84
votes

I want to convert/map some "data" class objects to similar "data" class objects. For example, classes for web form to classes for database records.

data class PersonForm(
    val firstName: String,
    val lastName: String,
    val age: Int,
    // maybe many fields exist here like address, card number, etc.
    val tel: String
)
// maps to ...
data class PersonRecord(
    val name: String, // "${firstName} ${lastName}"
    val age: Int, // copy of age
    // maybe many fields exist here like address, card number, etc.
    val tel: String // copy of tel
)

I use ModelMapper for such works in Java, but it can't be used because data classes are final (ModelMapper creates CGLib proxies to read mapping definitions). We can use ModelMapper when we make these classes/fields open, but we must implement features of "data" class manually. (cf. ModelMapper examples: https://github.com/jhalterman/modelmapper/blob/master/examples/src/main/java/org/modelmapper/gettingstarted/GettingStartedExample.java)

How to map such "data" objects in Kotlin?

Update: ModelMapper automatically maps fields that have same name (like tel -> tel) without mapping declarations. I want to do it with data class of Kotlin.

Update: The purpose of each classes depends on what kind of application, but these are probably placed in the different layer of an application.

For example:

  • data from database (Database Entity) to data for HTML form (Model/View Model)
  • REST API result to data for database

These classes are similar, but are not the same.

I want to avoid normal function calls for these reasons:

  • It depends on the order of arguments. A function for a class with many fields that have same type (like String) will be broken easily.
  • Many declarations are nesessary though most mappings can be resolved with naming convention.

Of course, a library that has similar feature is intended, but information of the Kotlin feature is also welcome (like spreading in ECMAScript).

8
Please describe how you want to use the mapped classes. What is the purpose of having two separate data formats?voddan
Never heard of duplication the data model (except for cases of legacy code). Usually the data you are working with (View Model) is the data you put into the database.voddan
@voddan one use case would be to expose only parts of the domain model to different API consumers. Having separate DTO per view of domain model is much cleaner than using i.e. JsonView IMHOmiensol
Do you really need these classes to be data classes?Alexey Andreev

8 Answers

74
votes
  1. Simplest (best?):

    fun PersonForm.toPersonRecord() = PersonRecord(
            name = "$firstName $lastName",
            age = age,
            tel = tel
    )
    
  2. Reflection (not great performance):

    fun PersonForm.toPersonRecord() = with(PersonRecord::class.primaryConstructor!!) {
        val propertiesByName = PersonForm::class.memberProperties.associateBy { it.name }
        callBy(args = parameters.associate { parameter ->
            parameter to when (parameter.name) {
                "name" -> "$firstName $lastName"
                else -> propertiesByName[parameter.name]?.get(this@toPersonRecord)
            }
        })
    }
    
  3. Cached reflection (okay performance but not as fast as #1):

    open class Transformer<in T : Any, out R : Any>
    protected constructor(inClass: KClass<T>, outClass: KClass<R>) {
        private val outConstructor = outClass.primaryConstructor!!
        private val inPropertiesByName by lazy {
            inClass.memberProperties.associateBy { it.name }
        }
    
        fun transform(data: T): R = with(outConstructor) {
            callBy(parameters.associate { parameter ->
                parameter to argFor(parameter, data)
            })
        }
    
        open fun argFor(parameter: KParameter, data: T): Any? {
            return inPropertiesByName[parameter.name]?.get(data)
        }
    }
    
    val personFormToPersonRecordTransformer = object
    : Transformer<PersonForm, PersonRecord>(PersonForm::class, PersonRecord::class) {
        override fun argFor(parameter: KParameter, data: PersonForm): Any? {
            return when (parameter.name) {
                "name" -> with(data) { "$firstName $lastName" }
                else -> super.argFor(parameter, data)
            }
        }
    }
    
    fun PersonForm.toPersonRecord() = personFormToPersonRecordTransformer.transform(this)
    
  4. Storing Properties in a Map

    data class PersonForm(val map: Map<String, Any?>) {
        val firstName: String   by map
        val lastName: String    by map
        val age: Int            by map
        // maybe many fields exist here like address, card number, etc.
        val tel: String         by map
    }
    
    // maps to ...
    data class PersonRecord(val map: Map<String, Any?>) {
        val name: String    by map // "${firstName} ${lastName}"
        val age: Int        by map // copy of age
        // maybe many fields exist here like address, card number, etc.
        val tel: String     by map // copy of tel
    }
    
    fun PersonForm.toPersonRecord() = PersonRecord(HashMap(map).apply {
        this["name"] = "${remove("firstName")} ${remove("lastName")}"
    })
    
23
votes

Is this are you looking for?

data class PersonRecord(val name: String, val age: Int, val tel: String){       
    object ModelMapper {
        fun from(form: PersonForm) = 
            PersonRecord(form.firstName + form.lastName, form.age, form.tel)           
    }
}

and then:

val personRecord = PersonRecord.ModelMapper.from(personForm)
16
votes

MapStruct lets kapt generate classes doing the mapping (without reflection).

