A catpanion library for cats-mtl and cats-effect providing:
- Easy composition of MTL-style functions
- MTL instances for cats-effect compatible datatypes (e.g.
IO
) and monix TaskLocal - Conflict-free implicits for sub-instances (e.g.
Stateful
=>Monad
)
Available for Scala 2.12 and 2.13, for Scala JVM and Scala.JS (1.x)
// Use %%% for scala.js or cross projects
// Classy lenses derivation (requires shapeless)
libraryDependencies += "com.olegpy" %% "meow-mtl-core" % "0.5.0"
// MTL instances for cats-effect Ref and effectful functions
libraryDependencies += "com.olegpy" %% "meow-mtl-effects" % "0.5.0"
// MTL instances for TaskLocal
libraryDependencies += "com.olegpy" %% "meow-mtl-monix" % "0.5.0"
Inspired by Next-level MTL talk and discussions on cats gitter.
You can also see demonstration of techniques this library enables in a post or a talk by Gabriel Volpe
type Headers = Map[String, String]
case class User(name: String)
case class AuthedRequest(headers: Headers, user: User)
def greetUser[F[_]: Functor](implicit F: Stateful[F, User]): F[String] = {
F.get.map(user => s"Hello, ${user.name}")
}
def addRequestIdHeader[F[_]: Sync](implicit F: Stateful[F, Headers]): F[Unit] =
for {
id <- Sync[F].delay(UUID.randomUUID().toString)
_ <- F.modify(_ + ("X-Request-ID" -> id))
} yield ()
Now, if you had AuthedRequest
as a state, that should mean that you
have a state of User
and Headers
too. This library allows you to call these
functions directly:
import com.olegpy.meow.hierarchy._
def handleGreetRequest[F[_]: Sync](implicit F: Stateful[F, AuthedRequest]) =
for {
_ <- addRequestIdHeader[F]
r <- greetUser[F]
} yield r
To get that Stateful
instance, it's possible to use StateT
transformer. But meow-mtl allows you to use Ref
from cats-effect
instead, yielding better performance. So at the edge of your application
it is possible to do this:
import com.olegpy.meow.effects._
def handleRequest: IO[String] =
for {
ref <- Ref[IO].of(AuthedRequest(Map(), User("John")))
res <- ref.runState { implicit monadState =>
handleGreetRequest[IO]
}
} yield res
Primary feature of meow-mtl is enabling boilerplate-free composition of
functions using cats-mtl typeclasses, in cases where instance clearly
either contains necessary fields (like State example above) or can be
converted to a necessary type. For example, it's possible to narrow
type of MonadError
from Throwable
to a custom exception type:
case class MyException(msg: String) extends Throwable
def handleOnlyMy[F[_], A](f: F[A], fallback: F[A])(implicit F: MonadError[F, MyException]) =
f.handleErrorWith(_ => fallback)
val io: IO[Int] = ???
handleOnlyMy(io, 42)
This is witnessed by Lens
and Prism
optics that meow-mtl generates
when you try to make a call to such method.
As another neat example, generated typeclasses can be used as ad-hoc lenses
case class Part(int: Int)
case class Whole(part: Part)
def modify[F[_]: Stateful[?[_], Whole]] =
Stateful[F, Part].set(Part(42)) // automatically "zooms" into Whole.part
All automatic derivation requires is a single import:
import com.olegpy.meow.hierarchy._
This needs to be done in every file where your call requires deriving an instance.
Supported typeclasses:
Typeclass | Required optic |
---|---|
ApplicativeError | Prism |
Handle | Prism |
MonadError | Prism |
Raise | Prism |
Tell | Prism |
Ask | Lens |
Local | Lens |
Stateful | Lens |
Don't use cats.mtl.implicits._
or cats.mtl.hierarchy.base._
imports.
Import cats.mtl.instances.all._
and cats.mtl.syntax.all._
if you
need it.
Failure to do this will result in ambiguous implicit instances.
In cats-mtl 0.4.0 hierarchy has been mostly replaced by subtyping. The remaining hierarchy imports will possibly be phased out
Alternatively, com.olegpy.meow.optics
can be used directly:
case class User(name: String)
type HasUser[A] = MkLensToType[A, User]
def isFred[A](a: A)(implicit mkLens: HasUser[A]) =
mkLens().get(a).name == "Fred"
In here, mkLens
is an object with 0-args apply
method, that creates
a shapeless Lens from A to User, e.g.:
case class RequestCtx(user: User, id: String)
assert { isFred(RequestCtx(User("Fred"), "0x42")) }
Prism works in similar way, but it's a custom class (not shapeless
Prism) with apply
and unapply
methods for construction and matching.
