Architecting Typed FP Applications & Libraries in Kotlin with Λrrow

https://youtu.be/VOZZTSuDMFE
https://github.com/47deg/arrow-architecture

This talk includes a comprehensive walkthrough of the most important patterns covered by the data types and type classes we find in Λrrow. Each pattern will be accompanied by code examples that illustrate how Λrrow brings Typed Functional Programming to the Kotlin Programming Language.

We will learn the fundamentals of Typed Functional Programming applied to Kotlin with the library Arrow and how we can architect applications and libraries that are polymorphic and composed of pure abstract functions using type classes.

Arrow provides a unified programming model in by which Kotlin practitioners can build programs relying on the traditional FP, MTL and Effect type classes in a Tagless Final style offering levels of flexibility and techniques new to the Kotlin FP community.

Transcript

1. Building Apps & Libraries with Λrrow / @raulraja !" @47deg !" Sources !" Slides 1
2. Who am I? # @raulraja @47deg • Co-Founder and CTO at 47 Degrees • Typed FP advocate (regardless of language) / @raulraja !" @47deg !" Sources !" Slides 2
3. Started as learning Exercise to learn FP in the spanish Android Community Slack / @raulraja !" @47deg !" Sources !" Slides 3
4. !!"then KΛTEGORY was born: Solution for Typed FP in Kotlin / @raulraja !" @47deg !" Sources !" Slides 4
5. KΛTEGORY + Funktionale = Λrrow / @raulraja !" @47deg !" Sources !" Slides 5
6. Type classes Λrrow contains many FP related type classes Error Handling ApplicativeError, MonadError Computation Functor, Applicative, Monad, Bimonad, Comonad Folding Foldable, Traverse Combining Semigroup, SemigroupK, Monoid, MonoidK Effects MonadDefer, Async, Effect Recursion Recursive, BiRecursive,!!" MTL FunctorFilter, MonadState, MonadReader, MonadWriter, MonadFilter, !!" / @raulraja !" @47deg !" Sources !" Slides 6
7. Data types Λrrow contains many data types to cover general use cases. Error Handling Option,Try, Validated, Either, Ior Collections ListK, SequenceK, MapK, SetK RWS Reader, Writer, State Transformers ReaderT, WriterT, OptionT, StateT, EitherT Evaluation Eval, Trampoline, Free, FunctionN Effects IO, Free, ObservableK Optics Lens, Prism, Iso,!!" Recursion Fix, Mu, Nu,!!" Others Coproduct, Coreader, Const, !!" / @raulraja !" @47deg !" Sources !" Slides 7
8. Let's build a simple library Requirements 1.Fetch Gists information given a github user 2.Immutable model • Allow easy in memory updates • Support deeply nested relationships without boilerplate 3.Support async non-blocking data types: • Observable, Flux, Deferred and IO • Allow easy access to nested effects 4.Pure: • Never throw exceptions • Defer effects evaluation / @raulraja !" @47deg !" Sources !" Slides 8
9. Fetch Gists information given a github user fun publicGistsForUser(userName: String): List<Gist> = TODO() / @raulraja !" @47deg !" Sources !" Slides 9
10. Immutable model • Allow easy in memory updates • Support deeply nested relationships without boilerplate data class Gist( val files: Map<String, GistFile>, val description: String?, val comments: Long, val owner: GithubUser) { override fun toString(): String = "Gist($description, ${owner.login}, file count: ${files.size})" } data class GithubUser(val login: String) data class GistFile(val fileName: String?) / @raulraja !" @47deg !" Sources !" Slides 10
11. Immutable model • Allow easy in memory updates • Support deeply nested relationships without boilerplate import arrow.intro.* val gist = Gist( files = mapOf( "typeclassless_tagless_extensions.kt" to GistFile( fileName = "typeclassless_tagless_extensions.kt" ) ), description = "Tagless with Λrrow & typeclassless using extension functions and instances", comments = 0, owner = GithubUser(login = "-__unkown_user1__-") ) / @raulraja !" @47deg !" Sources !" Slides 11
