Migrate to cats-effect 3 & http4s 0.23

[adamw]

See: #1050

[adamw]

@catostrophe if you'd have a better idea how to handle this case I'd be happy to take hints :) Here I've get a list of raw headers and I want to send them back (and only the headers). I couldn't get it to work without the (otherwise useless) map with a cast (triggering the implicit conversion), but it's probably because of friday evening not because it's necessary ;)

[catostrophe]

    case r @ GET -> Root / "api" / "echo" / "headers" =>
      val headers = r.headers.headers.map(h => h.copy(value = h.value.reverse))
      val filteredHeaders: Header.ToRaw = r.headers.headers.find(_.name == CIString("Cookie")) match {
        case Some(c) => headers.filter(_.name == CIString("Cookie")) :+ Header.Raw(CIString("Set-Cookie"), c.value.reverse)
        case None    => headers
      }
      okOnlyHeaders(List(filteredHeaders))
    case r @ GET -> Root / "api" / "echo" / "param-to-header" =>
      okOnlyHeaders(r.uri.multiParams.getOrElse("qq", Nil).reverse.map("hh" -> _: Header.ToRaw))
    case r @ GET -> Root / "api" / "echo" / "param-to-upper-header" =>
      okOnlyHeaders(r.uri.multiParams.map { case (k, v) =>
        k -> v.headOption.getOrElse("?"): Header.ToRaw
      }.toSeq)

[catostrophe]

I would write it this way

[adamw]

Thanks, much better :)

[catostrophe]

val headers = Headers(response.headers.map(h => h.name -> h.value))

[adamw]

I've got a problem converting the finatra-cats interpreter. It has the following:

FinatraServerInterpreter.toRoute(e.endpoint.serverLogic(i => eff.toIO(e.logic(new CatsMonadError)(i)).to[Rerunnable].run))

specifically, there's eff.toIO which enables to run the effect to a future. Is there something similar in ce3? That is, a way to run an arbitrary Async effect? Any typeclass to require?

[catostrophe]

I've got a problem converting the finatra-cats interpreter. It has the following:

FinatraServerInterpreter.toRoute(e.endpoint.serverLogic(i => eff.toIO(e.logic(new CatsMonadError)(i)).to[Rerunnable].run))

specifically, there's eff.toIO which enables to run the effect to a future. Is there something similar in ce3? That is, a way to run an arbitrary Async effect? Any typeclass to require?

@adamw I see. I will provide a solution

[andreamarcolin]

Is there something similar in ce3? That is, a way to run an arbitrary Async effect? Any typeclass to require?

@adamw I guess Dispatcher could be what you need here? https://typelevel.org/cats-effect/docs/std/dispatcher

[adamw]

@andreamarcolin Ah exactly, thanks! :)

[catostrophe]

@adamw things you should remember:

• Dispatcher is a resource. When you release it, it cancels all the spawned fibers (futures)
• you shouldn't pass it implicitly (it is not encouraged by the ce3 authors)
• you probably need to create one at first, and then create all the finatra routes within its use scope, i.e. it's a global resource with a wide scope, like the http server itself
• once you enter the use scope, the current runtime context is captured by the dispatcher, and all the further evalOns don't affect the ExecutionContext of the tasks run by the dispatcher

[catostrophe]

You also need a proper conversion to twitter's Future. catbird still doesn't have a ce3 compatible release

[catostrophe]

catbird is a blocker for this PR.
see: typelevel/catbird#267 - WIP for CE3-compatible release

[adamw]

@catostrophe thanks for the pointers. I think @kubinio123 migrated the code as far as possible, with the exception of the finatra-cats module.

[brbrown25]

@adamw I wanted to check in and see what the status of this pr is? Based on the comments it looks like this might be held up by catbird?

[adamw]

@brbrown25 yes, that's the only missing component. Another is the fact that we'd have to switch from "stable" (as long as you can call 0.21 stable) http4s releases to 1.x milestone releases. My current plan (which isn't set in stone) is to release tapir 0.18 using http4s 0.21, and then switch to http4s 1.x.

[brbrown25]

@adamw thanks for the quick reply! that sounds like a reasonable plan, I'll keep an eye on this one. have a great day and thanks for some awesome repositories!

[kubukoz]

Small note, PR title should say 0.23 not 1.0 :)

[adamw]

@kubukoz true, fixed :)

now the tests which started taking 3h instead of 40m ;)

[calvinlfer]

now the tests which started taking 3h instead of 40m ;)

🤣🙀

[adamw]

Yeah I have yet no idea what's going on, only a slight lead that there's some weird interaction between ce3 and netty. In tapir, the modules which take the longest to test (after the upgrade) are vertx and ziohttp, both of which are netty-based; while the test framework uses ce3. Similarly, in sttp-client, async-http-client backends in the fs2 flavor sometimes take a long time to test (and sometimes not). Maybe we're running something on the wrong thread, and hence the problems, but I didn't find anything specific yet :)

[adamw]

Luckily changing the http client to run the tests from ahc-fs2 to httpclient-fs2 solved, or rather minimized, the problem. The issue still exists in sttp-client, and the vertx tests in tapir are still suspiciously slow. But at least the build passes, so we can proceed.