This is a very bare-bones implementation of optics, having only minimal functionality needed to support automatic derivation without adding extra dependencies. If you need a full-fledged optics library, consider using monocle instead.
meow-mtl provides instances for cats-effect compatible data types like
cats-effect own IO
or monix Coeval
and Task
. These instances
reside in com.olegpy.meow.effects
package and provide a more flexible
and performant alternative to monad transformer stacks.
Because construction of such instances is typically effectful, they are locally scoped. That means, instead of being available by importing, they require a special method to be called with a lambda, which will receive an instance, i.e.:
// `unsafe` is used for the sake of an example. I don't recommend doing that.
Ref.unsafe[IO, Int](0).runAsk { implicit askInstance =>
??? // Ask[IO, Int] is available in this scope
}
Alternatively, you can pull it out with specific methods if you intend to use it explicitly or with better-monadic-for implicit patterns:
implicit val instance: Stateful[IO, Int] =
Ref.unsafe[IO, Int](0).stateInstance
// Stateful available below
???
Ref
is a referentially transparent variable added in cats-effect
1.0.0-RC2. It supports Stateful
, Ask
and Tell
effects (the latest requires a Semigroup
instance for type of
contained data).
Instances are provided by extension methods runState
, runAsk
and
runTell
respectively.
This is a simple example of using Stateful
instance of Ref
. Note
how updated state can be retrieved from ref
after executing operation.
def getAndIncrement[F[_]: Apply](implicit MS: Stateful[F, Int]) =
MS.get <* MS.modify(_ + 1)
for {
ref <- Ref.of[IO](0)
out <- ref.runState { implicit ms =>
getAndIncrement[IO].replicateA(3).as("Done")
}
state <- ref.get
} yield (out, state) == ("Done", 3)
Consumer
is a simple wrapper around A => F[Unit]
. It supports a
single effect - Tell
, and can be used for things like logging,
persistence, notifications, etc.
Consumer
instances are constructed with apply
method on a companion.
That logger only waits if a previous message is still being processed, to ensure correct ordering:
def greeter(name: String)(implicit ev: Tell[IO, String]): IO[Unit] =
ev.tell(s"Long time no see, \$name") >> IO.sleep(1.second)
def forever[A](ioa: IO[A]): IO[Nothing] = ioa >> forever(ioa)
for {
mVar <- MVar.empty[IO, String]
logger = forever(mVar.take.flatMap(s => IO(println(s)))
_ <- logger.start // Do logging in background
_ <- Consumer(mVar.put).runTell { implicit tell =>
forever(greeter("Oleg"))
}
} yield ()
Similar to Ref, but with TaskLocal scoping it's possible to provide Local
Here, you can build a middleware for a service that provides some additional data per call. This can be used for e.g. generating request IDs in HTTP server.
def service[F[_]: Monad](greeting: String, print: String => F[Unit])(implicit ev: Ask[F, String]): F[Unit] =
ev.ask.map(name => s"$greeting $name") >>= print
def middleware[F[_]: Monad, A](getName: F[String])(service: F[Unit])(implicit ev: Local[F, String]) =
getName.flatMap(n => ev.scope(n)(service))
// Can be looking up something in external system, or random ID
val getName = Task(if (Random.nextBoolean()) "Oleg" else "Olga")
def putStrLn(s: String) = Task(println(s))
val run =
for {
name <- TaskLocal("")
// note that you can create service separately from middleware,
// as long as they share the TaskLocal
svc = name.runLocal { implicit ev =>
service[Task]("Hello,", putStrLn)
}
withRandomName = name.runLocal { implicit ev => middleware(getName) _ }
// and run them in another place entirely that doesn't know about TaskLocal
// Randomly prints "Hello, Oleg" or "Hello, Olga"
_ <- withRandomName(svc)
// prints "Hello, " since we don't set a context
_ <- svc
} yield ()
// Don't forget to enable Local support!
run.executeWithOptions(_.enableLocalContextPropagation)
meow-mtl also provides a set of implicits which let you use
Monad
/Applicative
/Functor
instances if you have an MTL instance of
compatible type (e.g. Stateful
/Ask
/Tell
)
import cats.implicits._
import com.olegpy.meow.prelude._ // just this one import
// Can use pure and flatTap without having a Monad constraint or pull
// it out manually
def test[F[_]](implicit MS: Stateful[F, Int]): F[Int] =
42.pure[F].flatTap(MS.set)
It uses LowPriority
mechanism from shapeless
to ensure that having
a constraint does not result in ambiguities:
import cats.effect.Sync
// Uses Sync as a Monad instance, instead of getting it from Stateful
def test2[F[_]: Sync](implicit MS: Stateful[F, Int]): F[Int] =
42.pure[F].flatTap(MS.set)
MIT