12. Immutable model The data class synthetic copy is fine for simple cases gist.copy(description = gist.description!"toUpperCase()) !# Gist(TAGLESS WITH ΛRROW & TYPECLASSLESS USING EXTENSION FUNCTIONS AND INSTANCES, -__unkown_user1__-, file count: 1) / @raulraja !" @47deg !" Sources !" Slides 12
13. Immutable model As we dive deeper to update nested data the levels of nested copy increases gist.copy( owner = gist.owner.copy( login = gist.owner.login.toUpperCase() ) ) !" Gist(Tagless with Λrrow & typeclassless using extension functions and instances, -__UNKOWN_USER1__-, file count: 1) / @raulraja !" @47deg !" Sources !" Slides 13
14. Immutable model In Typed FP immutable updates is frequently done with Optics like Lens import arrow.optics.* val ownerLens: Lens<Gist, GithubUser> = Lens( get = { gist !" gist.owner }, set = { value !" { gist: Gist !" gist.copy(owner = value) }} ) val loginLens: Lens<GithubUser, String> = Lens( get = { user !" user.login }, set = { value !" { user !" user.copy(login = value) }} ) val ownerLogin = ownerLens compose loginLens ownerLogin.modify(gist, String!#toUpperCase) !$ Gist(Tagless with Λrrow & typeclassless using extension functions and instances, -__UNKOWN_USER1__-, file count: 1) / @raulraja !" @47deg !" Sources !" Slides 14
15. Immutable model Updating arbitrarily nested data with Λrrow is a piece of cake @optics data class Gist( val url: String, val id: String, val files: Map<String, GistFile>, val description: String?, val comments: Long, val owner: GithubUser ) { companion object } / @raulraja !" @47deg !" Sources !" Slides 15
16. Provide an immutable data model and means to update it Updating arbitrarily nested data with Λrrow is a piece of cake - val ownerLens: Lens<Gist, GithubUser> = - Lens( - get = { gist !" gist.owner }, - set = { value !" { gist: Gist !" gist.copy(owner = value) }} - ) - val loginLens: Lens<GithubUser, String> = - Lens( - get = { user !" user.login }, - set = { value !" { user !" user.copy(login = value) }} - ) - val ownerLogin = ownerLens compose loginLens - ownerLogin.modify(gist, String!#toUpperCase) + import arrow.optics.dsl.* + Gist.owner.login.modify(gist, String!#toUpperCase) / @raulraja !" @47deg !" Sources !" Slides 16
17. Let's build a simple library Requirements 1.Fetch Gists information given a github user 2.Immutable model • Allow easy in memory updates • Support deeply nested relationships without boilerplate 3.Support async non-blocking data types: • Observable, Flux, Deferred and IO • Allow easy access to nested effects 4.Pure: • Never throw exceptions • Defer effects evaluation / @raulraja !" @47deg !" Sources !" Slides 17
18. Support Async/Non-Blocking Popular data types A initial impure implementation that blocks and throws exceptions import arrow.intro.Gist import arrow.data.* import com.squareup.moshi.* import com.github.kittinunf.fuel.httpGet import com.github.kittinunf.result.Result fun publicGistsForUser(userName: String): ListK<Gist> { val (_,_, result) = "https:!"api.github.com/users/$userName/gists".httpGet().responseString() !" blocking IO return when (result) { is Result.Failure !# throw result.getException() !" blows the stack is Result.Success !# fromJson(result.value) } } / @raulraja !" @47deg !" Sources !" Slides 18
19. Let's build a simple library Requirements 1.Fetch Gists information given a github user 2.Immutable model • Allow easy in memory updates • Support deeply nested relationships without boilerplate 3.Support async non-blocking data types: • Observable, Flux, Deferred and IO • Allow easy access to nested effects 4.Pure: • Never throw exceptions • Defer effects evaluation / @raulraja !" @47deg !" Sources !" Slides 19
