cancel onAction when leaving the current state #175
Conversation
|
I broke something during the rebase, will look into it. |
|
It's fixed now the test failures are the already failing onEnter and CustomIsInStateDslTest tests |
| isInState: (S) -> Boolean, | ||
| getState: GetState<S>, | ||
| transform: (Flow<Action<S, A>>) -> Flow<Action<S, A>> | ||
| ) = channelFlow { |
There was a problem hiding this comment.
Why do we need a ChannelFlow here?. wouldn't it work with just a regular Flow as well?
There was a problem hiding this comment.
The reason is that we emit/send from within the launch. Just using flow { coroutineScope { ... } } violates the context preservation requirement of flow and crashes, you can see that in the commit from before the force push
| getState: GetState<S>, | ||
| transform: (Flow<Action<S, A>>) -> Flow<Action<S, A>> | ||
| ) = channelFlow { | ||
| var currentChannel: Channel<Action<S, A>>? = null |
There was a problem hiding this comment.
Do we need a mutex to ensure no 2 actions are processed in parallel or is there auch a guarantee already in place under the hood by channelFlow?
There was a problem hiding this comment.
Flow itself is sequential and under the assumption that collectors are not thread safe. So when we collect the upstream here we only get the next action when the previous one was handled which happens after we send it into currentChannel. So we sequentially put them into the channel and it also applies to creating/closing the channel. For the processing itself we are collecting the channel as a flow again, so it's als sequential and ordering of the action processing then depends on the transform. For onAction specifically it depends on the flatMapPolicy, CONCAT and LATEST are not parallel, while MERGE is.
Closes #128
This one is more complex then fixing onEnter. Just the boolean that we have there because we still need the action value and new actions while we are in the state should not cancel the previous flow. We basically want a
flatMapSometimesLatest. The new operator will put new upstream actions into aChannelwhenisInStateis true. ThisChannelis then turned into a Flow and we apply our filtering and transformation with the flatMapPolicy on it. The resultingFlowgets collected and it's emissions are emitted to the downstream. WhenisInStatebecomes false the channel is closed so that the resulting flow will get cancelled as well.