Skip to content

A Scala DSL to write type-safe programs for distributed computing

License

Notifications You must be signed in to change notification settings

tailcallhq/zio-compose

Repository files navigation

badge-workflow badge-sonatype-releases badge-sonatype-snapshots

ZIO Compose is a library that helps you write programs that can be serialized and sent over the wire.

Introduction

The basic idea behind having serializable programs is if code and data are on different machines, one of them must be moved to the other before the code can be executed on the data. Typically, in big-data applications it's much more efficient to move code than the other way around.

There are other use-cases that don't involve big-data where you would want a serializable program. For eg: Building a rule engine, where the rules are implemented using a DSL and the DSL is serialized and sent to the server for execution.

ZIO Compose intends to take care of such use cases. It intends to provide a complete DSL to write any kind of distributed computation using Scala in a type-safe manner. It's built on top of ZIO Schema.

Installation

Update your resolvers and add zio-compose as a dependency in your build.sbt.

libraryDependencies += "com.tusharmath" %% "zio-compose" % version

Getting started

  1. Here is a simple program that adds two numbers -

    import compose.Lambda._
    
    val program = constant(1) + constant(2)
  2. Programs can be executed using the default interpreter:

    import zio._
    
    object ZIOCompose extends ZIOAppDefault {
      val run = for {
        res <- Interpreter.eval(program)
        _   <- ZIO.succeed(println(s"Result: ${res}"))
      } yield ()
    }

Lambda

The core data type in ZIO Compose is Lambda. It is also type aliased by ~> (tilde, greater than). A lambda A ~> B represents a serializable unary function that takes in an input of type A and produces and output of type B. For eg:

val c1: Any ~> Int = Lambda.constant(100)

The above lambda c1 is a function that takes in any input and produces an Int value.

Another example of lambda is identity[A] which like the scala's identity function, takes in a type A and returns itself as output. The only difference is that zio-compose's identity is serializable.

Serialization

Any lambda from A ~> B can be serialized into JSON by performing a few steps.

// A program that adds two numbers
val program: Any ~> Int = constant(1) + constant(2)

// Call the `compile` method to create an execution plan
val compilation: ExecutionPlan = program.compile

// call `json` on the execution plan to encode it as JSON
val json: String = compilation.compile.json

Conditional

Conditional operations can be implemented on Lambdas that return a Boolean using the diverge operator. The following program returns "Yes" if the condition is true and "No" if the condition is false.

import Lambda._

val program = (constant(1) > constant(2)).diverge(
  isTrue = constant("Yes"),
  isFalse = constant("No")
)

Since 1 < 2 the condition is false and the output thus becomes "no".

Piping

Two lambdas can be composed using the pipe or compose operator. For eg: if there exists a lambda l1: A ~> B and a lambda l2: B ~> C then they can be composed using the pipe operator as —

val l1: A ~> B = ???
val l2: B ~> C = ???
val l12: A ~> C = A >>> B

This is the semantic equivalent of l2(l1(a)) , where a is of type A.

Lenses

ZIO Compose has support for lenses which allows very precise control over getting and setting values over record types. For eg: Let's say there is a type User and we want to get the age of that user. We could do something like this —

import zio.schema._
import compose.macros.DeriveAccessors

case class User(firstName: String, lastName: String, age: Int)

object User {

  // Derive the Schema
  implicit val schema: Schema[User] = DeriveSchema.gen[User]

  // Derive accessors
  val lens = DeriveAccessors.gen[User]
}

The schema field inside of User provides access to the meta-data and structure of the type User. Whereas lens internally uses schema to navigate through an instance to lookup or update it's fields in a type-safe manner. Let's see that in action —

val user: Any ~> User = constant(User("John", "Doe", 23))
val age: User ~> Int = User.lens.age.get
val program: Any ~> Int = user >>> age

Here we create a user using constant and then using the derived lens we create a Lambda from User ~> Int. We compose the two lambdas together using the >>> operator (alias to pipe). The final program is a type-safe, serializable function that can take anything and produce an integer.

Now let's look at an example where we are updating a field using lenses in the User type -

val user: Any ~> User = constant(User("John", "Doe", 23))
val program: Any ~> User = (user <*> constant(12)) >>> User.lens.age.set

The set methods on lens is a binary function, so it needs two arguments - 1. The whole object which needs to be updated and 2. the value it needs to set. In our case age.set would have a type like this - (User, Int) ~> User. That's why we use the <*> operator (alias to zip) to combine the two inputs and send it to the set function.

