Deprecate Service and remove MaybeResponse #1424

aeons · 2017-09-26T18:47:34Z

Following discussion on Gittter, this is an experiment to

deprecate the Service[F, A, B] type alias and object of the same name in favour of the raw Kleisli[F, A, B] type
remove the MaybeResponse[F] and Pass[F] types and changing the HttpService[F] type to be Kleisli[OptionT[F, ?], Request[F], Response[F]]

aeons · 2017-09-26T18:51:57Z

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

@@ -136,4 +77,13 @@ package object http4s { // scalastyle:ignore
      case Left(l) => f(l)
    }
  }
+
+  implicit def http4sHttpServiceSemigroup[F[_]: Monad]: Semigroup[HttpService[F]] =


This is just here to make stuff compile.

Are we aiming to provide a Semigroup or a SemigroupK?

In the specific case of HttpService, I guess a Semigroup works just as well.

This still needs to be manually in scope to work, I just wanted the rest to compile while we figure out what to do.

aeons · 2017-09-26T18:52:16Z

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

+        Kleisli(request => x(request).orElse(y(request)))
+    }
+
+  def FToOptionT[F[_]: Functor]: FunctionK[F, OptionT[F, ?]] =


Not sure where this should live. It's come in handy a couple of times.

This is a good question for the Cats room. If we can't get a satisfactory cats answer, then we look to org.http4s.util, I think.

aeons · 2017-09-26T20:30:36Z

There are quite a few places where MaybeResponse.orNotFound is used. That is a pure empty 404 response. The Response object has a def notFound which takes a request and renders s"${request.pathInfo} not found".

I suggest renaming that one to something like notFoundFor(request: Request[F]) and adding a notFound that returns a 404 with something like Not found.

The empty page on hitting a 404 is not always obvious what went wrong, and having the default 404 defined somewhere ensures consistency.

rossabaker · 2017-09-26T21:06:35Z

That sounds reasonable.

bryce-anderson · 2017-09-27T15:34:49Z

Drive-by comment and haven't looked at the details in depth, but I'm cool with the general idea.

rossabaker

I think this is on the right track.

rossabaker · 2017-09-27T17:42:15Z

blaze-server/src/main/scala/org/http4s/server/blaze/Http2NodeStage.scala

-          }
-          .map(renderResponse(_))
-        catch serviceErrorHandler(req).andThen(_.widen[MaybeResponse[F]])
+          .getOrElse(Response.notFound)


Not apropos of this commit, but this needs to look more like Http1ServerStage to thread shift properly. 😟

rossabaker · 2017-09-27T17:44:05Z

client/src/main/scala/org/http4s/client/middleware/Retry.scala

@@ -7,9 +7,12 @@ import cats.implicits._
 import fs2._
 import java.time.Instant
 import java.time.temporal.ChronoUnit
+
+import cats.data.Kleisli


These imports didn't sort correctly.

rossabaker · 2017-09-27T17:48:22Z

core/src/main/scala/org/http4s/instances/FunctionKInstances.scala

+import cats.arrow.FunctionK
+import cats.data.OptionT
+
+trait FunctionKInstances {


I am a little reluctant to provide this when it's 100% cats functionality. Have we talked to the Cats project about whether this would be a welcome function there?

I implemented this with an extension method on Kleisli instead, so instead of

kleisliOfF.transform(FToOptionT)

it's now

kleisliOfF.liftOptionT

rossabaker · 2017-09-27T17:49:12Z

core/src/main/scala/org/http4s/instances/KleisliInstances.scala

+
+trait KleisliInstances {
+
+  implicit def http4sKleisliSemigroupK[F[_]: SemigroupK, A]: SemigroupK[Kleisli[F, A, ?]] =


And this is what we can eliminate with typelevel/cats#1928.

