Alex / KATT

> There's something odd happening here, Undici doesn't have any constraints for forbidden headers in versions >=5.4.0 (see [docs](https://github.com/nodejs/undici#forbidden-and-safelisted-header-names)). And it never errored on them, it only filtered them out....

cc @sachinraja @3x071c It seems like the *only* thing we need to exclude is the `connection` header. See here for a fix in a example project: https://github.com/KATT/unidici-fetch-reserved-header-crap/pull/1

Closing because of the many conflicts it has, feel free to pick it up though

I am tentatively positive to this suggestion, but we should not consider it until we have an alpha of the next major. The next major's design would probably allow for...

With the [new major](https://github.com/trpc/trpc/blob/next/.tmp/v10-docs.md) we could potentially add a new method called `.inputCallback` or similar that could take whatever `input` takes but as a callback instead. 🤔

@sachinraja - a penny for your thoughts; do you think there's a way to support both? We could change the default behavior of the parser callback from `inputParser( input: unknown...

So usage could maybe be something like ```ts const router = t.router({ hello: t.procedure.input(z.string()).query(() => '...'), // kinda as above, but with a callback and we'll lose how to infer...

> Alternatively, if the input parser is called "late" (i.e. inside any middleware callback), potentially there could just be a recipe involving `async_hooks` and people could just do [...] It...

@zomars, this is awesome, but if you do, use the v10 alpha so it doesn't become obsolete straightaway ☺

> I put together a stripped down example using NX / tRPC v10 / Prisma > > https://github.com/nowlena/nx-trpc-test @nowlena please add to https://trpc.io/docs/awesome-trpc 🙏