I'm using promises and have code that looks like the following:
function getStuff() {
return fetchStuff().then(stuff =>
process(stuff)
).catch(err => {
console.error(err);
});
}
Or:
async function getStuff() {
try {
const stuff = await fetchStuff();
return process(stuff);
} catch (err) {
console.error(err);
}
}
I was doing this to avoid missing on errors but was told by a fellow user that I shouldn't be doing this and it is frowned upon.
- What's wrong with
return ….catch(err => console.error(err))
? - I've seen a lot of code that does this, why?
- What should I do instead?
error is much too generic, it is a catch all but there are only so many things that the operation would fail with, error is everything errorSomethingSpecific gives granularity
It returns a promise that will fulfill with
undefined
after you handled the error.On it's own, catching errors and logging them is fine at the end of a promise chain:
However if
getStuff()
does catch the error itself to log it and do nothing else to handle it like providing a sensible fallback result, it leads toundefined
showing up in the page instead of "Sorry".Historically, people were afraid of promise errors being handled nowhere which lead to them disappearing altogether - being "swallowed" by the promise. So they added
.catch(console.error)
in every function to make sure that they'd notice errors in the console.This is no longer necessary as all modern promise implementation can detect unhandled promises rejections and will fire warnings on the console.
Of course it's still necessary (or at least good practice, even if you don't expect anything to fail) to catch errors at the end of the promise chain (when you don't further return a promise).
In functions that
return
a promise to their caller, don't log errors and swallow them by doing that. Just return the promise so that the caller can catch the rejection and handle the error appropriately (by logging or anything).This also simplifies code a great lot:
If you insist on doing something with the rejection reason (logging, amending info), make sure to re-throw an error:
Same if you are handling some of the errors:
The reason you should not
catch
errors unless absolutely required (which is never) is thatImplications
Once an error is caught by a
catch
handler, it is considered as done and handled. All the successive promise subscribers in the promise chain will call their success handlers instead of failure or catch handlers. This leads to weird behaviours. This is never the intended code flow.If a function at lower level like a service method (
getStuff
) handles errors incatch
, it breaks the principle of Separation of Concerns. A responsibility of service handler should be solely to fetch data. When that data call fails, the application who is calling that service handler should manage the error.The catching of error in some function being caught by another one, results in weird behaviours all around and makes it really hard to track root causes of bugs. To track such bugs we have to enable the
Break on Caught Exceptions
in the Chrome dev console which will break on everycatch
and could take hours at an end to debug.It is always a good practice to handle promise rejections but We should always do that using
failure
handler overcatch
handler. A failure handler will only catchPromise rejections
and lets the application break, if any JS error occurs, which is how it should be.Why does old code do this?
Historically, older (pre 2013) promise libraries 'swallowed' unhandled promise rejections you have not handled yourself. This has not been the case in anything written since then.
What happens today?
Browsers and Node.js already automatically log uncaught promise rejections or have behaviour for handling them and will log them automatically.
Moreover - by adding the
.catch
you are signalling to the method calling the function thatundefined
is returned:The question one should be asking oneself when writing async code is usually "what would the synchronous version of the code do?":
I don't think a consumer would expect this function to return
undefined
if an error occurs.So to sum things up - it is frowned upon because it surprises developers in the application flow and browsers log uncaught promise errors anyway today.
What to do instead:
Nothing. That's the beauty of it - if you wrote:
In the first place you would get better errors all around anyway :)
What to do if I'm running an old version of Node.js?
Old versions of Node.js might not log errors or show a deprecation warning. In those versions you can use
console.error
(or proper logging instrumentation) globally:The most general statement here, that applies in languages beyond javascript, is don't 'catch' an error unless you plan to 'handle' the error. Logging is not handling.
i.e. In general, the best (only?) reason for a catch is to handle/'deal with' the error in a constructive way that allows the code to carry on without any further problems. And again, a line of logging probably never achieves that...