Async functions are great, especially if you have to call multiple functions in a row that return promises. With async / await, code becomes easier to reason about for humans, because the data flow mimics synchronous code, which is what we are used to reading.
So what are async functions exactly?
Syntactic sugar
When I first learned about async functions, the following sentence stuck with me:
Async / await is "just" syntactic sugar for promise chaining
β Someone, somewhen
This is mostly true, and if that's your mental model about async functions, it will get you quite far. To re-iterate, let's take an example and refactor it from promise chaining to an async function:
function fetchTodos() {
return fetch('/todos')
.then(response => response.json())
.then(json => json.data)
}
So far, so good. Nothing too difficult here, just our normal data fetching and extracting (error handling left out intentionally here). Still, even with this example, the callbacks are not so easy to read, so how would this look with an async function?
async function fetchTodos() {
const response = await fetch('/todos')
const json = await response.json()
return json.data
}
Ah, I believe that reads a lot better, because you can actually see where we are assigning variables to and what will be the final return value of that function.
So, if that is a good mental model for async functions, what's the problem with the above definition? Well, it's just not everything. There are a couple of subtle difference between promise chaining and async functions that I learned the hard way. Let's go through them:
They always return a promise
This is actually the defining trait of an async function. No matter what you do, it will always return a promise, even if you don't explicitly return one:
async function fetchRandom() {
// β
this will return `Promise<number>`
return Math.random()
}
This is necessary because you can use the await keyword in async functions, and once you do that, you enter promise-land, in which there is no escaping from. If code is async, you can't turn it back to sync code. I was personally quite confused by this, because in scala, Await.result actually takes a promise, blocks the execution for a certain amount of time and then lets you continue synchronously with the resolved value.
In JavaScript however, an async function will stay asynchronous, so the return value must be a promise, and the language construct makes sure of this out of the box. This brings us to the next point:
It transforms thrown Errors into rejected promises
You might have seen this in example code involving the fetch API, as fetch will not automatically give you a failed promise on erroneous status codes like other libraries, e.g. axios, do. To get to a failed promise, you just throw an Error (or anything, really), which will then be transformed into a failed promise. This is happening because, again, an async function always needs to return a promise:
async function fetchTodos() {
const response = await fetch('/todos')
if (!response.ok) {
// β
this will become a failed promise
throw new Error('Network response was not ok')
}
return response.json()
}
Now the same works if you are in a promise chain, but not if you are outside of it. Suppose you want to do some parameter validation and decide to throw an Error if the input is invalid in a non-async function:
function fetchTodo(id: number | undefined) {
if (!id) {
// π¨ this will NOT give you a failed promise
throw new Error("expected id")
}
return fetch('/todos')
.then(response => response.json())
.then(json => json.data)
}
If you make the same function async, it would give you a failed promise. These little nuances can be quite confusing, so I prefer to explicitly work with Promise.reject no matter which context I'm in:
function fetchTodo(id: number | undefined) {
if (!id) {
// β
this will work as expected, no matter where
return Promise.reject(new Error("expected id"))
}
return fetch('/todos')
.then(response => response.json())
.then(json => json.data)
}
They always return a new promise
I first stumbled upon this when working with query cancellation in react-query. Here, react-query wants us to attach a .cancel
method on our resulting promise. Surprisingly, this doesn't quite work in async functions:
async function fetchTodos() {
const controller = new AbortController()
const signal = controller.signal
const promise = fetch('/todos', {
signal,
})
promise.cancel = () => controller.abort()
// π¨ This will be a new promise without the cancel method!
return promise
}
Because we are in an async function, a new promise will be returned at the end of it, even if we already return a promise ourselves! Here is a great article if you want to see how query cancellation can work even with async functions.
Handling errors
The default way of handling errors in async functions is with try / catch, which I don't like very much, mainly because the scope of try / catches seems to get very large. If additional, synchronous code happens after the async operation that might fail, we are likely still treating it as if the fetch failed:
const fetchTodos = async (): Promise<Todos | undefined> => {
try {
const response = await axios.get('/todos')
// π¨ if tranform fails, we will catch it and show a toast :(
return transform(response.data)
} catch (error) {
showToast("Fetch failed: " + error.message)
return undefined
}
}
Sometimes, we even silently catch and discard the error, which will make debugging very hard.