Use MapStruct:

@Mapper
interface PersonConverter {

    @Mapping(source = "phoneNumber", target = "phone")
    fun convertToDto(person: Person) : PersonDto

    @InheritInverseConfiguration
    fun convertToModel(personDto: PersonDto) : Person

}


// Note this either needs empty constructor or we need @KotlinBuilder as dsecribe below
data class Person: this(null, null, null, null) (...)

Use:

val converter = Mappers.getMapper(PersonConverter::class.java) // or PersonConverterImpl()

val person = Person("Samuel", "Jackson", "0123 334466", LocalDate.of(1948, 12, 21))

val personDto = converter.convertToDto(person)
println(personDto)

val personModel = converter.convertToModel(personDto)
println(personModel)

Edit:

Now with @KotlinBuilder for avoiding constructor() issue:

GitHub: Pozo's mapstruct-kotlin

Annotate data classes with @KotlinBuilder. This will create a PersonBuilder class, which MapStruct uses, thus we avoid ruining the interface of the data class with a constructor().

@KotlinBuilder
data class Person(
    val firstName: String,
    val lastName: String,
    val age: Int,
    val tel: String
)

Dependency :

// https://mvnrepository.com/artifact/com.github.pozo/mapstruct-kotlin
api("com.github.pozo:mapstruct-kotlin:1.3.1.1")
kapt("com.github.pozo:mapstruct-kotlin-processor:1.3.1.1")

https://github.com/mapstruct/mapstruct-examples/tree/master/mapstruct-kotlin

4
votes

Do you really want a separate class for that? You can add properties to the original data class:

data class PersonForm(
    val firstName: String,
    val lastName: String,
    val age: Int,
    val tel: String
) {
    val name = "${firstName} ${lastName}"
}
4
votes

Using ModelMapper

/** Util.kt **/

class MapperDto() : ModelMapper() {
    init {
        configuration.matchingStrategy = MatchingStrategies.LOOSE
        configuration.fieldAccessLevel = Configuration.AccessLevel.PRIVATE
        configuration.isFieldMatchingEnabled = true
        configuration.isSkipNullEnabled = true
    }
}

object Mapper {
    val mapper = MapperDto()

    inline fun <S, reified T> convert(source: S): T = mapper.map(source, T::class.java)
}

Usage

val form = PersonForm(/** ... **/)
val record: PersonRecord = Mapper.convert(form)

You might need some mapping rules if the field names differ. See the getting started
PS: Use kotlin no-args plugin for having default no-arg constructor with your data classes

3
votes

This works using Gson:

inline fun <reified T : Any> Any.mapTo(): T =
    GsonBuilder().create().run {
        toJson(this@mapTo).let { fromJson(it, T::class.java) }
    }

fun PersonForm.toRecord(): PersonRecord =
    mapTo<PersonRecord>().copy(
        name = "$firstName $lastName"
    )

fun PersonRecord.toForm(): PersonForm =
    mapTo<PersonForm>().copy(
        firstName = name.split(" ").first(),
        lastName = name.split(" ").last()
    )

with not nullable values allowed to be null because Gson uses sun.misc.Unsafe..

1
votes

You can use ModelMapper to map to a Kotlin data class. The keys are:

  • Use @JvmOverloads (generates a constructor with no arguments)
  • Default values for data class member
  • Mutable member, var instead of val

    data class AppSyncEvent @JvmOverloads constructor(
        var field: String = "",
        var arguments: Map<String, *> = mapOf<String, Any>(),
        var source: Map<String, *> = mapOf<String, Any>()
    )
    
    val event = ModelMapper().map(request, AppSyncEvent::class.java)
    
1
votes

For ModelMapper you could use Kotlin's no-arg compiler plugin, with which you can create an annotation that marks your data class to get a synthetic no-arg constructor for libraries that use reflection. Your data class needs to use var instead of val.

package com.example

annotation class NoArg

@NoArg
data class MyData(var myDatum: String)

mm.map(. . ., MyData::class.java)

and in build.gradle (see docs for Maven):

buildscript {
  . . .
  dependencies {
    classpath "org.jetbrains.kotlin:kotlin-gradle-plugin:$kotlinVersion"
  }
}

apply plugin: 'kotlin-noarg'

noArg {
  annotation "com.example.NoArg"
}