This is in that same awkward position as KleisliOps, but it replicates the syntax we want people to have once that cats issue moves forward. I'm sad but totally okay with this one.

rossabaker · 2017-09-27T17:52:31Z

server/src/main/scala/org/http4s/server/middleware/PushSupport.scala

@@ -87,21 +90,20 @@ object PushSupport {
      service: HttpService[F],
      verify: String => Boolean = _ => true): HttpService[F] = {

-    def gather(req: Request[F], resp: Response[F]): Response[F] =
+    def gather(req: Request[F])(resp: Response[F]): Response[F] =


rossabaker · 2017-09-27T17:54:10Z

server/src/main/scala/org/http4s/server/package.scala

-          f(req, service)
-        }
-    }
+    def apply[F[_], A, B, C, D](f: (C, Kleisli[F, A, B]) => F[D]): Middleware[F, A, B, C, D] =


Does anybody even use these? They're kind of Finagle-like. I have to stare at it every time I remember it's here to understand how it works.

Please, lets get rid of funky type wrappers unless they are neccesary. Feel Similarly towards EntityBody not that anyone is asking.

codecov-io · 2017-09-28T07:07:45Z

Codecov Report

Merging #1424 into master will decrease coverage by 0.35%.
The diff coverage is 70.68%.

@@            Coverage Diff             @@
##           master    #1424      +/-   ##
==========================================
- Coverage   73.84%   73.48%   -0.36%     
==========================================
  Files         235      238       +3     
  Lines        5050     4997      -53     
  Branches      279      270       -9     
==========================================
- Hits         3729     3672      -57     
- Misses       1321     1325       +4

Impacted Files	Coverage Δ
...in/scala/org/http4s/server/middleware/Logger.scala	`72.22% <ø> (ø)`	⬆️
...ttp4s/client/asynchttpclient/AsyncHttpClient.scala	`92% <ø> (ø)`	⬆️
core/src/main/scala/org/http4s/Service.scala	`0% <ø> (-80%)`	⬇️
...ala/org/http4s/server/blaze/Http1ServerStage.scala	`73.68% <ø> (-0.35%)`	⬇️
...in/scala/org/http4s/client/blaze/BlazeClient.scala	`100% <ø> (ø)`	⬆️
...ala/org/http4s/server/blaze/WebSocketSupport.scala	`10.71% <ø> (-3.08%)`	⬇️
.../main/scala/org/http4s/server/syntax/package.scala	`0% <0%> (ø)`	⬆️
...main/scala/org/http4s/server/metrics/Metrics.scala	`0% <0%> (ø)`	⬆️
...ain/scala/org/http4s/server/middleware/Jsonp.scala	`0% <0%> (ø)`	⬆️
...ala/org/http4s/server/middleware/PushSupport.scala	`0% <0%> (ø)`	⬆️
... and 41 more

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update c3c3ef7...d2e497c. Read the comment docs.

ChristopherDavenport · 2017-09-29T20:02:56Z

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

      service: Kleisli[OptionT[F, ?], A, Response[F]]): KleisliResponseOps[F, A] =
    new KleisliResponseOps[F, A](service)
+
+  implicit def http4sKleisliSYntax[F[_]: Functor, A, B](


Weird Capitalization here. Although I thought we might be removing it.

Yes, I'd like to see whether Cats is interested in providing this. KleisliResponseOps is our jam, but this one has nothing to do with HTTP.

Capitalization was a typo.

I don't disagree that this should be provided somewhere else, but having it in a released version of cats will not happen overnight.

Inlining the functionality is basically .mapF(OptionT.liftF(_)). Would that be an acceptable solution?

aeons · 2017-10-02T07:55:33Z

blaze-server/src/main/scala/org/http4s/server/blaze/Http2NodeStage.scala

-        case Left(t) =>
-          IO(logger.error(t)("Error rendering response"))
-      }
+      executionContext.execute(new Runnable {


This bit mirrors how Http1 works and should do the same as #1435 did for 0.17.

cant that be done via F.shift(executionContext) >> currentRunMethodMinusUnsafeRun

I thought the same, and no, because the routing happens outside the task.

