Oookay I just took a look at this, and it's due to the new `assets_navigation_prefers_asset_serving`

Oookay I just took a look at this, and it's due to the new

assets_navigation_prefers_asset_serving

assets_navigation_prefers_asset_serving

assets_navigation_prefers_asset_serving

assets_navigation_prefers_asset_serving compat flag that was enabled by default on April 1st. Now, if requests include the

Sec-Fetch-Mode: navigate

Sec-Fetch-Mode: navigate

Sec-Fetch-Mode: navigate

Sec-Fetch-Mode: navigate header, which happens on page navigation, and in SPA mode, the

index.html

index.html

404.html

404.html

is served, instead of the Worker being invoked.

If you open your browser console and run

await (await fetch('/api/hello')).json()

await (await fetch('/api/hello')).json()

await (await fetch('/api/hello')).json()

await (await fetch('/api/hello')).json(), you'll see the expected response, and the Worker get invoked.

JamesOP•4/5/25, 9:10 PM

This is... very confusing behaviour imo. cc @GregBrimble @Sunil Pai

JamesOP•4/5/25, 9:13 PM

As a workaround if you wanted this to behave as you expect, add assets_navigation_has_no_effectassets_navigation_has_no_effect to your compatibility_flagscompatibility_flags

Silvan•4/5/25, 9:14 PM

funny enough i had this issue before that flag was introduced.
unless im doing something wrong but from my understanding 404 should just be redirected, which dosent seem to be the case?

JamesOP•4/5/25, 9:16 PM

That looks like it is invoking your worker, but you're expecting it not to?

JJames That looks like it _is_ invoking your worker, but you're expecting it not to?

Silvan•4/5/25, 9:18 PM

correct
https://developers.cloudflare.com/workers/static-assets/#routing-behavior

You can configure and override this default routing behaviour. For example, if you have a Single Page Application and want to serve index.html for all unmatched routes, you can set not_found_handling = "single-page-application":

You can configure and override this default routing behaviour. For example, if you have a Single Page Application and want to serve index.html for all unmatched routes, you can set not_found_handling = "single-page-application":

if i remove the catchAll/notFound from my worker i get no response at all

SSilvan correct https://developers.cloudflare.com/workers/static-assets/#routing-behavio...

JamesOP•4/5/25, 9:19 PM

Do you have a project I can clone to reproduce? I'm not able to with the starter one from C3 + React: https://i.james.pub/file/2025/04/1c0c5551-57fe-4d34-b070-156843babb6e.png

Silvan•4/5/25, 9:20 PM

unless this might be an issue with hono and using something like

app.use('*',...)

app.use('*',...)

Silvan•4/5/25, 9:20 PM

i will to make a repro!

JamesOP•4/5/25, 9:20 PM

thanks, happy to take a look

JJames This is... very confusing behaviour imo. cc @GregBrimble @Sunil Pai

JamesOP•4/5/25, 9:24 PM

Yeah this is really unintuitive. I completely understand technically why it works the way it does - it's very "smart". But this isn't intuitive to how most people treat the web and their applications - people don't expect different behaviour with a

fetch

fetch

to a URL vs hitting it in their browser.

JJames Do you have a project I can clone to reproduce? I'm not able to with the starter...

Silvan•4/5/25, 9:26 PM

to be 100% sure here. my worker is supposed to programtically return a 404 right?
like in the template as in return new Response(null, { status: 404 });return new Response(null, { status: 404 });

JamesOP•4/5/25, 9:27 PM

in most cases the worker is only hit if there's no asset match

JamesOP•4/5/25, 9:27 PM

SPA mode adds a few more conditions

JamesOP•4/5/25, 9:27 PM

https://developers.cloudflare.com/workers/static-assets/routing/

JJames in most cases the worker is only hit if there's no asset match

Silvan•4/5/25, 9:30 PM

yea this is where the wording gets confusing
example.com > obviously index.html (routing works fine)
example.com/some/sub/page > no asset match and no worker route match? so it should also go back to index.html

talking about pressing refresh or directly navigating to the page with a URL

JamesOP•4/5/25, 9:31 PM

the worker "route" will never be checked and then go back to assets, unless you're doing that in your own code. It'll check for exact match, handle spa behaviour if needed, and then fallback to your user worker as the final step

JamesOP•4/5/25, 9:31 PM

I honestly couldn't tell you exactly what the logic is for how that works with SPA mode

JamesOP•4/5/25, 9:33 PM

we're talking about 2 different questions/issues here though for sure. Your's is slightly different to what @matthewlsessions is observing, which is the weird

assets_navigation_prefers_asset_serving

assets_navigation_prefers_asset_serving

behaviour

JJames Oookay I just took a look at this, and it's due to the new `assets_navigation_pr...

GregBrimble•4/5/25, 9:34 PM

That’s correct, yes. I’m working on some more comprehensive docs at the moment (sorry for the delay in getting docs published before this change went live), but you can expect them to be live early next week. In the meantime, as @James says, you can add the assets_navigation_has_no_effectassets_navigation_has_no_effect flag if you wish to opt out. For most people though, we expect this change to be a welcome one. In traditional SPA setups, this will reduce the number of billable requests that you see in your application, without affecting functionality.

GregBrimble•4/5/25, 9:35 PM

Thanks, as ever, for fielding questions, @James !

GregBrimble•4/5/25, 9:36 PM

Definitely don’t dispute that it’s potentially confusing though! Hopefully updated docs help to clarify.

