Managing state in React is difficult. While many suggest that lifting state up to the highest level of the app, and frequently even outside of the component tree, most apps aren't that simple.
One case where this becomes complicated is when app authors need to interact with complex UI components within their application. I internally wrestle with this concept frequently as I build these UI components everyday at work.
One of the most recent cases where I am running into this is our new carousel component which I started work on about 6 months ago as of the time of writing this blog post. This component is perhaps the most complicated UI component we have at Wayfair.
It manages quite a bit of internal state, ranging from actual React state to many references to the rendered dom nodes as well. The problem here is that we also want end users (app developers) the ability to control the carousel from their business logic.
We started with the concept that render props and prop getters would be sufficient to allow end users to control the carousel, however over the months of development, we realized that many (maybe 90%) of frontend developers at Wayfair (and really anywhere) don't want to manage a lot of state and helpers that comes with a render prop.
So we mostly scrapped the render prop pattern on the carousel [1]. Since then we have continued to iterate on the carousel but we were still blocked by this battle of how to properly let end developers control the carousel while also not needing to manage a ton of internal logic (and possibly duplicating that across the codebase).
Skip forward ~ 2 months
On Memorial day weekend of 2018, I decided to spend some time in Codesandbox to work on this problem. One of the rough patterns that I still don't fully grasp from the Downshift component published by Paypal, is the idea of passing back up functions in event handlers to help compute the next state [2].
I started with a simple example, a component that can be both controlled as well as uncontrolled (a counter component). I broke the problem down into two pieces:
- Local state management
- External state management
The common logic between these two pieces is the actual methods behind how to update this state. Locally we call setState with either an object or a function. Using an object for setState generally means that we have all the information we need to determine the next state. Whereas using a function means that we need to know the current state of the world to derive the next state.
As with a counting component, a carousel only really needs to know the current slide, the action the user is taking (i.e. clicking next), and the number of slides to scroll by. With this information (which can be derived by the event as well as the current state and props) we can derive the next active index.
So the solution was a concept I will be calling State Updaters, these are functions which can be (and often are) curried methods that lastly return a function that takes in state and props and returns some new state. They are curried because they need additional information, such as selectors and transformers to return the correct state.
Lets break down this step by step in some code.
So we start off with some local state, a class method for handling the click
event from the button, and an isControlled
method for determining if an
implementer is controlling the component or not.
The next step is to handle the click logic:
Once we have the local handler resolved, now we can dive into the state updater method.
Now that we know what the stateUpdater looks like, we can dive back into the
event handler above. Inside the !this.isControlled()
check, we setState using
the returned function after calling stateUpdater
with undefined (results in
our default argument for transformState, selectState and fieldName).
There is one other thing we need to cover to fully make the UI component controllable:
Now we can put all this together into a working demo, which you can find here on Codesandbox.