model for this would be a kind of double-ended queue that is partly a LRU cache. You place a limit on the number of items in the structure. Then as prior items are loaded (user is scrolling up) the items at back end are pushed off, and when subsequent items are added (user is scrolling down), items at the front are pushed off.
Also, you would have a threshold where if you got within a few items of of one edge of the structure, a "loadNext" or "loadPrevious" event would fire and invoke a callback that you set up to push more data onto the edge of the structure.
So once you've figured out that your model is completely different, you realize that even RecyclerView isn't going to help you here because it's tied to the absolute indexing model. You need some sort of custom ViewGroup subclass that recycles item views like a ListView, but can adapt to the double-ended queue. And when you search code repos for something like this, there's nothing out there.
Sounds like fun. I'll post a link when I get a project started. (Sadly, it won't be done in any timely manner to help you right now, sorry.)
Something that might help you a little sooner: look at Google's Calendar implementation and see how they did it: Google Calendar Git repo
What you may be searching for is a FragmentStatePagerAdapter , where you can implement a swiped view, meaning when the user (for example)swipes to the right, a completely new view is displayed.
Using a FragmentStatePagerAdapter , you can handle a huge amount of views without overflowing the memory, because this specific PagerAdapter only keeps the views' states and is explicitly meant to handle large sets of views.
Keeping your example of a calendar, you can implement swiped navigation between for example weeks and generate the week views on demand while only keeping for example the year and the week's number as identifiers.
There are plenty of online tutorials for Android, maybe you have a look at this one
I have a list of lists of lists, implemented as a MvxListView.
It looks like
-Parent-
MvxListView containing items of Fruit
-HumanCell-
Some forms and an MvxListView containing items of ExistingFruitColors
-ExistingFruitColors-
Some forms and an MvxListView containing items of ExistingShapes
-ExistingShapes
Some forms
I noticed that when I'm navigating in my app (this closing and reopening the ViewModel) as well as when I'm scrolling in the listview, I'm seeing some garbage collecting GC_BRIDGE appearing in the debug screen. In 2 minutes, memory consumption can grow from 6MB to 12MB for a simple 30 cells list.
The fact that a lot of object are created is not a problem, but I'm curious if it's possible to override a method either in the adapter or in the list to garbage or dispose of some views and listeners.
There is no changes happening in the list, but there's a lot of memory variations.
I'm thinking about listening to ViewModel changes and only update the listview canvas when a real change happens, but that's an ugly solution.
Is there something I should be aware when nesting binded lists into binded lists?
Can I force the MvxListView to just draw my items and not attempt to do anything else with it?
Is there something I should be aware when nesting binded lists into binded lists?
IMO, generally, this is normally better done as a single list with sections/groups. It's easy to use header/footer cells for each section and to use a custom adapter to decide which cell to draw.
Having lists within lists can also give a strange touch experience for the user - since the lists can scroll independently - but if this is the UX you desire, then this should be OK.
Can I force the MvxListView to just draw my items and not attempt to do anything else with it?
Sorry, but I'm not entirely sure what this question means.
In general, MvxListView (and ListView) only draw the cells/items currently on screen - recycling is used on cells to reduce the memory used.
When a Activity is Finished, then the MvxListView's ItemSource is set to null - which should result in the ListView memory being cleared up when the Activity is complete.
One thing to be aware of is that in order to estimate the total list scroll size, then the list view does initially have to consider all the list items - it does have to estimate how high each cell is.