JamesOP•4/5/25, 9:37 PM

Yeah I think once people understand how it works, it might be welcomed, but for folks new to the platform or the web in general, it's very unintuitive. Expecting different responses when hitting a route in your browser vs

fetch

fetch

took me some time to figure out and I follow the platform very closely - without something like debug logs from

wrangler

wrangler

, I can imgine people are going to hit this very frequently.

JamesOP•4/5/25, 9:38 PM

I don't think anyone really expects to see a different response in their browser when they hit /api/foo/api/foo vs when they

fetch

fetch

it, especially when there's no code in the worker that explicitly handles this. As standard and completely transparent behaviour, that's just weird.

GregBrimble•4/5/25, 9:39 PM

Yup, totally understand. And I think your idea of more comprehensive logging is absolutely a good one. If you’ve got specific suggestions here, I’m all ears. Otherwise, I’ll take a stab at something myself next week and see what we can do to make the request flows more obvious in

wrangler dev

wrangler dev

JJames Do you have a project I can clone to reproduce? I'm not able to with the starter...

Silvan•4/5/25, 9:42 PM

https://github.com/Sillyvan/asset_handling_spa

GGregBrimble Yup, totally understand. And I think your idea of more comprehensive logging is ...

JamesOP•4/5/25, 9:42 PM

I think if there was a log that very clearly stated GET /api/foo - Received Sec-Fetch-Mode: navigate, serving <>GET /api/foo - Received Sec-Fetch-Mode: navigate, serving <> or words to that effect, it'd help a lot with the confusion.

Isaac McFadyen•4/5/25, 9:46 PM

^ and would be immediately visible if users post their logs in here and we need to help them troubleshoot.

GregBrimble•4/5/25, 9:48 PM

Logic is:

Check if request “exactly matches” an asset. If yes, serve.
If no, but request is a navigation request, and new compat flag is enabled and assets.not_found_handlingassets.not_found_handling is something other than “none”, that fallback is served (index.html for SPA, 404.html for 404-page).
If not navigation request or flag is disabled, worker script is invoked if present. Since this is now happening after the fallback serving in typical SPA modes, an assets binding and final env.ASSETS.fetch(request)env.ASSETS.fetch(request) will likely no longer be necessary in most applications. They can now just return 404s or whatever.
Finally, if the worker script either isn’t present or does env.ASSETS.fetch(request)env.ASSETS.fetch(request), a response according to assets.not_found_handlingassets.not_found_handling is returned.

GregBrimble•4/5/25, 9:48 PM

I’m on mobile right now, so can’t try to repro, but if something isn’t following that, I’m pretty sure it’s a legit bug.

GregBrimble•4/5/25, 9:49 PM

“Exactly matches” is in quotes to hand wave over HTML handling.

GregBrimble•4/5/25, 9:51 PM

Again, this’ll all be documented more formally next week. Apologies for this rough first draft

JamesOP•4/5/25, 9:58 PM

Thanks. When you're next available, taking a look at the example @Silvan posted in https://canary.discord.com/channels/595317990191398933/1358194995076796507/1358195325306802306 might help a bit. I think what's happening is technically expected, but not what I would expect intuitively. I couldn't tell you why the SPA behaviour isn't working for them at all on a compat date prior to April 1st - I must be missing something.

JamesOP•4/5/25, 10:00 PM

In the future, it might be preferrable to not enable things like this so quickly by default when the documentation/education/logs around it aren't complete yet

BumblebeeSquare•4/5/25, 10:00 PM

not sure where to post this but please...

can we please allow upload .env directly for build variables ? it is annoying we have to upload one by one thorugh the ui ( regular env var seems like support pasting dotenv) - when using sveltekit we have to provide them for build .. and i think in pages, all env variables are also used as build variables
is there an max limit for build variable for workers ? once my build var reach 50+ i got Initializing build environment... internal error occurred. Please retry your build. If this problem persists, contact Cloudflare support.

JJames Thanks. When you're next available, taking a look at the example @Silvan posted ...

Silvan•4/5/25, 10:02 PM

technically it makes 100% sense else i dont get how this would even work
but this needs to be documented very clearly in 1 place else its very hard to find the connection between all these things

JamesOP•4/5/25, 10:10 PM

If I'm reading correctly, it looks like the

404.html

404.html

traverses up to find one, but

index.html

index.html

doesn't in SPA mode, which might explain your weirdness @Silvan: https://github.com/cloudflare/workers-sdk/blob/d4c11710fd36286be8587379d659e19e91778777/packages/workers-shared/asset-worker/src/handler.ts#L867

JamesOP•4/5/25, 10:11 PM

I've not tested and haven't looked that closely at this code, but that looks like what it's doing

zwily•4/6/25, 3:57 AM

i reran npx wrangler types to generate a new worker-configuration.d.ts, and instead of doing this:
interface Env {
...
}

it did:
declare namespace Cloudflare {
interface Env {
...
}
}
interface Env extends Cloudflare.Env {}

and now my env doesn't work anywhere... (context.cloudflare.env doesn't have any of my types on it) Has anyone run into this or know how to fix it? (this is in a remix app, but I've updated to wrangler 4 and been using it fine for a bit...)

IG•4/8/25, 12:44 PM

I'm having the exact problem. Spent an hour reading docs, forums, but didn't find any solution. Came here (discord) in hopes of fix, but seems you didn't get any help either. Not a good first impression for this tool..