Conversation
540cd24 to
af160bd
Compare
...tioncomponent/src/main/java/com/mapbox/maps/plugin/locationcomponent/LocationProviderImpl.kt
Outdated
Show resolved
Hide resolved
|
|
||
| ## main | ||
| ## Bug fixes 🐞 | ||
| * Fix memory leak in location component. ([#1093](https://github.com/mapbox/mapbox-maps-android/pull/1093)) |
There was a problem hiding this comment.
@Chaoba could you explain how was this reproduced?
There was a problem hiding this comment.
Open and close LocationComponentActivity several times could reproduce it.
There was a problem hiding this comment.
Yeah I mean - aren't we hiding the actual problem here? Why isn't unRegisterLocationConsumer called here? Shouldn't lifecycle call it always?
There was a problem hiding this comment.
Actually, we do call unRegisterLocationConsumer, but still have this issue.
There was a problem hiding this comment.
Hmm, how is this possible at all?
There was a problem hiding this comment.
This can only be reproduced with compass engine, guess locationEngine holds the reference of location plugin as well as sensorManager.
There was a problem hiding this comment.
Maybe we don't unsubscribe from compass engine somehow? I'd say compass is also very common component so if memory leaks were reproducible it would leak in many apps out there
There was a problem hiding this comment.
The root cause should be locationEngine holds the reference the callback during request location update even we remove it. And then the location compass is kept and causes the leak.
Only request location update or have location compass can't reproduce this issue, must combine these two together.
There was a problem hiding this comment.
I've digged it a bit and it seems found the root cause. Thing is - we use Activity context to create SensorManager instance at CompassEngine.
And callbacks, that we add / remove to the LocationEngine indeed are retained, even after remove is called.
So the issue became visible with the CompassEngine addition, since it holds SensorManager, which holds Context, which holds (and is) Activity.
It seems to me that we should use application context everywhere, so that in future such problems won't be possible at all.
PS
Context for the issue is interesting as well. It seems that our implementation of LocationEngine in mapbox-android-core lacks certain logic cleaning the callback or otherwise such issues wouldn't be possible as well.
The issue with LocationEngine keeping its listeners even after removeLocationUpdates has very long story, raised here, reworked here and followed with the documented recommendation to avoid leaks using WeakReferences here published here.
I'll cut a ticket in the lib repo to apply the fix as well.
There was a problem hiding this comment.
...tioncomponent/src/main/java/com/mapbox/maps/plugin/locationcomponent/LocationProviderImpl.kt
Show resolved
Hide resolved
af160bd to
c3b6992
Compare
c3b6992 to
14a9300
Compare
...tioncomponent/src/main/java/com/mapbox/maps/plugin/locationcomponent/LocationProviderImpl.kt
Outdated
Show resolved
Hide resolved
14a9300 to
ae404ce
Compare
ae404ce to
af258e8
Compare
Summary of changes
This PR adds a new class
CurrentLocationEngineCallbackto holdLocationProviderImplinstance with a weak reference to avoid memory leak issue.Invoke
removeCompassListenerwhen there is noLocationConsumerregisted.User impact (optional)
Pull request checklist:
@JvmOverloads,@file:JvmName, etc).mapbox-maps-androidchangelog:<changelog></changelog>.v10.[version]release branch fix / enhancement, merge it tomainfirstly and then port tov10.[version]release branch.Fixes: < Link to related issues that will be fixed by this pull request, if they exist >
PRs must be submitted under the terms of our Contributor License Agreement CLA.