So if you also think that async / await is cool, but try / catch is not, you can try combining async functions with "traditional" catch methods:
const fetchTodos = async (): Promise<Todos | undefined> => {
const response = await axios.get('/todos').catch(error => {
// π showing the toast is scoped to catching the response error
showToast("Fetch failed: " + error.message)
return undefined
})
return transform(response?.data)
}
In summary
I hope this gives you a bit of a deeper understanding of what async / await is doing under the hood. I have seen lots of code where the async keyword is just stuck on a function for no good reason, so lastly, here are some examples of patterns that I think should be avoided:
// π¨ the async keyword doesn't do anything -
// except creating a new unneccessary promise
const fetchTodos = async () => axios.get('/todos')
const fetchTodos = async () => {
const response = await axios.get('/todos')
// β οΈ awaiting a non-promise is possible, but doesn't do anything
return await response.data
}
// π I don't even π€·ββοΈ
const fetchTodos = async () =>
await axios.get('/todos').then(response => response.data)
const fetchTodos = async () => {
try {
// π¨ await is redundant here, too
return await axios.get('/todos')
} catch (error) {
// π¨ the catch-and-throw is totally unnecessary
throw error
}
}
That's it for today. Feel free to reach out to me on twitter
if you have any questions, or just leave a comment below β¬οΈ
Top comments (16)
I thought I had a pretty good handle on async JS, but the 'I don't even...' example at the end has me a little confused. Could you please detail the antipattern you're illustrating there?
I've even been using promises with arrays, async mapping and reducing to my heart's content... but just because it 'works' doesn't mean I necessarily understand the nuances of HOW.
(self taught, JS first language outside of bash)
Taking an early guess, is it the combo of async/await and .then()? That doesn't feel quite right tho, because await only works in async functions so I imagine you have to wrap the .... oh wait, my logic only makes sense if the callback inside .then() is also async. The resolved value of the getter is equivalent to the parameter passed to the callback, unless that callback returns a promise (which, in that case, is what you'd be awaiting)
it's multiple things here really:
1) combining async/await with .then(). I think you'd want to choose one way and then stick to it
2) awaiting the last (and in this case only) Promise is unnecessary. You can just return the Promise, instead of awaiting it and then returning a new Promise due to the nature of an async function.
3) Since the await is unnecessary (see point 2), unless you get rid of the the .then() chain (see point 1), the function being async is also unnecessary.
All in all, that combination is just unnecessarily verbose and shows that whoever has written it doesn't understand what async functions are really doing :)
I'll be doing a lot more reading up on all of this now that I think I'm scraping at what you're saying...
My code has consistently worked up to a recent project - I wrote a function to update state which returns successfully but doesn't produce the desired side-effects. I'm almost certain now it has to do with an unresolved promise somewhere.
I actually scrapped it and chose a different approach, fearing it might be an API bug with the component since I had successfully implemented a similar solution elsewhere in my app.
Thanks for your post and reply - I'm just now getting to the level of confidence to even reach out to other developers, so I really appreciate your thoughtful response!
EDIT: I took one 'await' out of my init function just to see if I was understanding things right - and it still works, but faster by a factor of 3! Thanks again :D
Just saw your edit and thatβs incredible π
Actually in your first example I find the
then()
based approach much more readable than theasync/await
approach, because the data is being chained or piped through the lines. I also think about it as a more functional way, whilstasync/await
is the procedural way.I wouldn't always prefer
then()
overasync/await
. The latter is especially more suitable when there's just a single promise.Yep, it's definitely a stylistic thing from time to time. To be honest, if you use TypeScript, the types flow through
.then
chains very nicely as well, so I don't really have a problem with that. I just don't really like that people sometimes stickasync
on a function for no apparent reason, possibly without knowing what it does and what it's for, so that was my main motivation to write this blog post :)I think important to add that async / await doesnβt work inside Array built in iterator functions, like .forEeach.
What is really bad that this behavior often not mentioned in tutorials, not get caught by linters, and didnβt throw error.
Good news is that there is a way to make it work, but itβs a topic for another article:)
good point. Am I right in thinking that it works with awaiting Promise.all of the result array, if used with
.map
instead of with.forEach
?Be careful when using Promise.all because if one Promise fails in the Promises set, the others will still run.
I think that your example after "If you make the same function async, it would give you a failed promise." should have async in code. Both examples are the same.
The next example replaces throw with Promise.reject, both in non-async functions on purpose. The takeaway is that throw is only consistently transformed to failed promises in async functions, which is a detail that is easy to miss. Promise.reject is more explicit and works everywhere.
That's right, sorry didn't nice that. I was looking at missing async keyword. Good article.
Or .reduce, starting with Promise.resolve().
The issue is not complicated, easy to find a few good solutions on StackOverflow.
Worth just pointing out that
for...of
andreduce
cause the promises to be started in sequence where theall
variants start them without waiting for the earlier ones to finish.My 2 cents:
for...of
lets you chain promises and terminate early, that's my way to go if that's my need or if the reduce function would be long, as I find that more readable in this circumstance.Thank you, happy that you like it :)