Monads composition API that just works. For OOP developers. It is well suited to compose also Kotlin nullables.

I know, we have Arrow that is the best functional library around. Anyway if you only want to do simple tasks, like validating your domain classes, Arrow is a bit of an overkill.

Also, Arrow is a real functional library, with a plenty of functional concepts that you need to digest before being productive. For the typical OOP developer, it has a quite steep learning curve.

Here it comes Konad. It has only three classes:

• Result: can be Result.Ok or Result.Errors.
• Validation: can be Validation.Success or Validation.Fail.
• Maybe: you know this... yet another Optional/Option/Nullable whatever. (But read the Maybe section below, it will get clear why we need it)

These are monads and applicative functors, so they implement the usual map, flatMap and ap methods.

Konad exists with the only purpose to let you easily compose these three classes.

Advanced use-cases examples are described here:

• Nice Kotlin Nullables and Where to Find Them
• Type-safe Domain Modeling in Kotlin

Add the dependency

add in pom.xml

<dependency>
    <groupId>io.github.lucapiccinelli</groupId>
    <artifactId>konad</artifactId>
    <version>1.2.6</version>
</dependency>

add in build.gradle

dependencies {
    implementation "io.github.lucapiccinelli:konad:1.2.6"
}

For an exaustive list of usage examples, please refer to test suite CreateNewUserTests.kt and to ResultTests.kt

Let's say you have a User class, that has an Email and a PhoneNumber. Email and PhoneNumber are built so that they can only be constructed using a factory method. It will return a Result.Errors type if the value passed is not valid.

data class User(val username: String, val email: Email, val phoneNumber: PhoneNumber, val firstname: String)

data class Email private constructor (val value: String) {
    companion object{
        fun of(emailValue: String): Result<Email> = if (Regex(EMAIL_REGEX).matches(emailValue))
            Email(emailValue).ok()
            else "$emailValue doesn't match an email format".error()
    }
}

data class PhoneNumber private constructor(val value: String){
    companion object {
        fun of(phoneNumberValue: String): Result<PhoneNumber> = if(Regex(PHONENUMBER_REGEX).matches(phoneNumberValue))
            PhoneNumber(phoneNumberValue).ok()
            else "$phoneNumberValue should match a valid phone number, but it doesn't".error()
    }
}

Email and PhoneNumber constructors are private, so that you can be sure that it can't exist a User with invalid contacts. However, the factory methods give you back a Result<Email>/Result<PhoneNumber>.

In order to compose them and get a Result<User> you have to do the following

    val userResult: Result<User> = ::User  
        "foo.bar"  
        Email.of("foo.bar")   // This email is invalid -> returns Result.Errors
        PhoneNumber.of("xxx")   // This phone number is invalid -> returns Result.Errors
        "Foo"
    
    when(userResult){
        is Result.Ok -> userResult.toString()
        is Result.Errors -> userResult.toList().joinToString(" - ")
    }.run(::println) // This is going to print "foo.bar doesn't match an email format - xxx should match a valid phone number, but it doesn't
    
    // or
    
    userResult
       .map{ user -> user.toString() }
       .ifError { errors -> errors.description(errorDescriptionsSeparator = " - ") }
       .run(::println)

Composition happens thanks to concepts named functors and applicative Functors.

I chose to stay simple and practical, then all the methods that implement composition are called on (See applicativeBuilders.kt). However, for those who love the functional naming, you can choose this other style. (See applicativeBuildersPureStyle.kt)

    val user: Result<User> = ::User.curry()
        .apply("foo.bar")
        .map(Email.of("foo.bar")) 
        .ap(PhoneNumber.of("xxx"))
        .pure("Foo")
        .result

Maybe is needed only to wrap Kotlin nullables and bring them to a higher-kinded type (see unaryHigherKindedTypes.kt). In this way on, can be used to compose nullables.

Its constructor is private because you should avoid using it in order to express optionality. Kotlin nullability is perfect for that purpose.

