Weird runtime crash with BindingReducer as a parent and on a child

[ricardomsm]

Hello 👋🏼,

We're using TCA, version 0.50.0, and we're experiencing a weird runtime crash, which much to our surprise and without great understanding only happens on Staging and Release configuration, to note, in theory(in what regards to build settings), Staging is exactly equal to DEBUG.

We have the following parent and child reducers:
Parent

Child

And this is the weird runtime stack trace we get when we dismiss the child one through a viewStore.send in which we catch in the parent and nil out the child:

Through searching and trying out some possible solutions, we found this answered issue #1564 which gave us a hint so we tried these two solutions that solved our problem:

Declaring the case paths explicitly without the use of the / operator prevented the crash,

As well as changing the order of our actions in the Action type, by moving the binding action to be at the bottom, the crash was avoided 🤷🏼‍♂️🥲

Although we "solved" this in our case, we'd like to know if anyone else has encountered a similar issue, or if there's any hint someone can give in regards to what is actually happening behind the scenes.

Thanks in advance 🙏🏼

[stephencelis]

Hey @ricardomsm! You mention this being an issue in 0.50.0. Does this mean the same code paths cause no problem in 0.49?

Also, is there any chance for you to whittle down a reproduction that you can share and we can use to help troubleshoot a fix?

[ricardomsm]

@tgrapperon Indeed adding an indirect case to my enum causes the crash as well, although in a little example I'm trying to craft to expose the issue, I can't reproduce it yet even with the "same" structure in terms of child and parent.

[farvalhoSH]

Hi everyone, I'm a colleague of @ricardomsm and am helping him navigate these waters.
I was following @tgrapperon's suggestion and found that while adding dummy cases and/or making some actions indirect didn't prevent the crash, specifically making the BindingAction case indirect did. I've tested all manner of scenarios in our project and it seems to always work as long as that specific case is indirect.

[tgrapperon]

Ah interesting, I always assumed that having one indirect case was equivalent to mark the whole enum indirect.

[ricardomsm]

@stephencelis We managed to wip up a little demo that has the identic stack trace we're facing.
BindingActionCrashDemo.zip

[stephencelis]

This is a Swift bug around existentials and enum layouts:

#1881

We can work around it in the library, luckily, but thanks for the repro and catch!