Lifted transformers

While Chimney transformers wrap total functions of type From => To, they don’t really support partial transformations, where depending on the input value, transformation may succeed or fail.

Let’s take a look at the following example.

case class RegistrationForm(email: String,
                            username: String,
                            password: String,
                            age: String)

case class RegisteredUser(email: String,
                          username: String,
                          passwordHash: String,
                          age: Int)

We get field age: String as an input, but we would like to parse it into correct Int or signal an error, if provided value is not valid integer. This is simply not possible with total Transformer. This is a moment when lifted transformers, provided by TransformerF type class come into play.

val okForm = RegistrationForm("", "John", "s3cr3t", "40")

  .intoF[Option, RegisteredUser] // (1)
  .withFieldComputed(_.passwordHash, form => hashpw(form.password))
  .withFieldComputedF(_.age, _.age.toIntOption) // (2)
  .transform // (3)
// Some(RegisteredUser("", "John", "...", 40)): Option[RegisteredUser]

There are few differences to total transformers in the example above:

  1. Instead of into[RegisteredUser], we use intoF[Option, RegisteredUser], which tells Chimney that Option type will be used for handling partial transformations.

  2. Instead of withFieldComputed, we use withFieldComputedF, where second parameter is a function that wraps result into a type constructor provided in (1) - Option in this case.

  3. Result type of transform call is not RegisteredUser, but Option[RegisteredUser].

As you expect, when provided age which is not valid integer, this code evaluates to None.

val badForm = RegistrationForm("", "John", "s3cr3t", "not an int")

  .intoF[Option, RegisteredUser]
  .withFieldComputed(_.passwordHash, form => hashpw(form.password))
  .withFieldComputedF(_.age, _.age.toIntOption)
// None: Option[RegisteredUser]

Lifted DSL operations

Similar to withFieldConst, withFieldComputed, withCoproductInstance operations in DSL, there are lifted counterparts available:

  • withFieldConstF

  • withFieldComputed

  • withCoproductInstanceF

Analogously to Transformer definition DSL for Transformer, we can define above transformation as implicit TransformerF[Option, RegistrationForm, RegisteredUser]. In order to do this, we use TransformerF.define (or equivalently Transformer.defineF).

implicit val transformer: TransformerF[Option, RegistrationForm, RegisteredUser] =
  TransformerF.define[Option, RegistrationForm, RegisteredUser]
    .withFieldComputed(_.passwordHash, form => hashpw(form.password))
    .withFieldComputedF(_.age, _.age.toIntOption)

As commonly, as with total transformers, this instance may be later picked up and used other, lifted transformations. In the following example it’s used for transforming array of registration forms into list of registered users.

Array(okForm, badForm).transformIntoF[Option, List[RegisteredUser]]
// None: Option[List[RegisteredUser]]

Note that following error handling semantics for collections, we’ve got None as a result (because not all of array elements were valid forms, according to the defined lifted transformer).

Capturing validation errors

Usually, when partial transformation failed, we would like to know why it failed. Thus, we must use different wrapper type than Option that allows to capture error information.

Chimney supports out of the box Either[C[E], +*], as the wrapper type, where

  • E - type of a single error occurrence

  • C[_] - collection type to store all the transformation errors (like Seq, Vector, List, etc.)

If we pick error type as String (as validation error message) and collection as Vector, we obtain wrapper type Either[Vector[String], +*].


Type syntax with +* is only available with kind-projector compiler plugin. If you don’t want to (or can’t) use it, you may either use type-lambda with weird syntax:

({type L[+X] = Either[Vector[String], X]})#L

or define type alias:

type EitherVecStr[+X] = Either[Vector[String], X]

and use type EitherVecStr as a lifted wrapper type.

Let’s enhance our RegistrationForm to RegisteredUser lifted transformer with few additional validation rules:

  • email field should contain @ character

  • age must be at least 18 years

implicit val transformer: TransformerF[EitherVecStr, RegistrationForm, RegisteredUser] = {
  Transformer.defineF[EitherVecStr, RegistrationForm, RegisteredUser]
    .withFieldComputedF(, form => {
      if('@')) {
      } else {
        Left(Vector(s"${form.username}'s email: does not contain '@' character"))
    .withFieldComputed(_.passwordHash, form => hashpw(form.password))
    .withFieldComputedF(_.age, form => form.age.toIntOption match {
      case Some(value) if value >= 18 => Right(value)
      case Some(value) => Left(Vector(s"${form.username}'s age: must have at least 18 years"))
      case None => Left(Vector(s"${form.username}'s age: invalid number"))

Then, trying to transform multiple registration forms, we can validate all them at once:

  RegistrationForm("", "John", "s3cr3t", "10"),
  RegistrationForm("", "Alice", "s3cr3t", "19"),
  RegistrationForm("", "Bob", "s3cr3t", "21.5")
).transformIntoF[EitherVecStr, List[RegisteredUser]]
// Left(
//   Vector(
//     "John's email: does not contain '@' character",
//     "John's age: must have at least 18 years",
//     "Bob's age: invalid number",
//   )
// )

In case when all the provided forms are correct, we obtain requested collection of registered users, wrapped in Right.

