Upgrade core to cats-effect-3 #3784

domaspoliakas · 2020-10-22T22:19:39Z

Hello.

Took me a little bit of time to get my bearings with regards to figuring out the right artifacts to import and such. For now this is a pretty simple set of changes just to make everything compile.

I got rid of AsyncSyntax; I don't think it made any sense to keep it.
Removed all mention of Blocker and ContextShift (naturally)
There's a couple of instances where unsafeRunSync was done where I used the global IORuntime. I'm not sure if that's the right thing to do. If that could be avoided core could probably import just kernel and std.
I don't know if I used the dispatcher correctly. That is, I added Dispatcher as an implicit in a couple of places, which I'm concerned about since it might have a knock-on effect and I don't know if that's desired.

I was primarily focused on getting this to compile and trying to understand how to migrate without changes, but now that it compiles I think that some places where Sync is used could be loosened due to the presence of Files.

Anyway, let me know how I could improve this. I will try to run tests tomorrow.

SystemFw · 2020-10-22T22:23:50Z

core/src/main/scala/org/http4s/EntityDecoder.scala

-  def textFile[F[_]](file: File, blocker: Blocker)(implicit
-      F: Sync[F],
-      cs: ContextShift[F]): EntityDecoder[F, File] =
+  def textFile[F[_]: Files](file: File)(implicit F: Sync[F]): EntityDecoder[F, File] =


why Sync here? Also if Sync is unavoidable, you can remove Files from there

~~There's a single call that needs to be suspended downstream. I'll try to see if I can eliminate that.~~
Nevermind, I was thinking of a different place. This can be changed to Concurrent.

SystemFw · 2020-10-22T22:29:27Z

core/src/main/scala/org/http4s/Message.scala

@@ -186,6 +186,7 @@ sealed trait Message[F[_]] extends Media[F] { self =>
 object Message {
  private[http4s] val logger = getLogger
  object Keys {
+    import cats.effect.unsafe.implicits.global


we can avoid this by using SyncIO for this

SystemFw · 2020-10-22T22:29:35Z

core/src/main/scala/org/http4s/Message.scala

@@ -495,6 +496,7 @@ object Request {
  final case class Connection(local: InetSocketAddress, remote: InetSocketAddress, secure: Boolean)

  object Keys {
+    import cats.effect.unsafe.implicits.global


SystemFw · 2020-10-22T22:30:52Z

core/src/main/scala/org/http4s/StaticFile.scala

@@ -25,15 +26,13 @@ object StaticFile {

  val DefaultBufferSize = 10240

-  def fromString[F[_]: Sync: ContextShift](
+  def fromString[F[_]: Files: Sync](


you don't want Files: Sync. It it works with just Files plus non effect capturing classes (e.g. Monad or Concurrent), then great, if not Sync is enough.

SystemFw · 2020-10-22T22:32:09Z

core/src/main/scala/org/http4s/StaticFile.scala

@@ -116,37 +111,31 @@ object StaticFile {
      Sync[F].delay(
        if (f.isFile) s"${f.lastModified().toHexString}-${f.length().toHexString}" else "")

-  def fromFile[F[_]: Sync: ContextShift](
+  def fromFile[F[_]: Files: Sync](


SystemFw · 2020-10-22T22:41:31Z

core/src/main/scala/org/http4s/multipart/MultipartParser.scala

      .handleErrorWith { err =>
        logger.error(err)("Caught error during file cleanup for multipart")
        //Swallow and report io exceptions in case
        F.unit
      }

-  private[this] def tailrecPartsFileStream[F[_]: Sync: ContextShift](
+  private[this] def tailrecPartsFileStream[F[_]: Sync: Files](


SystemFw · 2020-10-22T22:41:53Z

core/src/main/scala/org/http4s/multipart/MultipartParser.scala

      case None => Part(hdrs, body)
    }

  /** Split the stream on `values`, but when
    */
-  private def splitWithFileStream[F[_]](
+  private def splitWithFileStream[F[_]: Files](


SystemFw · 2020-10-22T22:42:11Z

core/src/main/scala/org/http4s/multipart/Part.scala

@@ -7,10 +7,10 @@
 package org.http4s
 package multipart

-import cats.effect.{Blocker, ContextShift, Sync}
+import cats.effect.kernel.Sync


