kotlin-sum-types（科特林求和类型）

STARTER_CHARACTER = 🔀

Parse, Don't Validate with Kotlin Sealed Classes

Represent validation states explicitly using sealed classes. This makes invalid states unrepresentable in your domain logic and pushes validation to system boundaries.

Core Principle

Parse, don't validate means transforming untyped input into strongly-typed domain objects at the boundary, carrying proof of validity through the type system.

Instead of:

fun processEmail(email: String) {
    if (!isValid(email)) throw ValidationException()
    // Every function must revalidate or assume validity
}

Do this:

fun processEmail(email: ValidEmail) {
    // Type proves it's valid, no need to check
}

Basic Pattern: Valid/Invalid States

Use sealed classes to represent parsed data that can be either valid or invalid:

sealed class Email {
    data class ValidEmail(
        val user: String,
        val domain: String,
    ) : Email() {
        fun stringRepresentation(): String = "$user@$domain"
    }

    data class InvalidEmail(
        val value: String,
        val _errors: List<ValidationError>,
    ) : Email(), InvalidDataClass {
        override fun getErrors(): List<ValidationError> = _errors
    }

    companion object {
        @JvmStatic
        @JsonCreator
        fun create(createValue: String): Email =
            if (createValue.contains("@")) {
                createValue.split("@").let { ValidEmail(it.first(), it.last()) }
            } else {
                InvalidEmail(
                    createValue,
                    listOf(ValidationError("", "Not a valid Email", createValue))
                )
            }
    }
}

Key elements:

• Sealed class as parent (Email)
• ValidEmail with parsed structure (user, domain)
• InvalidEmail preserving original value + errors
• Static create() factory for parsing
• @JsonCreator for Jackson integration

Validation Error Model

Standard error representation:

data class ValidationError(
    val path: String,      // Field path (e.g., "address.city")
    val message: String,   // Human-readable message
    val value: String,     // The invalid value
)

interface InvalidDataClass {
    fun hasErrors(): Boolean = getErrors().isNotEmpty()
    fun getErrors(): List<ValidationError>
}

All invalid states implement InvalidDataClass to expose errors uniformly.

Composite Validation

Build complex types by composing validated types:

sealed class Address {
    data class ValidAddress(
        val streetName: String,
        val city: String,
        val postCode: String,
        val country: String,
    ) : Address()

    data class InvalidAddress(
        val streetName: String?,
        val city: String?,
        val postCode: String?,
        val country: String?,
        val _errors: List<ValidationError>,
    ) : Address(), InvalidDataClass {
        override fun getErrors(): List<ValidationError> = _errors
    }

    companion object {
        @JvmStatic
        @JsonCreator
        fun create(
            streetName: String?,
            city: String?,
            postCode: String?,
            country: String?,
        ): Address {
            if (streetName.isNullOrEmpty() || city.isNullOrEmpty() ||
                postCode.isNullOrBlank() || country.isNullOrBlank()) {
                return InvalidAddress(
                    streetName, city, postCode, country,
                    listOf(ValidationError("", "Missing required fields", "..."))
                )
            }
            return ValidAddress(streetName, city, postCode, country)
        }
    }
}

InvalidAddress preserves all input: Even invalid data is kept so you can return meaningful error messages to users.

Nested Valid States

Valid states can have their own hierarchy:

sealed class RegistrationForm {
    data class Invalid(
        val email: Email,
        val anonymous: Boolean,
        val name: String?,
        val address: Address?,
        val _errors: List<ValidationError>,
    ) : RegistrationForm(), InvalidDataClass {
        override fun getErrors(): List<ValidationError> = _errors
    }

    sealed class Valid(
        open val email: Email.ValidEmail,  // Only ValidEmail allowed!
    ) : RegistrationForm() {
        data class AnonymousRegistration(
            val _email: Email.ValidEmail,
        ) : Valid(_email)

        data class Registration(
            val _email: Email.ValidEmail,
            val name: String,
            val address: Address.ValidAddress,  // Only ValidAddress allowed!
        ) : Valid(_email)
    }

    companion object {
        @JvmStatic
        @JsonCreator
        fun create(
            email: Email,
            anonymous: Boolean,
            name: String?,
            address: Address?,
        ): RegistrationForm {
            // Collect errors from nested validated types
            val errors = mapOf("email" to email, "address" to address)
                .filter { it.value is InvalidDataClass }
                .flatMap { (key, value) ->
                    (value as InvalidDataClass).getErrors()
                        .map { error ->
                            error.copy(
                                path = key + if (error.path.isNotEmpty()) ".${error.path}" else ""
                            )
                        }
                }

            return when {
                errors.isNotEmpty() -> Invalid(email, anonymous, name, address, errors)
                anonymous -> Valid.AnonymousRegistration(email as Email.ValidEmail)
                name != null -> Valid.Registration(
                    email as Email.ValidEmail,
                    name,
                    address as Address.ValidAddress
                )
                else -> Invalid(
                    email, anonymous, name, address,
                    listOf(ValidationError("", "Invalid combination", ""))
                )
            }
        }
    }
}

Note the types:

• Valid.Registration requires Email.ValidEmail and Address.ValidAddress
• Invalid case can contain any Email and Address (valid or invalid)
• Errors are propagated with paths ("email", "address.city")