  RegistrationForm("", "John", "s3cr3t", "40"),
  RegistrationForm("", "Alice", "s3cr3t", "19"),
  RegistrationForm("", "Bob", "s3cr3t", "21")
).transformIntoF[EitherVecStr, List[RegisteredUser]]
// Right(
//   List(
//     RegisteredUser("", "John", "...", 40)
//     RegisteredUser("", "Alice", "...", 19),
//     RegisteredUser("", "Bob", "...", 21)
//   )
// )


Note that collection type where you gather errors is independent of any eventual collection types that takes part in the transformation.

For Either wrappers, Chimney supports practically any Scala standard collection type, but depending on your choice, you may obtain different performance characteristics. Thus, collections with reasonably fast concatenation should be preferred on the error channel.

If you prefer to use Cats library, you might be interested in Validated support for lifted transformers.

TransformerF type class

Similar to the Transformer type class, Chimney defines a TransformerF type class, which allows to express partial (lifted, wrapped) transformation of type From => F[To].

trait TransformerF[F[+_], From, To] {
  def transform(src: From): F[To]

The whole library functionality that refers to total transformers, is also supported for lifted transformers. This especially means:

  • local implicit instances of TransformerF are preferred in the first place, before deriving as instance by a macro (read more about it in Deriving lifted transformers)

  • all the enable/disable flags are respected by lifted transformers

  • you can customize lifted transformers using any operation described in Customizing transformers which works as well for total transformers, as for lifted ones

  • all the Standard transformers rules are provided for lifted transformers too

  • derivation for case classes, tuples, Java beans are supported too


Note that for convenience of some operations, F is defined with as covariant type constructor.

Supporting custom F[_]

Chimney provides pluggable interface that allows you to use your own F[_] type constructor in lifted transformations.

The library defines TransformerFSupport type class, as follows.

trait TransformerFSupport[F[+_]] {
  def pure[A](value: A): F[A]
  def product[A, B](fa: F[A], fb: => F[B]): F[(A, B)]
  def map[A, B](fa: F[A], f: A => B): F[B]
  def traverse[M, A, B](it: Iterator[A], f: A => F[B])(implicit fac: Factory[B, M]): F[M]


Chimney macros, during lifted transformer derivation, resolve implicit instance of TransformerFSupport for requested wrapper type constructor and use it in various places in emitted code.

In order to be able to use wrapper type of your choice, you need to implement an instance of TransformerFSupport and put it as implicit term in the scope of usage.

For those familiar with applicative functors and traversable type classes, implementation of these methods should be obvious. Yet it gives some choice about semantics of error handling.

Chimney supports Option, Either and (in Cats integration) just exactly by providing implicit instaces of TransformerFSupport implemented for those wrapper types.

Emitted code

Curious how the emitted code for lifted transformers looks like?

Let’s first refactor the transformation defined above, which is equivalent to the previous one, but with few functions extracted out - their implementation is not really important at this point.

def validateEmail(form: RegistrationForm): EitherVecStr[String] = ...
def computePasswordHash(form: RegistrationForm): String = ...
def validateAge(form: RegistrationForm): EitherVecStr[Int] = ...

implicit val transformer: TransformerF[EitherVecStr, RegistrationForm, RegisteredUser] = {
  Transformer.defineF[EitherVecStr, RegistrationForm, RegisteredUser]
    .withFieldComputedF(, validateEmail)
    .withFieldComputed(_.passwordHash, computePasswordHash)
    .withFieldComputedF(_.age, validateAge)

The .buildTransformer call generates implementation of TransformerF, which is semantically equivalent to the following, hand-crafted version.

implicit val transformer: TransformerF[EitherVecStr, RegistrationForm, RegisteredUser] = {

  val tfs: TransformerFSupport[EitherVecStr] = ... // resolved implicit instance

  new TransformerF[EitherVecStr, RegistrationForm, RegisteredUser] {
    def transform(form: RegistrationForm): EitherVecStr[RegisteredUser] = {
        tfs.product(validateEmail(form), validateAge(form)),
        { case (email: String, age: Int) =>

tfs.product is used to combine results of successful validations into a tuple type (email, age): (String, Int). In case that some validations failed, validation errors are combined together also by tfs.product.

Then, if all validations passed, transforms their results to a target value of type RegisteredUser. Otherwise, just passes validation errors as a final result.


  • only functions provided by withFieldComputedF are working with the wrapper type F

  • remaining fields transformations (indentity transformer for username and a function provided by withFieldComputed for password) work without any wrapping with F

This strategy leads to generating particularly efficient code.

Deriving lifted transformers

When deriving a TransformerF[F, From, To] instance, where:

  • type From consists of some type F1

  • type To consists of some type T1

  • F1 in From is a counterpart of T1 in To

…we need to have transformation from F1 to T1 in order to be able to derive requested TransformerF.

The rule is that:

  1. we first check for function F1 => F[T1] passed to lifted DSL operations (withFieldConstF, withFieldComputedF, etc.) or function F1 => T1 passed to total DSL operations (withFieldConst, withFieldComputed, etc.)

    • whichever was found, it’s used in the first place

    • the last one passed in DSL for given field/type wins

  2. then we look for implicit instances for TransformerF[F, F1, T1] and Transformer[F1, T1]

    • if both of them were found, ambiguity compilation error is reported

    • if only one of them was found, it’s used

  3. we try to derive total Transformer[F1, T1] using library rules

  4. we try to derive lifted TransformerF[F, F1, T1] using library rules