SystemFw · 2020-10-22T22:42:19Z

core/src/main/scala/org/http4s/multipart/Part.scala

-      blocker: Blocker,
-      headers: Header*): Part[F] =
-    fileData(name, file.getName, readAll[F](file.toPath, blocker, ChunkSize), headers: _*)
+  def fileData[F[_]: Sync: Files](name: String, file: File, headers: Header*): Part[F] =


SystemFw · 2020-10-22T22:42:56Z

core/src/main/scala/org/http4s/syntax/KleisliSyntax.scala

@@ -9,7 +9,7 @@ package syntax

 import cats.{Functor, ~>}
 import cats.syntax.functor._
-import cats.effect.Sync
+import cats.effect.kernel.Sync


SystemFw · 2020-10-22T22:44:33Z

Left a lot of comments but they are all instances of the same 4-5 things. @djspiewak two interesting points btw:

do we want to encourage importing from kernel?
look at Dispatcher, there are a couple of "mini-ends-of-the-world" cases here

djspiewak · 2020-10-23T00:00:42Z

IMO if you have core, import from it. Kernel imports are for when you don't depend on core.

djspiewak · 2020-10-23T00:12:49Z

I'll look at the dispatcher call sites when I get back to my laptop

djspiewak · 2020-10-23T02:56:29Z

core/src/main/scala/org/http4s/internal/package.scala

+  private[http4s] def unsafeRunAsync[F[_]: Async, A](fa: F[A])(
+      f: Either[Throwable, A] => F[Unit])(implicit
+      dispatcher: Dispatcher[F],
+      ec: ExecutionContext): Unit =


Do you actually need to thread the ExecutionContext through anymore? Cats Effect contains its own EC, so in theory it shouldn't need to appear anywhere in the signatures.

djspiewak · 2020-10-23T02:57:35Z

core/src/main/scala/org/http4s/internal/package.scala

-      F: ConcurrentEffect[F]): Unit =
-    F.runAsync(F.start(F.delay(f)).flatMap(_.join))(loggingAsyncCallback(logger)).unsafeRunSync()
+  private[http4s] def invokeCallback[F[_]](logger: Logger)(
+      f: => Unit)(implicit F: Async[F], dispatcher: Dispatcher[F]): Unit =


I wouldn't generally recommend making the Dispatcher implicit. You can, but it's a bit weird given that it needs to live within a Resource.

djspiewak · 2020-10-23T02:58:20Z

core/src/main/scala/org/http4s/internal/package.scala

-  @deprecated(
-    "Replaced by cats.effect.Async.fromFuture. You will need a ContextShift[F].",
-    "0.21.4")
+  @deprecated("Replaced by cats.effect.Async.fromFuture", "0.21.4")


I'm guessing this should just be removed entirely since this has to be a bincompat breaking release anyway.

👍 I like to err on the side of keeping deprecations for graceful migrations, but this is private, so we can be more aggressive here.

djspiewak · 2020-10-23T02:59:03Z

core/src/main/scala/org/http4s/internal/package.scala

-          })
-        F.delay { cf.cancel(true); () }
+          })) >>
+          F.delay(Some(F.delay(cf.cancel(true)).void))


You probably want F.pure for the outer call here.

rossabaker

Thanks. This is an excellent start.

As you address the reviews (thanks, @SystemFw, @RaasAhsan, and @djspiewak. We're in good hands here!), please just add commits instead of force-pushing. This will make it easier for everyone to review the incremental changes as this progresses.

rossabaker · 2020-10-23T04:26:02Z

core/src/main/scala/org/http4s/internal/package.scala

+      f: Either[Throwable, A] => F[Unit])(implicit
+      dispatcher: Dispatcher[F],
+      ec: ExecutionContext): Unit =
+    dispatcher.unsafeRunSync(fa.evalOn(ec).attemptTap(f).void)


All the usages are in blaze. We could punt it to blaze-core right now, and see if we can eliminate entirely when we get to that backend.

rossabaker · 2020-10-23T04:31:29Z

core/src/main/scala/org/http4s/internal/package.scala

-  @deprecated(
-    "Replaced by cats.effect.Async.fromFuture. You will need a ContextShift[F].",
-    "0.21.4")
+  @deprecated("Replaced by cats.effect.Async.fromFuture", "0.21.4")


👍 I like to err on the side of keeping deprecations for graceful migrations, but this is private, so we can be more aggressive here.

djspiewak · 2020-10-23T04:39:46Z

Do you actually need to thread the ExecutionContext through anymore? Cats Effect contains its own EC, so in theory it shouldn't need to appear anywhere in the signatures.

