Status Update
No update yet.
Comments
ma...@google.com
as...@google.com
So there does seem to be a bug in Navigation that causes this to fail.
When it navigates with popUpTo and removes SecondFragment, that removal in fragment is set for a frame later and before it is actually removed the navigate call in the onViewCreated of SecondFragment enqueues another call to add it back. But we still need to destroy SecondDetailFragment which ends up marking the NavBackStackEntry associated with SecondFragment as complete before navigation has received the call from the fragment that it has been added back. So by the time it hits this check, the SecondFragment is in the correct state, but Navigation has no way of referencing it so it assumes that it is not associated with an entry.
But this is not the way this type of situation should be implemented. Instead of having the destinations as part of the NavGraph, they should:
- be managed by the childFragmentManager of
SecondFragment. So instead of going back toSecondFragmentjust to go somewhere else, each of the fragments take the entire content ofSecondFragmentand are swapped out. - for some sort of a/b testing, set different graphs in the Activity. The graphs would keep the same
SecondFragmentid, but swap out the names and then you choose the correct graph based on the condition.
br...@monzo.com
as...@google.com
- Have
SecondFragmentas part of your graph and then inside ofSecondFragmentcall thechildFragmentManagerandreplacewith the fragment you actually want to show, whether that isSecondTypeOneorSecondTypeTwo. - Create different graphs and follow along with the
docs
br...@monzo.com
shardViewModel.onGetTypeSecond.observe(viewLifecycleOwner) { isType1 ->
if (isType1) {
val graph = findNavController().graph
val carInstallmentNavGraph =
graph.findNode(R.id.nav_second) as NavGraph
carInstallmentNavGraph.setStartDestination(R.id.secondTypeOneFragment)
findNavController().navigate(SecondFragmentDirections.actionSecondFragmentToSecondTypeOneFragment())
} else {
val graph = findNavController().graph
val carInstallmentNavGraph =
graph.findNode(R.id.nav_second) as NavGraph
carInstallmentNavGraph.setStartDestination(R.id.secondTypeTwoFragment)
findNavController().navigate(SecondFragmentDirections.actionSecondFragmentToSecondTypeTwoFragment())
}
}
and removed app:popUpTo="@id/secondFragment" and app:popUpToInclusive="true"
<fragment
android:id="@+id/secondFragment"
android:name="com.example.navigationdeeplink.SecondFragment"
android:label="fragment_second"
tools:layout="@layout/fragment_second" >
<action
android:id="@+id/action_secondFragment_to_secondTypeOneFragment"
app:destination="@id/secondTypeOneFragment" />
<action
android:id="@+id/action_secondFragment_to_secondTypeTwoFragment"
app:destination="@id/secondTypeTwoFragment"
app:popUpTo="@id/secondFragment" />
</fragment>
ap...@google.com
Is there a planned fix for this issue? We've been stuck on Navigation 2.5.3 for some time due to issues in 2.6.x and now this issue, and the suggested workaround of restructuring our navigation graph is a pretty big lift. If that's the only recommended path it would be helpful to know so we can try to plan for that work, otherwise we'd want to just wait for the fix that doesn't require restructuring.
br...@monzo.com
This was fixed by
na...@google.com
Branch: androidx-main
commit b292676aa37122f0ca0215023dcdbe27168b2469
Author: Jeremy Woods <jbwoods@google.com>
Date: Wed Sep 27 21:25:35 2023
Add a test for navigate with popUpTo nested graph
Adding a test to ensure that if the start destination of a nested graph
is popped off the back stack when a fragment is created, Navigation does
not crash and gets to the correct position.
Bug: 287133013
Test: adding a test
Change-Id: Ie1e03aeeb00b730b7688311741847b2312bdbaa0
M navigation/navigation-fragment/src/androidTest/java/androidx/navigation/fragment/NavControllerWithFragmentTest.kt
Description
Jetpack Compose version: I've tested on 1.3.0-alpha01 and 1.2.0-beta03
Jetpack Compose component(s) used: LazyColumn
Android Studio Build: 221.3427.89.2211.8689873
Kotlin version: 1.7.10
Steps to Reproduce or Code Sample to Reproduce:
Context:
We have a rather large (10k+ items) LazyColumn where scrolling (eventually) leads to the application grinding to a halt, constantly trying to free memory. It can't, we just keep getting lots of these:
Background concurrent copying GC freed 51964(1414KB) AllocSpace objects, 6(120KB) LOS objects, 0% free, 190MB/192MB, paused 101us total 1.483sI took at heap dump at this point and basically all of the heap is used up by instances of
SnapshotMutableStateImpl$StateStateRecord.Replace LazyColumn with a RecyclerView (with each item being a ComposeView) and the memory leak is gone. App performance is night and day better.
I've been trying for hours to reproduce this in a trivial sample project and haven't had any luck yet. Raising this anyway just in case you have any ideas about what might be causing it. Obviously this makes LazyColumn completely unusable for us, so we'll stick to RecyclerView for now!