20. Don't throw exceptions When learn FP we usually start with exception-free but synchronous Try and Either like types. import arrow.core.* fun publicGistsForUser(userName: String): Either<Throwable, ListK<Gist!" { val (_,_, result) = "https:!#api.github.com/users/$userName/gists".httpGet().responseString() !# blocking IO return when (result) { is Result.Failure !$ result.getException().left() !#exceptions as a value is Result.Success !$ fromJson(result.value).right() } } publicGistsForUser("-__unkown_user__-") !# Left(a=com.github.kittinunf.fuel.core.HttpException: HTTP Exception 404 Not Found) / @raulraja !" @47deg !" Sources !" Slides 20
21. Let's build a simple library Requirements 1.Fetch Gists information given a github user 2.Immutable model • Allow easy in memory updates • Support deeply nested relationships without boilerplate 3.Support async non-blocking data types: • Observable, Flux, Deferred and IO • Allow easy access to nested effects 4.Pure: • Never throw exceptions • Defer effects evaluation / @raulraja !" @47deg !" Sources !" Slides 21
22. Support Async/Non-Blocking Popular data types Many choose to go non-blocking with Kotlin Coroutines, a great and popular kotlin async framework import kotlinx.coroutines.experimental.* fun publicGistsForUser(userName: String): Deferred<Either<Throwable, ListK<Gist!!" = async { val (_, _, result) = "https:!#api.github.com/users/$userName/gists".httpGet().responseString() when (result) { is Result.Failure !$ result.getException().left() is Result.Success !$ fromJson(result.value).right() } } !#by default `async` when constructed runs and does not suspend effects publicGistsForUser("-__unkown_user1__-") !# DeferredCoroutine{Active}@514149e1 / @raulraja !" @47deg !" Sources !" Slides 22
23. Let's build a simple library Requirements 1.Fetch Gists information given a github user 2.Immutable model • Allow easy in memory updates • Support deeply nested relationships without boilerplate 3.Support async non-blocking data types: • Observable, Flux, Deferred and IO • Allow easy access to nested effects 4.Pure: • Never throw exceptions • Defer effects evaluation / @raulraja !" @47deg !" Sources !" Slides 23
24. Support Async/Non-Blocking Popular data types But now we have to dive deep into the Deferred and Either effects to get to the value we care about suspend fun allGists(): List<Gist> { val result1: Either<Throwable, ListK<Gist!" = publicGistsForUser("-__unkown_user1__-").await() val result2: Either<Throwable, ListK<Gist!" = publicGistsForUser("-__unkown_user2__-").await() return when { result1 is Either.Right !# result2 is Either.Right !$ result1.b + result2.b else !$ emptyList<Gist>() } } / @raulraja !" @47deg !" Sources !" Slides 24
25. Support Async/Non-Blocking Popular data types Λrrow Monad Transformers help with syntax in the world of nested effects. import arrow.effects.* import arrow.instances.* import arrow.typeclasses.* import arrow.effects.typeclasses.* fun allGists(): DeferredK<Either<Throwable, List<Gist!!" = EitherT .monad<ForDeferredK, Throwable>(DeferredK.monad()) .binding { val result1 = EitherT(publicGistsForUser("-__unkown_user1__-").k()).bind() val result2 = EitherT(publicGistsForUser("-__unkown_user2__-").k()).bind() result1 + result2 }.value().fix() !# Λrrow's delegation to `async` is always lazy allGists() !# DeferredK(deferred=LazyDeferredCoroutine{New}@5113d1f2) / @raulraja !" @47deg !" Sources !" Slides 25
