React Lifecycle Edge Cases

Dec 13

I have a scratch wrapper I wrote for React as an alternative to setState for controlled components. The idea being that it’s nice to keep all data in Redux, but it’s kind of a pain in the butt to do so manually for every single form, since generally you will also want that data to be cleared from the store when the form is unmounted. My ConnectWithScratch higher order component adds lifecycle hooks on component mount and unmount to create and delete scratch space for the component. As I’ve been using it, though, I’ve been discovering some interesting edge cases.

So my current ConnectWithScratch component calls createScratch in constructor with a user provided scratchKey. (I’m not going to share the current implementation with you because it’s a hairball and I’m still finding edge cases.) It passes an updateScratch prop down to the component to use. And it calls deleteScratch when the component unmounts. In one case I’m using my scratch component for an edit task form, where the scratch key might be something like editTask-114, where 114 is the task id. When I introduced this edit task form on another page, I ran into an interesting issue. The one on the new page has the same key, but when navigating from one page to the other, it tries to create the scratch for the new page before the old one has been removed. Console logging out revealed the following:

1
2
3
4
5
6
7
8
ShowView constructor // the new view is being rendered
ConnectWithScratch constructor:  editTask-114 // new view setting up scratch
creating scratch:  editTask-114 // so far so good...
FocusView componentWillUnmount // and the old view is about to unmount
ConnectWithScratch componentWillUnmount:  editTask-114 // old component scratch unmounting
deleting scratch:  editTask-114 // old component deletes the scratch space!!!
ConnectWithScratch componentDidMount:  editTask-114 // show view scratch mounted
ShowView componentDidMount // and the show view mounts, but has no scratch to work with

So, in short, React starts constructing the next component before the previous one is told it will be unmounted. Which maybe makes sense in terms of how React needs to work to reconcile and see if it can re-use anything. But it creates a pain point here. A few options that come to mind for me:

1) I might be able to move the createScratch call to componentDidMount. This will cause extra renders and maybe performance issues, per their docs, but the user won’t see any intermediary state. It also means we can’t touch the scratch space in the constructor.

2) I could assign each scratch component a unique id instead and use that as the scratch key, or part of it. This loses some of the benefit of the scratch space, though, in that we no longer have a readable key for debugging, nor can we rehydrate components in a straightforward way.

3) I could require the parent component to pass down a part of the scratch key that would make it unique, such as ShowView-editTask-114, FocusView-editTask-114. This maintains readability and rehydratability (sic), but adds additional complexity to actually using the scratch container. Now both the parent and the child need to know about the scratch component.

I don’t love any of these options, though I’m leaning towards option 1 for my first attempt at a fix. Option 2 is my last resort.


Comments

comments powered by Disqus