This site runs best with JavaScript enabled.

Why I avoid nesting closures

Photo by Anna Sjöblom


A little tendency I have to reduce cognitive load for readers of my code.

Watch "Reduce cognitive load for readers of your code by avoiding nesting of closures" on egghead.io

If I come across code like this:

1function getDisplayName({firstName, lastName}) {
2 const upperFirstCharacter = string =>
3 string.slice(0, 1).toUpperCase() + string.slice(1)
4
5 return `${upperFirstCharacter(firstName)} ${upperFirstCharacter(lastName)}`
6}

Chances are, I'll refactor it to this:

1const upperFirstCharacter = string =>
2 string.slice(0, 1).toUpperCase() + string.slice(1)
3
4function getDisplayName({firstName, lastName}) {
5 return `${upperFirstCharacter(firstName)} ${upperFirstCharacter(lastName)}`
6}

I tend to put functions as close to the left side of the screen as reasonably possible. By that I mean, I like to reduce nesting of closures. A closure is what is created when a function is created. It allows the function to access all variables defined external to itself. And that's actually the reason I like to avoid nesting closures: So I don't have to think about as many variables when I'm working in a single function.

In our first code sample (where upperFirstCharacter is nested in getDisplayName), it's possible for my function to access the firstName and lastName variables. This means that while I'm working with it, I'm uncertain whether I need to keep their values in mind. In addition to that, I have to consider that it could access module-level definitions as well (imports/variables/functions).

However, when I move it out (in the second example), I don't have to worry about those variables because it's impossible to access them anyway (I'll probably not even give it a moment's notice or thought). The only thing that function can access is what's defined within it as well as what is defined at the module-level (imports/variables/functions).

In this simple example, it's not a big deal because it's such a small function, but in more complex scenarios the cognitive load can be a problem (and sometimes you even have trouble with variable shadowing which can increase cognitive load as well).

There are other arguments for doing this kind of thing as well:

  1. Performance: not having to re-create the closure every time getDisplayName is called. This argument doesn't really hold water in typical scenarios. Unless you're calling that one billions of times then you're probably fine.
  2. Testing: we could export upperFirstCharacter and test it in isolation. In this situation I wouldn't bother and I'd test getDisplayName instead. But sometimes if the code is complicated this can be useful.

In general, I'm interested in reducing the amount of trivial things my brain has to think about so I can reserve my brain space for more important things and that's my biggest argument for avoiding nesting of closures. That said, I'm not religious about it and don't feel super strongly about this. It's just a tendency I have.

Also, sometimes it's just unavoidable because you really need access to those variables. For example:

1function addThings(additive, ...numbers) {
2 const add = n => n + additive
3 return numbers.map(add)
4}
5
6addThings(3, 1, 2, 3, 4)
7// ↳ [4, 5, 6, 7]

In this case we can't move add out of addThings because it depends on the additive parameter. We could extract add and accept an additional argument and sometimes that can be useful for more complicated functions, but like I said, I'm not religious about the "avoid nesting closures" rule so I think this code is simpler the way that it is now and I'll probably leave it as-is.

I've had some conversations about this on twitter that have been interesting and I think the most interesting insight so far has been from a thread where Lily Scott described a side-effect which Jed Fox summed up well in this tweet:

So yes, this concept has nuance and there are trade-offs. I think I still prefer extracting things because most of the time my files are only a few hundred lines at most, so pulling something out does not increase the number of things that can depend on it and I also feel like it's easier to think about "what can depend on this thing" than "what can affect this thing." But again, it's too nuanced to make a rule for.

Finding ways to offload thinking of trivial things is one skill that I'm always trying to develop (automation plays a big role in that). I really don't want anyone to make an ESLint rule about this idea (please don't do that), but it's something to think about when you're trying to simplify some complicated code. Try pulling things over to the left of the screen a bit. Good luck!

Discuss on TwitterEdit post on GitHub

Share article
loading relevant upcoming workshops...
Kent C. Dodds

Kent C. Dodds is a JavaScript software engineer and teacher. He's taught hundreds of thousands of people how to make the world a better place with quality software development tools and practices. He lives with his wife and four kids in Utah.