26. Let's build a simple library Requirements 1.Fetch Gists information given a github user 2.Immutable model • Allow easy in memory updates • Support deeply nested relationships without boilerplate 3.Support async non-blocking data types: • Observable, Flux, Deferred and IO !" What about all other data types? • Allow easy access to nested effects 4.Pure: • Never throw exceptions • Defer effects evaluation / @raulraja !" @47deg !" Sources !" Slides 26
27. Support Async/Non-Blocking Popular data types Turns out we don't need concrete data types if we use Type classes and Polymorphism / @raulraja !" @47deg !" Sources !" Slides 27
28. Support Async/Non-Blocking Popular data types Λrrow can abstract away the computational container type emulating higher kinded types. Kind<F, A> denotes an A value inside an F type contructor: Ex: List<A>, Deferred<A>, IO<A>, Observable<A> import arrow.Kind interface GistApiDataSource<F> { fun publicGistsForUser(userName: String): Kind<F, ListK<Gist!" } / @raulraja !" @47deg !" Sources !" Slides 28
29. Support Async/Non-Blocking Popular data types Emulating higher kinded types is based on defunctionalization Lightweight higher-kinded polymorphism by Jeremy Yallop and Leo White + @higherkind + class Option<A> : OptionOf<A> - class ForOption private constructor() { companion object } - typealias OptionOf<A> = arrow.Kind<ForOption, A> - inline fun <A> OptionOf<A>.fix(): Option<A> = - this as Option<A> / @raulraja !" @47deg !" Sources !" Slides 29
30. Support Async/Non-Blocking Popular data types How can we implement a computation in the context of F if we don't know what F is? class DefaultGistApiDataSource<F> : GistApiDataSource<F> { override fun publicGistsForUser(userName: String): Kind<F, ListK<Gist!" = TODO() } / @raulraja !" @47deg !" Sources !" Slides 30
31. Support Async/Non-Blocking Popular data types Ad-Hoc Polymorphism and type classes! A type class is a generic interface that describes behaviors that concrete types can support interface Functor<F> { !" Λrrow projects type class behaviors as static or extension functions over kinded values fun <A, B> Kind<F, A>.map(f: (A) !$ B): Kind<F, B> fun <A, B> lift(f: (A) !$ B): (Kind<F, A>) !$ Kind<F, B> = { fa: Kind<F, A> !$ fa.map(f) } } / @raulraja !" @47deg !" Sources !" Slides 31
32. Support Async/Non-Blocking Popular data types Ad-Hoc Polymorphism and type classes! A data type may be able to implement such abstract interfaces @extension interface DeferredFunctor : Functor<ForDeferredK> { override fun <A, B> Kind<ForDeferredK, A>.map(f: (A) !" B): DeferredK<B> = fix().map(f) } / @raulraja !" @47deg !" Sources !" Slides 32
33. Support Async/Non-Blocking Popular data types Ad-Hoc Polymorphism and type classes! A data type may be able to implement such abstract interfaces @extension interface IOFunctor : Functor<ForIO> { override fun <A, B> Kind<ForIO, A>.map(f: (A) !" B): IO<B> = fix().map(f) } / @raulraja !" @47deg !" Sources !" Slides 33
34. Support Async/Non-Blocking Popular data types Ex. Functor allows us to transform the contents regardless of the concrete data type. listOf(1).map { it + 1 } !" [2] Option(1).map { it + 1 } !" Some(2) Try { 1 }.map { it + 1 } !" Success(value=2) Either.Right(1).map { it + 1 } !" Right(b=2) / @raulraja !" @47deg !" Sources !" Slides 34
35. Support Async/Non-Blocking Popular data types Λrrow includes a comprehensive list of type classes Type class Combinator Semigroup combine Monoid empty Functor map, lift Foldable foldLeft, foldRight Traverse traverse, sequence Applicative just, ap ApplicativeError raiseError, catch Monad flatMap, flatten MonadError ensure, rethrow MonadDefer delay, suspend Async async Effect runAsync / @raulraja !" @47deg !" Sources !" Slides 35