Controller Pattern

Handle valid/invalid cases at the boundary using when:

sealed class ControllerResponse {
    data class OkResponse(val result: String) : ControllerResponse()
    data class ErrorResponse(val errors: List<ValidationError>) : ControllerResponse()
}

class RegistrationController(
    private val registrationService: RegistrationService,
) {
    private val mapper = jacksonObjectMapper()

    fun registerUser(jsonString: String): ControllerResponse =
        when (val parsed: RegistrationForm = mapper.readValue(jsonString)) {
            is RegistrationForm.Valid -> {
                registrationService.createNewRegistration(parsed)
                when (parsed) {
                    is RegistrationForm.Valid.Registration ->
                        ControllerResponse.OkResponse("Congrats ${parsed.name}!")
                    is RegistrationForm.Valid.AnonymousRegistration ->
                        ControllerResponse.OkResponse("Congrats!")
                }
            }
            is RegistrationForm.Invalid ->
                ControllerResponse.ErrorResponse(parsed.getErrors())
        }
}

Key aspects:

• Parse JSON at the boundary
• when expression handles valid/invalid exhaustively
• Service receives only valid types
• Errors automatically collected and returned

Jackson Integration

Use @JsonCreator to hook into Jackson parsing:

companion object {
    @JvmStatic
    @JsonCreator
    fun create(param1: Type1, param2: Type2): SealedClass {
        // Validation logic here
    }
}

Jackson calls create() during deserialization, giving you control over validation.

Dependencies:

implementation("com.fasterxml.jackson.module:jackson-module-kotlin")

Usage:

val mapper = jacksonObjectMapper()
val parsed: RegistrationForm = mapper.readValue(jsonString)
// parsed is either Valid or Invalid

Benefits

Type Safety:

• Invalid states are unrepresentable in domain logic
• Compiler prevents passing invalid data to functions expecting valid types
• when expressions ensure all cases are handled

Error Collection:

• Multiple validation errors collected in one pass
• Nested errors preserve field paths
• Original invalid values preserved for debugging

Maintainability:

• Validation logic centralized in create() factories
• Domain logic operates only on valid types
• Clear boundary between validated and unvalidated data

Refactoring Safety:

• Adding new fields to valid types is a compiler error if not handled
• Changing validation rules doesn't affect domain logic
• IDE autocomplete shows all valid/invalid states

When to Use This Pattern

Use sealed classes for validation when:

• Parsing external input (JSON, CSV, user forms)
• Multiple fields must be validated together
• You need to collect multiple validation errors
• Invalid data should be preserved for error reporting
• Validation rules are complex or change frequently

Don't use when:

• Simple non-null checks (use Kotlin's ? types)
• Single-field validation with no composition
• Data is already validated by external system (database constraints)
• Performance is critical (adds allocation overhead)

Testing Strategy

Test valid state creation:

@Test
fun shouldParseValidEmail() {
    val parsed = Email.create("user@example.com")

    assertThat(parsed).isInstanceOf(Email.ValidEmail::class.java)
    (parsed as Email.ValidEmail).let {
        assertThat(it.user).isEqualTo("user")
        assertThat(it.domain).isEqualTo("example.com")
    }
}

Test invalid state creation:

@Test
fun shouldParseInvalidEmail() {
    val parsed = Email.create("not-an-email")

    assertThat(parsed).isInstanceOf(Email.InvalidEmail::class.java)
    (parsed as Email.InvalidEmail).let {
        assertThat(it.value).isEqualTo("not-an-email")
        assertThat(it.getErrors()).isNotEmpty()
    }
}

Test composite validation:

@Test
fun shouldCollectNestedErrors() {
    val form = RegistrationForm.create(
        email = Email.create("invalid"),
        anonymous = false,
        name = null,
        address = Address.create(null, null, null, null)
    )

    assertThat(form).isInstanceOf(RegistrationForm.Invalid::class.java)
    (form as RegistrationForm.Invalid).let {
        val errorPaths = it.getErrors().map { e -> e.path }
        assertThat(errorPaths).contains("email", "address")
    }
}

Test controller handling:

@Test
fun shouldReturnErrorResponseForInvalidInput() {
    val response = controller.registerUser("""{"email": "bad"}""")

    assertThat(response).isInstanceOf(ControllerResponse.ErrorResponse::class.java)
    (response as ControllerResponse.ErrorResponse).let {
        assertThat(it.errors).isNotEmpty()
    }
}

Real-World Example

See complete working example:

• Domain: RegistrationDomain.kt
• Controller: ControllerLikeRegistrationController.kt
• Tests: ParsingTest.kt

Anti-Patterns

Don't validate in the domain:

// BAD - validation scattered throughout domain
fun processRegistration(email: String, name: String) {
    require(email.contains("@")) { "Invalid email" }
    require(name.isNotBlank()) { "Name required" }
    // ... business logic
}

Do validation at boundaries, pass validated types to domain:

// GOOD - validation at boundary, domain receives valid types
fun processRegistration(registration: RegistrationForm.Valid.Registration) {
    // registration.email is ValidEmail, no validation needed
}

Don't lose original invalid values:

// BAD - can't tell user what they sent
data class InvalidEmail(val _errors: List<ValidationError>) : Email()

Keep the original value:

// GOOD - can show user what they sent
data class InvalidEmail(
    val value: String,  // Preserve original input
    val _errors: List<ValidationError>
) : Email()

Don't use exceptions for expected validation:

// BAD - exceptions for expected invalid input
fun create(email: String): ValidEmail {
    if (!isValid(email)) throw ValidationException()
    return ValidEmail(...)
}

Return sealed class representing valid/invalid:

// GOOD - invalid input is expected, not exceptional
fun create(email: String): Email {  // Returns Valid or Invalid
    if (!isValid(email)) return InvalidEmail(...)
    return ValidEmail(...)
}

Related Patterns

• Railway-Oriented Programming: Result<T, E> types (similar but more functional)
• Type-Driven Development: Using types to guide design
• Domain-Driven Design: Validation at aggregate boundaries

References

• Parse, don't validate by Alexis King
• Making Illegal States Unrepresentable by Scott Wlaschin
• Type-Driven Development talk at JavaZone 2022