All the usages are in blaze. We could punt it to blaze-core right now, and see if we can eliminate entirely when we get to that backend.

Oh is this the worker pool for blaze? Or is it blaze's Future-shifting pool? If it's the latter, then Async[F].executionContext should give you what you need (an F[ExecutionContext] that produces the compute pool). If it's the former, then continuing to thread it through makes sense, though it does result in some unfortunate API shenanigans higher up in the stack (my fault, I know!).

domaspoliakas · 2020-10-24T22:09:19Z

I think I resolved most of the feedback now.

Farewell kernel imports
: Sync: Files eliminated; it is now either just : Files or : Files: Concurrent
global IORuntime imports removed in favour of SyncIO
couple of misc things

The things that are incomplete right now:

the unsafeRunAsync that @rossabaker mentioned should go into blaze core
One section on multipart parser (line 756). I'll leave a comment outlining my confusion there.

domaspoliakas · 2020-10-24T22:14:14Z

core/src/main/scala/org/http4s/multipart/MultipartParser.scala

-            } yield split)
-              .handleErrorWith(e => Pull.eval(cleanupFile(path)) >> Pull.raiseError[F](e))
-          }
+        ???


My confusion is as thus: it seems a temp file is created but not always cleaned up. That does not sound right to me, so I tried to read through it to get a complete picture but I have to admit - I don't quite understand what's supposed to be happening here so I am left a little confused. Any help would be greatly appreciated.

Actually, I might be able to work it through. I'll give it a shot now.

domaspoliakas · 2020-10-24T22:38:00Z

core/src/main/scala/org/http4s/multipart/MultipartParser.scala

-              split <- streamAndWrite(s, state, Stream.empty, racc, 0, path)
-            } yield split)
-              .handleErrorWith(e => Pull.eval(cleanupFile(path)) >> Pull.raiseError[F](e))
+          .eval(Files[F].tempFile(???, "", "").allocated)


If someone could check if this logic correctly replaces previous logic that'd be very good

The ??? is the system property "java.io.tmpdir". The underlying Java API supports null there, and I don't know what else you'd do if that property is None, but I don't know what fs2 would do with that.

I think the logic is right. .allocate is always a smell. There's not a stream finalizer on the happy path to clean up the resource, which makes me think this function should return a Resource, which would bubble up and force the caller to use the temp files before they are disposed. I don't know what ramifications that would have, but we could track that as a separate issue.

I just checked java.io.File temp file creation allows nulls, but the nio temp file creation requires the dir property to be non-null and forces you to use the method which has no dir property for the default directory. So one way or the other the fs2 interface has to change unless we introduce a Sync requirement here again, but that'd be rather unfortunate. I'll see if I can get a change merged to fs2 to allow this.

rossabaker

Great progress.

rossabaker · 2020-10-25T03:07:30Z

core/src/main/scala/org/http4s/internal/package.scala

+      f: Either[Throwable, A] => F[Unit])(implicit
+      dispatcher: Dispatcher[F],
+      ec: ExecutionContext): Unit =
+    dispatcher.unsafeRunSync(fa.evalOn(ec).attemptTap(f).void)


Let's move it to blaze-core, under its natural package. That will help get core clean so other ports can commence, and give me more time to relearn and answer to @djspiewak's question. 😄

rossabaker · 2020-10-25T03:14:14Z

core/src/main/scala/org/http4s/multipart/MultipartParser.scala

-              split <- streamAndWrite(s, state, Stream.empty, racc, 0, path)
-            } yield split)
-              .handleErrorWith(e => Pull.eval(cleanupFile(path)) >> Pull.raiseError[F](e))
+          .eval(Files[F].tempFile(???, "", "").allocated)


The ??? is the system property "java.io.tmpdir". The underlying Java API supports null there, and I don't know what else you'd do if that property is None, but I don't know what fs2 would do with that.

rossabaker · 2020-10-25T03:22:38Z

core/src/main/scala/org/http4s/multipart/MultipartParser.scala

-              split <- streamAndWrite(s, state, Stream.empty, racc, 0, path)
-            } yield split)
-              .handleErrorWith(e => Pull.eval(cleanupFile(path)) >> Pull.raiseError[F](e))
+          .eval(Files[F].tempFile(???, "", "").allocated)


I think the logic is right. .allocate is always a smell. There's not a stream finalizer on the happy path to clean up the resource, which makes me think this function should return a Resource, which would bubble up and force the caller to use the temp files before they are disposed. I don't know what ramifications that would have, but we could track that as a separate issue.