36. Λrrow includes a comprehensive list of type classes Data types may support all or a subset of type classes based on capabilities: Type class Combinators List Functor map, lift ✓ Applicative just, ap ✓ ApplicativeError raiseError, catch ✕ Monad flatMap, flatten ✓ MonadError ensure, rethrow ✕ MonadDefer delay, suspend ✕ Async async ✕ Effect runAsync ✕ / @raulraja !" @47deg !" Sources !" Slides 36
37. Λrrow includes a comprehensive list of type classes Data types may support all or a subset of type classes based on capabilities: Type class Combinators List Either Deferred IO Functor map, lift ✓ ✓ ✓ ✓ Applicative pure, ap ✓ ✓ ✓ ✓ ApplicativeError raiseError, catch ✕ ✓ ✓ ✓ Monad flatMap, flatten ✓ ✓ ✓ ✓ MonadError ensure, rethrow ✕ ✓ ✓ ✓ MonadDefer delay, suspend ✕ ✕ ✓ ✓ Async async ✕ ✕ ✓ ✓ Effect runAsync ✕ ✕ ✓ ✓ / @raulraja !" @47deg !" Sources !" Slides 37
38. Support Async/Non-Blocking Popular data types We can use the Async type class to lift async computations into the abstract context of F class DefaultGistApiDataSource<F>(private val async: Async<F>) : GistApiDataSource<F>, Async<F> by async { override fun publicGistsForUser(userName: String): Kind<F, ListK<Gist!" = async { proc: (Either<Throwable, ListK<Gist!") !# Unit !# "https:!$api.github.com/users/$userName/gists".httpGet().responseString { _, _, result !# when (result) { is Result.Failure !# proc(result.getException().left()) is Result.Success !# proc(fromJson(result.value).right()) } } } } / @raulraja !" @47deg !" Sources !" Slides 38
39. Support Async/Non-Blocking Popular data types If we have more than one logical services we can group them into a module abstract class Module<F>( val async: Async<F>, val logger: Logger<F> = DefaultConsoleLogger(async), private val dataSource: GistApiDataSource<F> = DefaultGistApiDataSource(async, logger), val api: GistsApi<F> = DefaultGistApi(dataSource) ) / @raulraja !" @47deg !" Sources !" Slides 39
40. Support Async/Non-Blocking Popular data types Our library now supports all data types that provide a type class instance for Async. This pattern allow you to keep code in a single place while providing compile "com.biz:mylib-coroutines:$version" object KotlinCoroutinesRuntime : Module<ForDeferredK>(DeferredK.async()) import arrow.intro.runtime.* KotlinCoroutinesRuntime.api.publicGistsForUser("-__unkown_user1__-") !" DeferredK(deferred=LazyDeferredCoroutine{New}@2e2d965) / @raulraja !" @47deg !" Sources !" Slides 40
41. Support Async/Non-Blocking Popular data types Our library now supports all data types that provide a type class instance for Async. This pattern allow you to keep code in a single place while providing compile "com.biz:mylib-reactor:$version" object ReactorRuntime : Module<ForFluxK>(FluxK.async()) import arrow.intro.runtime.* ReactorRuntime.api.publicGistsForUser("-__unkown_user1__-") !" FluxK(flux=FluxFlatMap) / @raulraja !" @47deg !" Sources !" Slides 41
42. Support Async/Non-Blocking Popular data types Our library now supports all data types that provide a type class instance for Async. This pattern allow you to keep code in a single place while providing compile "com.biz:mylib-arrow-io:$version" object IORuntime : Module<ForIO>(IO.async()) import arrow.intro.runtime.* IORuntime.api.publicGistsForUser("-__unkown_user1__-") !" Bind(cont=Suspend(thunk=() !# arrow.effects.IO.Pure<A>), g=(A) !# arrow.effects.IO<B>) / @raulraja !" @47deg !" Sources !" Slides 42
43. Support Async/Non-Blocking Popular data types Our library now supports all data types that provide a type class instance for Async. This pattern allow you to keep code in a single place while providing compile "com.biz:mylib-rx2:$version" object Rx2Runtime : Module<ForObservableK>(ObservableK.async()) import arrow.intro.runtime.Rx2Runtime Rx2Runtime.api.publicGistsForUser("-__unkown_user1__-") !" ObservableK(observable=io.reactivex.internal.operators.observable.ObservableFlatMap@fb152c5) / @raulraja !" @47deg !" Sources !" Slides 43