Transformations

Transformations from one type to another are easily possible using the lens API, however it can become a bit verbose and boilerplate sometimes. ZIO Compose provides a DSL to simplify transformations. Here is an example of converting User to Customer, we start by defining the types, schema and it's lens.

case class Customer(name: String, age: Int, allowed: Boolean)

object Customer {
  implicit val schema = DeriveSchema.gen[Customer]
  val lens = DeriveAccessors.gen[Customer]
}

We then take each field of the user, perform some transformations on the field themselves and then set it in a customer. A transformation can be defined using the ->> operator.

val t1: User ~> Customer = (User.lens.age.get + constant(10)) ->> Customer.lens.age.set
,

A Transformation, is nothing but a pair of a getter and setter. We can combine multiple transformations using the transform operator -

import Lambda._

val user2Customer: User ~> Customer = transform(
  (User.lens.age.get + constant(10)) ->> Customer.lens.age.set,
  (User.lens.firstName.get ++ constant(" ") ++ Person.lens.lastName.get) ->> Customer.lens.name.set,
  (User.lens.age.get > constant(18)) ->> Customer.lens.isAllowed.set,
)

The final output of the transformation is a function from User ~> Customer. We can then pipe in an actual user instance to produce a customer as follows —

val program: Any ~> Customer = constant(User("John", "Doe", 20)) >>> user2Customer

Looping

With ZIO Compose one can loop over a lambda in multiple ways. For eg: Let's say I want to add all numbers between 0 to 10. We can do this by creating a type Sum which maintains intermediary state of our program like this —

import compose.macros.DeriveSchema

case class Sum(count: Int, result: Int)

object Sum {
  implicit val schema = DeriveSchema.gen[Sum]
  val lens = DeriveAccessor.gen[Sum]
}

Then we make a lambda of type Sum ~> Sum to represent one iteration of our loop. In the iteration we perform two operations -

  1. Increase the value of count by one.
  2. Add the value of count to result.
import Lambda._

val iteration: Sum ~> Sum = transform(
  Sum.lens.count.get.inc ->> Sum.lens.count.set,
  Sum.lens.result.get + Sum.lens.count.get ->> Sum.lens.result.set
)

We then use the repeatWhile operator to keep iterating while the condition is true.

val sum: Any ~> Sum = iteration.repeatWhile(Sum.lens.count.get < constant(10))

To get the exact value of the sum we can again use the lens API as follows —

val program: Any ~> Int = sum >>> Sum.lens.result.get

Scopes

Scopes allows us to define and update variables within a given context. They turn out to be pretty handy when we want to share some data across different part of our program without having to pass it using pipe. Below we take an arbitrary example where have two numbers and we want to check if their sum is greater than their product.

import Lambda._

val program = scope { implicit ctx =>
  val a = Ref.make(key = "a", value = 10)
  val b = Ref.make(key = "b", value = 5)
  val result = Ref.make(key = "result", value = false)

  (a.get + b.get) > (a.get * b.get) >>> result.set
}

A Ref is like a ZRef with get and set methods on it. It needs a unique key within the scope of it's usage and a default value at the time of initialization. However, it can only be initialized inside a scope { } block. The {implicit ctx => provides context in which the scoped variable is available.

Fibonacci

Here is an advanced example of a program that calculates fibonacci numbers and is completely serializable.

import compose._
import zio.schema._

case class Fib(a: Int, b: Int, i: Int)

object Fib {
  implicit val schema: Schema[Fib] = DeriveSchema.gen[Fib]
  val lens = DeriveAccessor.gen[Fib]
}

def fib = constant(Fib(0, 1, 0)) >>>
  transform(
    Fib.lens.b.get ->> Fib.lens.a.set,
    Fib.lens.a.get + Fib.lens.b.get ->> Fib.lens.b.set,
    Fib.lens.i.get.inc ->> Fib.lens.i.set,
  ).repeatWhile(Fib.lens.i.get =!= constant(20)) >>> Fib.lens.b.get

PS: If you like what you see, give the repo a ⭐️ 🙏

About

A Scala DSL to write type-safe programs for distributed computing

Topics

Resources

License

Stars

Watchers

Forks

Packages

No packages published

Languages