domaspoliakas · 2020-10-27T13:32:24Z

core now compiles; I will proceed to fixing tests

rossabaker

Fantastic work. I'm happy going module-by-module on the PRs to keep them tractable and let others jump in.

rossabaker · 2020-10-27T14:21:11Z

project/Http4sPlugin.scala

    val catsEffectTesting = "0.4.1"
    val circe = "0.13.0"
    val cryptobits = "1.3"
    val disciplineSpecs2 = "1.1.0"
    val dropwizardMetrics = "4.1.13"
-    val fs2 = "3.0-5158029"
+    val fs2 = "3.0-cd73a32"


Oh, one question: Does this have anything that -M2 doesn't?

yeah, the tempFile and tempDirectory changes which unblock this PR, and were added today

#3784 (comment)

domaspoliakas · 2020-10-27T14:23:10Z

Great! I presume I need to "undraft" this PR now, so I'll do that and then make a new branch for test upgrade.

SystemFw · 2020-10-27T20:39:06Z

blaze-core/src/main/scala/org/http4s/blazecore/package.scala

+  // have an ExecutionContext but not a Timer.
+  private[http4s] def unsafeRunAsync[F[_], A](fa: F[A])(
+      f: Either[Throwable, A] => IO[Unit])(implicit F: Effect[F], ec: ExecutionContext): Unit =
+    F.runAsync(Async.shift(ec) *> fa)(f).unsafeRunSync()


Really keen to get to blaze and destroy this

SystemFw · 2020-10-27T21:00:21Z

core/src/main/scala/org/http4s/StaticFile.scala

            .handleError(_ => ())
            .as(Some(Response(NotModified)))
      }
    })
  }

-  def calcETag[F[_]: Sync]: File => F[String] =
+  def calcETag[F[_]: Files: MonadError[*[_], Throwable]]: File => F[String] =


Why not just Functor here?

I must've mistakenly replaced it along with some other Functors with MonadError while testing the waters. I'll get it changed to Functor.

domaspoliakas and others added 4 commits October 22, 2020 15:39

Updated to CE3/FS3; Migration WIP

903c036

core now compiles

e16ece1

Restored sbt files to their unformatted form

d9a6a4c

Updated cats-effect/fs2 versions

e9beb9b

SystemFw reviewed Oct 22, 2020

View reviewed changes

djspiewak reviewed Oct 23, 2020

View reviewed changes

rossabaker reviewed Oct 23, 2020

View reviewed changes

domaspoliakas added 3 commits October 24, 2020 20:21

CE3 migration: addressed some of the PR comments

6b25599

Loosened sync constraint on StaticFile

31b54aa

Restored asynchronicity that was lost

32837b3

domaspoliakas commented Oct 24, 2020

View reviewed changes

MultipartParser; continued migration to ce3

b49fb57

domaspoliakas commented Oct 24, 2020

View reviewed changes

rossabaker reviewed Oct 25, 2020

View reviewed changes

domaspoliakas added 2 commits October 25, 2020 08:01

Moved a method from core to blaze-core

1a10f62

Updated fs2/ce3 versions and removed ???

9f6f516

rossabaker approved these changes Oct 27, 2020

View reviewed changes

rossabaker reviewed Oct 27, 2020

View reviewed changes

domaspoliakas marked this pull request as ready for review October 27, 2020 14:23

rossabaker changed the title ~~[WIP] Core upgrade~~ Upgrade core to cats-effect-3 Oct 27, 2020

SystemFw reviewed Oct 27, 2020

View reviewed changes

Loosened constraint from MonadError to Functor

5e3747c

rossabaker merged commit ffc0444 into http4s:cats-effect-3 Oct 28, 2020

Upgrade core to cats-effect-3 #3784

Upgrade core to cats-effect-3 #3784

Conversation

domaspoliakas commented Oct 22, 2020

Choose a reason for hiding this comment

domaspoliakas Oct 23, 2020 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

SystemFw commented Oct 22, 2020 • edited Loading

djspiewak commented Oct 23, 2020

djspiewak commented Oct 23, 2020

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

rossabaker left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

djspiewak commented Oct 23, 2020

domaspoliakas commented Oct 24, 2020

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

rossabaker left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

domaspoliakas commented Oct 27, 2020

rossabaker left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

domaspoliakas commented Oct 27, 2020

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

domaspoliakas Oct 23, 2020 •

edited

Loading

SystemFw commented Oct 22, 2020 •

edited

Loading