Flatten changes #2976
Flatten changes #2976
Conversation
When used by concat, if the current subscription completed at the same time as a new signal was was received, the new signal could be subscribed to immediately in addition to the queued up signal that was supposed to be subscribed to next. This commit fixes the race condition, making sure that signals are only subscribed to in the order they are received by the flatten.
Had an app crash on the line that invokes the subscribeToSignal after it looked like things were starting to be disposed of. Added a check to prevent trying to call a block that's already been set to nil.
It looks like this is half of #2433, which I somehow missed when I was looking for patches that hadn't been back ported. Would you mind updating the changes here to match those or opening the other changes against |
| @@ -1907,6 +1907,61 @@ + (void)configure:(Configuration *)configuration { | |||
| expect(@(flattenDisposable.disposed)).toEventually(beTruthy()); | |||
| } | |||
| }); | |||
|
|
|||
| qck_it(@"only subscribes to one signal at a time - even when adding new signals at the same time as one completes", ^{ | |||
| for (int i = 0; i < 100; i++) { | |||
There was a problem hiding this comment.
I'm not wild about adding a long test like this. 😕 It takes 1.6s on CI.
|
@baconpat Are you still interested in pursuing this? |
|
I'd still like to try to get these changes in but I haven't had a chance to address your concerns. I'll try to get to it soon. |
|
The Objective-C ReactiveCocoa APIs have been moved to https://github.com/ReactiveCocoa/ReactiveObjC/. If you're interested in continuing this PR, please reopen it against that repository. Thanks! |
I ran into a couple of issues I was able to trace back to
-flatten:. This pull request fixes them.-flatten:when it's used by-concat. If the current subscription completes at the same time as a new signal is added, the new signal could be subscribed to immediately in addition to the queued up signal that was supposed to be subscribed to next.subscribeToSignalwasn't being cleaned up.subscribeToSignalbeing called after it had been set to nil.