scala> def foo(i: Int): IO[List[String]] = { val x = Thread.currentThread.getName; IO(List(x, Thread.currentThread.getName)) } e; IO(List(x, Thread.currentThread.getName)) }Thread.currentThread.getNam <console>:27: warning: parameter value i in method foo is never used def foo(i: Int): IO[List[String]] = { val x = Thread.currentThread.getName; IO(List(x, Thread.currentThread.getName)) } ^ foo: (i: Int)cats.effect.IO[List[String]] scala> (Effect[IO].shift(global) >> foo(42)).unsafeRunSync (Effect[IO].shift(global) >> foo(42)).unsafeRunSync res7: List[String] = List(run-main-0, scala-execution-context-global-125)

I think it could be done with an F.shift if the service were called inside F and then flattened.

aeons · 2017-10-02T08:34:01Z

The last couple of commits inlines the F ~> OptionT[F, ?] and deprecates {Authed,Http}Service.lift.

rossabaker · 2017-10-02T15:47:53Z

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

+      service: Kleisli[OptionT[F, ?], A, Response[F]]): KleisliResponseOps[F, A] =
+    new KleisliResponseOps[F, A](service)
+
+  implicit def http4sKleisliSYntax[F[_]: Functor, A, B](


Still miscapitalized

rossabaker · 2017-10-02T15:51:24Z

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

+}
+
+final class KleisliOps[F[_]: Functor, A, B](self: Kleisli[F, A, B]) {
+  def liftOptionT: Kleisli[OptionT[F, ?], A, B] =


I'm still a little hesitant to expose implicit syntax for types we don't own and aren't in our domain, but I see the use case for it. Would it be better to make this private[http4s] until Cats provides a better solution?

rossabaker · 2017-10-02T15:54:43Z

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

+    * handle all requests it is given.  If `f` is a `PartialFunction`, use
+    * `apply` instead.
+    */
+  def lift[F[_]: Functor](f: Request[F] => F[Response[F]]): HttpService[F] =


This is one I thought could be deprecated.

rossabaker · 2017-10-02T15:56:36Z

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

+  def lift[F[_]: Functor](f: Request[F] => F[Response[F]]): HttpService[F] =
+    Kleisli(f).liftOptionT
+
+  def liftF[F[_]](f: Request[F] => OptionT[F, Response[F]]): HttpService[F] =


We call it liftF because it lifts a Function? I don't have a better name. Cats calls this apply, but that's already taken. lift is confusing for the existing lift and for this, because Kleisli.lift takes just the F[B].

Does this help with type inference, or could we even deprecate this one and just use Kleisli for things that are already Kleisli functions and leave our apply as the DSL for partiality?

rossabaker · 2017-10-02T15:57:58Z

core/src/main/scala/org/http4s/instances/KleisliInstances.scala

+
+trait KleisliInstances {
+
+  implicit def http4sKleisliSemigroupK[F[_]: SemigroupK, A]: SemigroupK[Kleisli[F, A, ?]] =


This is in that same awkward position as KleisliOps, but it replicates the syntax we want people to have once that cats issue moves forward. I'm sad but totally okay with this one.

rossabaker · 2017-10-02T15:59:19Z

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

+    * handle all requests it is given.  If `f` is a `PartialFunction`, use
+    * `apply` instead.
+    */
+  def lift[F[_]: Functor, T](f: AuthedRequest[F, T] => F[Response[F]]): AuthedService[F, T] =


Another one to deprecate.

rossabaker · 2017-10-02T16:32:15Z

Alright, I'm happy with this in current form, with the possible exception of the existence of liftF alias for Kleisli.apply. I wonder if those just add more confusion, like the old Service alias.

aeons · 2017-10-02T16:35:28Z

I would be perfectly fine with deprecating .liftF as well and instructing people to use Kleisli.apply.

ChristopherDavenport · 2017-10-02T16:51:19Z

I agree with the deprecations. Let's make things explicit as this PR suggests.

aeons · 2017-10-02T17:28:03Z

liftF is now removed. Should be good to merge on green.

