Open PaulWild opened 5 months ago
Interesting... so the problem is not as phrased by the title, thought the problem exists.
One way to see the issue is with the devtools network, and preserve logs, but another is to modify the api endpoint to:
import { NextRequest, NextResponse } from 'next/server';
export const GET = (req: NextRequest) => {
console.log('redirect called', req.nextUrl.searchParams.get('_rsc'));
return NextResponse.json({ hello: 'there' });
};
And now you see, where this is going. When you have the loading
file, Next.js serves that suspense boundary to the client, so Loading
is seen by the client.
Then the framework is making a request for _rsc data to the endpoint, not that this endpoint serves 200 + json, an at the same time the framework is making a request to the endpoint, without the _rsc query.
I'd seem that this PR, https://github.com/vercel/next.js/pull/63786, referenced in #57257 should indeed take care of this 🤔
ah nice, what would be a more appropriate title?
I think the issue is still, a Double redirect when using a loading boundary and Route Handlers
From the test I see that, the intention had been to redirect to another page, not a Route Handler.
Though I am not sure I have the time to investigate a whole lot, but the issue is interesting to say the least.
Thanks hopefully a more appropriate title will help attract some more interest, it is slightly different as this issue is present if you redirect to another Page and a route handler. My example was a route handler as that is a cut down version of our use case (We need to do some cookie clean up and that was the only way we could think in the flow we have).
There is a work around that if the redirect is down in a child client component then you do not get the double render. However this is a bit messy and is not ideal
Alright, I see, the odd thing is that the test introduced by #63786 should really be guarding against that... I'll try to run the tests on this machine, to see what's up
Ok, I got them tests running here, and isolated to the test introduced in #63786 plus a router handler test redirect I wrote, and I can see that, the test is making sure the UI doesn't change, because the redirected to page (or my route handler), render with Date.now, so the assumption is made, should you show two different dates to the client, then you definitely did a double redirect, but that's not the entire picture.
I can see that the server makes a request with _rsc
, and one without it, where it'd seem that the UI ignores one, and uses the other, and that makes the test assumption pass.
I'll see if I can change the check to make the test fail.
We're experiencing the very same issue, which seems to have been introduced in 14.2.2
or 14.2.3
.
We temporarily downgraded to 14.2.1
, which doesn't show this weird behavior. For our use case it is critical, as the redirect involves a third party data generation API which doesn't allow the same url (incl. query params) to be called twice.
The same issue is present in 15.0.1 as well.
I have a client component with a Link (prefetch=false) that makes a call to an API route when the user clicks the button. The API route returns a downloadable image/png. The API call is being made twice, one after another. The first call has the _rsc parameter.
Subsequent clicks on the Link invoke the URL with a _rsc parameter. The browser seems to ignore the return result whenever _rsc is set.
Link to the code that reproduces this issue
https://github.com/PaulWild/multiple-redirects-in-rsc/
To Reproduce
/
/api/redirect
Current vs. Expected behaviour
The api route should only be called once
Provide environment information
Which area(s) are affected? (Select all that apply)
Navigation
Which stage(s) are affected? (Select all that apply)
next dev (local), Vercel (Deployed)
Additional context
This doesn't happen if there aren't any loading pages. This seems to be related to https://github.com/vercel/next.js/issues/57257