Everything You Thought You Knew About React Functional Components Is Wrong

Thank you for coming to React Summit and I appreciate you taking the time to come to my presentation. This presentation just came about because I was writing about React and React Hooks and a lot of people seemed to think that what was going on was very simple and self-explanatory and I don't see it that way so I wanted to create a little more clarity around the subject, not just for everybody else but for myself, and hopefully after you finish this presentation you'll at least have some tools that you can use when things are going wrong in React and you just can't quite figure out why.

So my name is Amy Blankenship and you can find me on LinkedIn at that URL, you can find me on Medium at Amy Blankenship and in fact this presentation originated as an article there. You can find me on GitHub, this is just my personal GitHub stuff, nothing I've done professionally and you can email me if you want to.

So the first thing I think is conceptually the cause of this issue is thinking that functional components are actually a thing that exists and I think that is kind of a mental trap that we should probably dive into a little bit deeper. So let's look at that. When we have a React class component which is how most of us started out doing React, you have React creating a new component. So then it has an instance of that component that it keeps and it runs component did mount, component will receive props, component did update and all of that stuff and finally render. It will continue running that entire life cycle and producing JSX on that instance that actually exists that it has for the lifetime of your component until it's unmounted.

Now let's look at a functional component. The first thing that React does is it takes the props and it passes those props into a function and this function is what we think of as your component and then that component will produce JSX. If the JSX is the same as the JSX from the previous time that it was called, that it called that function, it will re-render it otherwise it doesn't do anything with the result of calling that function. So but at any rate, as soon as it has that JSX, that render of your function is done, you're out of that world. And so yes, it will call your component function again, but it's like an entirely different new thing that is happening.

So let's go a little bit deeper into how creation and destruction and things works in JavaScript so we can really get a handle on what's going on here. So let's start out with a simple case where we create a function that logs a date. So if we have this log date function and every time we run it, it creates a new date, then if you run this function 99 times, you will create a new date. I'm sorry, if you run it 100 times, you will create a new date every one of those 100 times. Now let's look at this a little closer. Let's say that we have a function that's designed to create a new date and we store the result of that function in a variable and we're going to also run it the other 99 times and we're just going to run it. We're just not going to store the date. So we still created 100 dates, but we only keep one of them and we'll talk about what happens to the rest of them later, but in essence the only one that we can do anything with is the one that we stored in the variable. The rest of them just go away.

So now let's look at, we do the same thing, but then we set my date to undefined. So in this case, we created 100 dates and we don't reference any of those dates. So they were created, but they are actually no longer serving any purpose in our program.

Now let's go a little bit further because React now is very skewed towards the functional side. We're going to create a function that creates the date logger that we just looked at and returns the function. So if we run date logger creator 100 times, whether we keep a reference to it or not, we don't create any dates. We create the function, but we don't create any dates. So by running this function, we actually create 100 versions of this function, but no new dates. Now let's change it a little bit again. Now we're going to create our date logger creator so that it creates a new date, and then that is the date that will always be logged by the date logger. So what's going to happen here is we create the date logger 2, and it will create a new date, but every time you run the my date logger, it's going to only log this date.

And if you create 99 more of these, you're creating 99 more dates and 99 more versions of this function. And this is going to be important to come back to later, but because we're not using any of these other 99 functions, they just get thrown away. Now if we then take this one that we stored and we call it 99 or 100 times, then it's going to log the first date that was closed around with this closure 100 times. So we only have one date that we're using here. So what happens to all of those functions and dates and everything that we are creating and we're not using? Well, those things get garbage collection collected. And garbage collection used to be like a much bigger thing because it could like even stop your program from running for a minute while the garbage collection ran. And a lot of people feel like we no longer need to think about garbage collection because it runs in much smaller increments and it's less likely to interrupt your program. But you kind of still need to think about the fact that we're creating all of this stuff that we never use and that's not free in terms of system resources.

Another thing that we need to think about is that when you do store a reference to something, that's going to be kept in memory. And there needs to be like a direct path from window to your variable for it to be kept in memory. If there is no path like that, it will be released. And React handles a lot of that stuff. So if you have things in React variables like useState, useMemo, you can consider that as long as your component is mounted that there is a direct path from window to the variable. Whenever you do an add event listener, for example, in useEffect, that is creating a reference from whatever the component is that you added the event listener to, to the function that you're calling as the parameter, the second parameter on addEventListener. It's really important for you to remove those or you can create memory leaks.