44. Let's build a simple library Requirements 1.Fetch Gists information given a github user 2.Immutable model • Allow easy in memory updates • Support deeply nested relationships without boilerplate 3.Support async non-blocking data types: • Observable, Flux, Deferred and IO • Allow easy access to nested effects 4.Pure: • Never throw exceptions • Defer effects evaluation / @raulraja !" @47deg !" Sources !" Slides 44
45. Recap Requirements 1.FUNC REQ Fetch Gists information given a github user 2.OPTICS Immutable model • Allow easy in memory updates • Support deeply nested relationships without boilerplate 3.POLYMORPHISM Support async non-blocking data types: • Observable, Flux, Deferred and IO • Allow easy access to nested effects 4.EFFECT CONTROL Pure: • Never throw exceptions • Defer effects evaluation / @raulraja !" @47deg !" Sources !" Slides 45
46. Λrrow is modular Pick and choose what you'd like to use. Module Contents typeclasses Semigroup, Monoid, Functor, Applicative, Monad!!" core/data Option, Try, Either, Validated!!" effects Async, MonadDefer, Effect, IO!!" effects-rx2 ObservableK, FlowableK, MaybeK, SingleK effects-coroutines DeferredK mtl MonadReader, MonadState, MonadFilter,!!" free Free, FreeApplicative, Trampoline, !!" recursion-schemes Fix, Mu, Nu optics Prism, Iso, Lens, !!" meta @higherkind, @deriving, @extension, @optics / @raulraja !" @47deg !" Sources !" Slides 46
47. We want to make Typed FP in Kotlin even easier / @raulraja !" @47deg !" Sources !" Slides 47
48. Thanks to @tomasruizlopez we have a POC for KEEP-87: https:!"github.com/arrow-kt/kotlin/pull/6 / @raulraja !" @47deg !" Sources !" Slides 48
49. KEEP-87 Proposes the following changes to Kotlin Type class declarations are simple plain interfaces and have a expanded usage beyond FP interface Repository<A> { fun A.save(): A fun cache(): List<A> } / @raulraja !" @47deg !" Sources !" Slides 49
50. KEEP-87 Proposes the following changes to Kotlin Multiple data types can implement the behavior without resorting to inheritance extension object UserRepository : Repository<User> { fun User.save(): User = TODO() fun cache(): List<User> = TODO() } / @raulraja !" @47deg !" Sources !" Slides 50
51. KEEP-87 Proposes the following changes to Kotlin We can write polymorphic code with compile time verified dependencies fun <A> persistCache(with R: Repository<A>): List<A> = cache().map { it.save() } persistCache<User>() !" compiles and runs because there is a [Repository<User>] persistCache<Invoice>() !" fails to compile: No `extension` [Repository<Invoice>] found persistCache(UserRepository) !" java compatible persistCache(InvoiceRepository) !" compiles and runs because extension context is provided explicitly / @raulraja !" @47deg !" Sources !" Slides 51
52. KEEP-87 The Λrrow team plans to submit this proposal once it's solid and it has properly addressed feedback from the community and the jetbrains compiler team. / @raulraja !" @47deg !" Sources !" Slides 52
53. Credits Λrrow is inspired in great libraries that have proven useful to the FP community: • Cats • Scalaz • Freestyle • Monocle • Funktionale / @raulraja !" @47deg !" Sources !" Slides 53
54. Join us! Github https:!"github.com/arrow-kt/arrow Slack https:!"kotlinlang.slack.com/messages/ C5UPMM0A0 Gitter https:!"gitter.im/arrow-kt/Lobby We are beginner friendly and provide 1:1 mentoring for both users & new contributors! +90 Contributors and growing! / @raulraja !" @47deg !" Sources !" Slides 54
55. Join us at lambda.world for more FP in Kotlin! / @raulraja !" @47deg !" Sources !" Slides 55
56. Thanks! Thanks to everyone that makes Λrrow possible! / @raulraja !" @47deg !" Sources !" Slides 56