Deprecate Service and remove MaybeResponse

aeons added the don't merge yet label Sep 26, 2017

aeons force-pushed the kleislis-all-over branch from a303055 to dfec3fb Compare September 26, 2017 18:49

aeons commented Sep 26, 2017

View reviewed changes

rossabaker mentioned this pull request Sep 27, 2017

Deprecate Monoid[MaybeResponse] #1429

Merged

aeons added 7 commits September 27, 2017 15:42

core: Deprecate Service and remove MaybeResponse

9727a29

server: Deprecate Service and remove MaybeResponse

ecd4e32

Rearrange imports

34707c6

client: Deprecate Service and remove MaybeResponse

64cfeb4

blaze-server: Deprecate Service and remove MaybeResponse

54a38fc

Refactor Response.notFound

7fd472c

rest: Deprecate Service and remove MaybeResponse

30ec1f4

aeons force-pushed the kleislis-all-over branch from ad5d82e to 30ec1f4 Compare September 27, 2017 13:42

aeons changed the title ~~WIP: Deprecate Service and remove MaybeResponse~~ Deprecate Service and remove MaybeResponse Sep 27, 2017

Rearrange imports, again

2a514de

Move SemigroupK[Kleisli] and FToOptionT instances

eef3194

rossabaker reviewed Sep 27, 2017

View reviewed changes

aeons added 3 commits September 27, 2017 20:09

Remember the tut

90e6618

Address review comments

87121d8

Make it actually compile

e7caca7

Use an extension method on Kleisli instead of a FunctionK

02f1c10

ChristopherDavenport mentioned this pull request Sep 29, 2017

Response Monoid Instance #1434

Closed

ChristopherDavenport reviewed Sep 29, 2017

View reviewed changes

rossabaker mentioned this pull request Oct 1, 2017

Release automation for 0.18 #1439

Merged

aeons mentioned this pull request Oct 2, 2017

Thread shift the service function in HTTP/2 server #1435

Merged

aeons commented Oct 2, 2017

View reviewed changes

aeons added 2 commits October 2, 2017 10:14

Inline F ~> OptionT[F, ?] on Kleisli

57b40a8

Deprecate {Authed,Http}Service.lift

ff26231

aeons removed the don't merge yet label Oct 2, 2017

rossabaker reviewed Oct 2, 2017

View reviewed changes

Remove {Authed,Http}Service.liftF

d2e497c

rossabaker approved these changes Oct 2, 2017

View reviewed changes

aeons merged commit 0d1ebe8 into http4s:master Oct 2, 2017

aeons deleted the kleislis-all-over branch October 2, 2017 17:58

rossabaker mentioned this pull request Mar 8, 2018

API change ideas, per gitter discussion #1692

Closed

rossabaker pushed a commit to http4s/http4s-servlet that referenced this pull request Apr 3, 2022

Merge pull request http4s/http4s#1424 from aeons/kleislis-all-over

3f479de

Deprecate Service and remove MaybeResponse

rossabaker pushed a commit to http4s/http4s-servlet that referenced this pull request Apr 4, 2022

Merge pull request http4s/http4s#1424 from aeons/kleislis-all-over

ddd3551

Deprecate Service and remove MaybeResponse

armanbilge pushed a commit to http4s/http4s-async-http-client that referenced this pull request May 20, 2022

Merge pull request http4s/http4s#1424 from aeons/kleislis-all-over

8953aef

Deprecate Service and remove MaybeResponse


		trait KleisliInstances {

		implicit def http4sKleisliSemigroupK[F[_]: SemigroupK, A]: SemigroupK[Kleisli[F, A, ?]] =

Deprecate Service and remove MaybeResponse #1424

Deprecate Service and remove MaybeResponse #1424

Conversation

aeons commented Sep 26, 2017

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

aeons commented Sep 26, 2017

rossabaker commented Sep 26, 2017

bryce-anderson commented Sep 27, 2017

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

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

aeons Sep 27, 2017 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

codecov-io commented Sep 28, 2017 • edited Loading

Codecov Report

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

aeons commented Oct 2, 2017

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 commented Oct 2, 2017

aeons commented Oct 2, 2017

ChristopherDavenport commented Oct 2, 2017

aeons commented Oct 2, 2017

aeons Sep 27, 2017 •

edited

Loading

codecov-io commented Sep 28, 2017 •

edited

Loading