If ever you tried to compose nullables in Kotlin, then probably you ended up having something like the following

val foo: Int? = 1
val bar: String? = "2"
val baz: Float? = 3.0f

fun useThem(x: Int, y: String, z: Float): Int = x   y.toInt()   z.toInt()

val result1: Int? = foo
    ?.let { bar
    ?.let { baz
    ?.let { useThem(foo, bar, baz) } } }

// or

val result2: Int? = if(foo != null && bar != null && baz != null) 
    useThem(foo, bar, baz) 
    else null

This is not very clean. And it gets even worse if would like to give an error message when a null happens.

Using Konad, nullables can be composed as follows

val result: Int? = ::useThem   foo   bar   baz

or you can choose to give an explanatory message when something is null

val result: Result<Int> = ::useThem  
    foo.ifNull("Foo should not be null")  
    bar.ifNull("Bar should not be null")  
    baz.ifNull("Baz should not be null")

Validation<A, B> is like an Either monad, but with the left case accumulation. It is similar to Result<T> but instead of fixing the error case as a string description, it lets you decide how you represent the error. Example:

sealed class ResourceError {
    data class BadInput(val description: String) : ResourceError()
    object NotFound : ResourceError()
    object Forbidden : ResourceError()
}

fun readUser(id: String): Validation<ResourceError, User> =
    if (id.isBlank()) ResourceError.BadInput("id should not be blank").fail()
    else repository.findById(id)?.success() ?: ResourceError.NotFound.fail()

readUser("xxx")
    .map { user: User -> println(user) }
    .ifFail { failures: Collection<ResourceError> -> println(failures) }

What if you have a List<Result<T>> and you want a Result<List<T>>? Then use flatten extension method.

val r: Result<Collection<Int>> = listOf(Result.Ok(1), Result.Ok(2)).flatten()

Errors get cumulated as usual

 val r: Result<Collection<Int>> = listOf(Result.Errors("error1"), Result.Ok(1), Result.Errors("error2"))
    .flatten()

when(r){
    is Result.Ok -> r.value.toString()
    is Result.Errors -> r.description
}.run(::println) // will print error1 - error2

Obviously it works also on nullables: Collection<T?> -> Collection<T>?

val flattened = setOf("a", null, "c").flatten()

flattened shouldBe null

and on Validation

val v: Validation<String, Collection<Int>> = listOf("error1".fail(), 1.success(), "error2".fail()).flatten()

Sometime you need to add some details on an error, or to transform it. Result and Validation monads have convenience method for this case. Examples:

fun checkNotEmpty(value: String) = if(value.isBlank()) "value should not be blank".error() else value.ok()

data class User private constructor(val firstName: String, val lastname: String){
    companion object{
        fun of(firstname: String, lastname: String): Result<User> = ::User.curry()
            .on(checkNotEmpty(firstname))
            .on(checkNotEmpty(lastname))
            .result
    }
}

in this example, if both firstname and lastname are blank, then you will get two errors. Unfortunately both of those errors will have the same description, and you will not be able to distinct which value should not be empty. To fix, there is the method Result::errorTitle

fun of(firstname: String, lastname: String): Result<User> = ::User.curry()
    .on(checkNotEmpty(firstname).errorTitle("firstname"))
    .on(checkNotEmpty(lastname).errorTitle("lastname"))
    .result

You can find a more detailed specification here: ResultTests

Similarly, Validation has the mapFail method, to apply a tranformation on the error case. Examples here ValidationTests

In case of accumulated errors, both errorTitle and mapFail are applied to the entire list of errors.

Since version 1.2.3, there exist an extension method Result<T>.field that enables to add an error title in a type-safe manner.

fun of(firstname: String, lastname: String): Result<User> = ::User  
    checkNotEmpty(firstname).field(User::firstname)   
    checkNotEmpty(lastname).field(User::lastname)

In this example, field will add the name of the property as an error title, while also checking at compile time if the type of the property matches the type of the corresponding constructor parameter

If you wish to implement your own monads and let them be composable through the on Konad applicative builders, then you need to implement the interfaces that are here: Higher-kinded types

Actually, to let your type be composable, it is enough to implement the ApplicativeFunctorKind interface.

Kotlin doesn't natively supports Higher-kinded types. To implement them, Konad is inspired on how those are implemented in Arrow. That is why there is the need of .result and .nullable extension properties.