Fix #1484 mutation not calling onSettled if onError throws - #4939

The latest updates on your projects. Learn more about Vercel for Git ↗︎

This pull request is automatically built and testable in CodeSandbox.

To see build info of the built libraries, click here or the icon next to each commit SHA.

Latest deployment of this branch, based on commit 125fc29:

Not sure about your bug reproduction, can't get it to work (filter is not defined). I simplified it a bit:

• mutationFn throws new Error('foo')
• onError throws new Error('bar')

we're winding up correctly in an error state due to the existing finally clause, and the error displayed is 'foo'. This was a conscious decision because when you have two separate errors, we can only "store" one of them, and onError is a method to execute side-effects which mostly shouldn't have any errors. It is not meant to "transform" errors any further.

https://codesandbox.io/p/sandbox/icy-wave-wguwm9?file=%2Fsrc%2Findex.jsx&selection=%5B%7B%22endColumn%22%3A1%2C%22endLineNumber%22%3A11%2C%22startColumn%22%3A1%2C%22startLineNumber%22%3A11%7D%5D

I've tested the link here in incognito and it reproduced the bug correctly, the mutation fails but gets stuck in a loading state, you can see it in this video:
0fd0cd0a-98a7-4879-a200-c49ba13dfb7b.webm

we're winding up correctly in an error state due to the existing finally clause, and the error displayed is 'foo'.

Yes getting to an error state is what should happen, in reality it winds up in a loading state which does not change unless another mutation is triggered

can you please try to make it fail in my reduced example please

can you please try to make it fail in my reduced example please

I've reproduced it in your example but quite inconsistently, we could be dealing with two different bugs here:

1- The one I reported which might only exist in React Query and not TanStack. I strongly believe that might be the case because it's easily reproduced as you can see in the codesandbox I've forked from the gallery and modified, no timeouts needed. Still it might be important to fix it for users that haven't migrated;

2 - The one I reproduced in the first sandbox and again here but with no throws in onError, filter does not exist and trying to parse it inside a JSON on reject seems to cause it to hang in loading state, which does not happen if you define a value to filter or remove the timeout.

but with no throws in onError, filter does not exist and trying to parse it inside a JSON on reject seems to cause it to hang in loading state

yeah I see it now - the promise never gets rejected because accessing filter, which doesn't exist, throws and then promise.reject is never executed, so the promise still hangs in loading state.

if you'd use an async function instead and added some sleep in there instead of creating a new Promise that never resolves or rejects (because of the Error), there wouldn't be a problem.

I'm gonna close this PR because I don't think there is an issue in the current version of Tanstack query.