Back to the cases of creating objects and having them get either garbage collected or stored. This is a really, really simple React component where we have a useState and the useState is just a simple true-false toggle. And we're going to create a new date in line. Every time this function runs, that new date will be created. Just like in our previous examples, every time we ran one of these date loggers or created one of these date loggers, it would create a new date, even if we threw away the date logger later.

Then we're going to create a useMemo that is going to store the timestamp every time that toggle changes. And finally, we're just going to have an anonymous function, clickHandler, that is going to change the toggle every time the user clicks the button.

So what's going on here when we set up our component like this? So on the first render, the toggle, the useMemo, stores the time. So what's going to happen is this function that always returns that time is stored somewhere in the bowels of React. And then the renders happen, or the calls of your function happen in between. The toggle hasn't changed, toggle hasn't changed, toggle hasn't changed. And so then when we go to render it later, that is the version of that closure that we are going to use in the render. Does that make sense? I hope so. And that kind of just looks like this in the dev tool, as you can see. The toggle state is here, and then your memo is storing this value of the particular date that was in scope when this closure was created by useMemo is stored in useMemo.

One of the things that we need to think about is the useCallback versus function literal. In this case I used a function literal for two reasons. One because it's a slide, there's not a whole lot of room that useCallback syntax. But the more important reason is it's only going into a button that really doesn't have any bad side effects if you re-render it. However, if you have a callback that's going into something with a giant state tree, every time your component is called it's going to create a new version of this function, just like when we had the date logger creator creating the 99 versions of the function that just got thrown away. And so if you put that in some kind of callback that's going further down the tree and it's an expensive place for that to re-render, you're going to cause unnecessary renders by not using useCallback. And one thing you do not want to do is use useCallback here and then call that useCallback with an anonymous function that you add here because that is going to negate any positive effect that you had by using useCallback. It's still going to create a new instance of that function every time the function is called.

Okay, so now we know that we're actually not really storing state. Instead, we're asking React to create a closure that then returns a version of our previous something that was current during a previous render. So dealing with stale or inconsistent state is something that happens more often than I think most of us would like to. One instance is that calls can complete out of order. So if you look at this particular example, every time the user ID for this component is changed, like for instance this might be contact details for a given user and before the call comes back someone clicks on a different user and now the async function runs again. You haven't completed the first function, the first fetch. The second fetch might come back first and then if you do the setToDos without any modification then you're going to wind up with the wrong user's details.

So this is a pattern that the React team has put forward in the past. I'm not sure if it's best current practice, but it is one way that you can handle this. I know that also they have moved towards not fetching directly in views anymore. So that is another way that you can handle this problem, but it also comes with its set of problems.

One thing that I encountered relatively recently is that I couldn't actually get the most current version of my data that was stored in Redux into the server side data source that was being used by AG Grid. I'm not going to go into a lot of depth of how AG Grid works and why this happened, but the point is that the getRows function needed to have access to the return function of the from doing the fetch call in Redux. So it would have needed to be created as a new...I should have used use callback here, but it would have been needed to be created again once the data came back. And the problem with doing that is that if you create the server side data source again when the getRows now has access to the return value, then it clears out the grid. So that was something that wasn't going to work.

Getting the data outside of my view may have solved one problem, but now it created a different problem in that I could not get the return data into the rows that needed to go into the server side data source. And the way that you can handle this is with useRef. So useRef is kind of an escape hatch around the closure based system in React. If you start with the first version in your getRows of your data, then when the data comes in with the new version of the rows, then you just update ref current to the new rows that you got, and then when the server side data source needs the rows, then you can read the current data from that new version. That mutable ref object, and then everything works as designed.

So sometimes the normal functional system doesn't serve you, and you will occasionally need to use ref current to work around that. Hopefully this talk has been helpful, and you've got some better tools to understand what's going on when your React functional components aren't working exactly the way you thought they would. So it's really important to realize there's more going on with hooks than meets CI. It's really, you need to think in terms of disconnected runs of a specific function rather than a thing that actually exists. React is constantly running that function, and the only thing that you have left over from the previous function is what's stored in the closures. Those closures are things like use state, use memo, and use callback, and they are actually storing a function that returns a particular version of a previous renderer. So be aware of what that renderer is, what the value of the state was when you created the closure. Make sure that you're using the return values of use effect to help make sure that if you've called some kind of data fetching functionality that you're able to ignore it and not update your state in cases where it's been overridden, and also to prevent memory leaks in your use event listeners, I mean add event listeners, sorry about that. And when all else fails and you don't have a great way to get the value from one render to where you need it to be in a later render, you can use use ref because the mutability there means that that object is going to carry through all of the renders, and you can just pull out the most recent version of it. And thank you very much. I really appreciate you coming to my talk.