Fixed
Status Update
Comments
ch...@gmail.com <ch...@gmail.com> #3
I think I've worked it out, it's a bug in the generated DAO method. A single RoomSQLiteQuery _statement is acquired outside of the Flow Callable, and is re-used for every Flow call. I assume that the acquired RoomSQLiteQuery is re-used later by a different query so breaks the Flow.
Should be a simple fix to move the acquire() within the call() method, so a fresh one is used each time.
Should be a simple fix to move the acquire() within the call() method, so a fresh one is used each time.
da...@google.com <da...@google.com> #4
The issue is not that the query is being acquired early outside the callable, is that we are finalizing it and returning it to the pool too early (on the first cursor iteration actually). Should be an easy fix, sorry about that.
ap...@google.com <ap...@google.com> #5
Project: platform/frameworks/support
Branch: androidx-master-dev
commit f854749a335aa41b1620cc4341f5bf13492bc7f8
Author: Daniel Santiago Rivera <danysantiago@google.com>
Date: Mon Aug 12 09:39:46 2019
Make CoroutineFlowResultBinder extend BaseObservableQueryResultBinder
Enabling the binder to generate code to correctly recycle the SQL
statement when the Callable is GCed instead of the first cursor
iteration.
Additionally upgraded Kotlin Coroutines to 1.3.0-RC2
Bug: 139175786
Test: Manual
Change-Id: Icf194ae5be50d1e2494304662a30481b60f21403
M buildSrc/src/main/kotlin/androidx/build/dependencies/Dependencies.kt
M room/compiler/src/main/kotlin/androidx/room/solver/query/result/CoroutineFlowResultBinder.kt
https://android-review.googlesource.com/1100157
https://goto.google.com/android-sha1/f854749a335aa41b1620cc4341f5bf13492bc7f8
Branch: androidx-master-dev
commit f854749a335aa41b1620cc4341f5bf13492bc7f8
Author: Daniel Santiago Rivera <danysantiago@google.com>
Date: Mon Aug 12 09:39:46 2019
Make CoroutineFlowResultBinder extend BaseObservableQueryResultBinder
Enabling the binder to generate code to correctly recycle the SQL
statement when the Callable is GCed instead of the first cursor
iteration.
Additionally upgraded Kotlin Coroutines to 1.3.0-RC2
Bug: 139175786
Test: Manual
Change-Id: Icf194ae5be50d1e2494304662a30481b60f21403
M buildSrc/src/main/kotlin/androidx/build/dependencies/Dependencies.kt
M room/compiler/src/main/kotlin/androidx/room/solver/query/result/CoroutineFlowResultBinder.kt
da...@google.com <da...@google.com>
ch...@gmail.com <ch...@gmail.com> #6
Just confirming that the fix works in the latest snapshot
al...@gmail.com <al...@gmail.com> #7
May I ask, how can you access the latest snapshot?
Description
Version used: 2.2.0-alpha02
Devices/Android versions reproduced on: API 29
I'm hitting a weird issue where a Flow<T> from Room will randomly stop dispatching updates (usually after a few successful emissions). I've dug into this a bit and found this in my logs:
2019-08-10 15:29:39.928 19259-19367/app.tivi.debug E/DiscoverViewModel: Fail() during execute
java.lang.IllegalArgumentException: column '`show_id`' does not exist. Available columns: [id, request, entity_id, timestamp]
at android.database.AbstractCursor.getColumnIndexOrThrow(AbstractCursor.java:340)
at androidx.room.util.CursorUtil.getColumnIndexOrThrow(CursorUtil.java:108)
at app.tivi.data.daos.PopularDao_Impl$13.call(PopularDao_Impl.java:257)
at app.tivi.data.daos.PopularDao_Impl$13.call(PopularDao_Impl.java:249)
at androidx.room.CoroutinesRoom$Companion$createFlow$1$1.invokeSuspend(CoroutinesRoom.kt:81)
at kotlin.coroutines.jvm.internal.BaseContinuationImpl.resumeWith(ContinuationImpl.kt:33)
at kotlinx.coroutines.DispatchedTask.run(Dispatched.kt:241)
at androidx.room.TransactionExecutor$1.run(TransactionExecutor.java:45)
at java.util.concurrent.ForkJoinTask$RunnableExecuteAction.exec(ForkJoinTask.java:1412)
at java.util.concurrent.ForkJoinTask.doExec(ForkJoinTask.java:285)
The columns listed in the exception have no relevance to the Flow query (they're from a completely different table). The table in question is also not mentioned in the generated calling code.
I *think* this is probably a cache invalidation issue, maybe in RoomSQLiteQuery.acquire, where the acquired query isn't properly invalidated so the "old" query is still being used?