Skip to content

Troubleshooting

Already using Chimney and you've got some issues? This page might help you with it.

Migration from 0.7.x to 0.8.0

Version 0.8.0 is the first version that cleaned up the API. It introduced several breaking changes.

Replacing Lifted Transformers (TransformerF) with PartialTransformers

Lifted Transformers (TransformerT), deprecated in 0.7.0, got removed in favor of PartialTransformers

Chimney's Lifted Transformers were historically the first experimental attempt to express transformations that may potentially fail. Despite their great expressiveness, they lacked several basic features and had a few design flaws that make them unattractive/difficult for wider adoption.

Let's have a look at the type signatures of both Lifted and Partial Transformers.

Example

package io.scalaland.chimney

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

// Partial Transformer
// partial comes from io.scalaland.chimney.partial
trait PartialTransformer[From, To] {
   def transform(src: From, failFast: Boolean): partial.Result[To]
}
  • Lifted Transformers provided abstraction over the target transformation type container (F[+_]), while Partial Transformers fix resulting type to built-in partial.Result[_]
    • as a consequence of this abstraction, Lifted Transformers required a type class instance (TransformerFSupport) in scope for every specific F[_+] used
    • Partial Transformers rely on built-in behavior and provide convenience methods to convert between more familiar data types (Option, Either, etc.)
    • abstraction over the resulting container type in the Lifted Transformer allowed for having custom error types; this is not easily possible with Partial Transformer, which focuses on a few most common error types
  • Partial Transformer has built-in support for fail-fast (short-circuiting) semantics by passing failFast boolean parameter, while in Lifted Transformers it was barely possible (only by providing a supporting type class that had such a fixed behavior)
  • Error path support in Lifted Transformers required providing another type class instance (TransformerFErrorPathSupport) for your error collection type, while in partial transformers it is a built-in feature

To migrate your code from Lifted Transformers to Partial Transformers, you may take the following steps.

  • replace all the occurrences of TransformerF type with PartialTransformer and remove the first type argument (F[_]) which is not used for Partial Transformers.
  • for your transformations find corresponding DSL methods. Their name usually differs on the suffix, for example:
    • replace withFieldConstF with withFieldConstPartial
    • replace withFieldComputedF with withFieldComputedPartial
    • etc.
  • adjust the types passed to the customization methods. In Lifted Transformers, they were expecting values of your custom type F[T], while in Partial Transformers they work with partial.Result[T]. See the partial.Result companion object for ways of constructing success and failure instances, for example:
    • partial.Result.fromValue
    • partial.Result.fromOption
    • partial.Result.fromEither
    • partial.Result.fromTry
    • and so on...
  • the resulting type of a call to .transform is also a partial.Result[T]. If you don't want to work with partial.Result directly, figure out ways to convert it to other, more familiar data structures. Some of the ways may include:
    • result.asOption
    • result.asEither
    • result.asErrorPathMessages
    • other data structures using Cats' integration

Removal of .enableUnsafeOption

.enableUnsafeOptionwas removed - if Option unwrapping is needed, it is recommended to use PartialTransformer

This option allowed calling .get on Option to enable conversion from Option to non-`Option:

Example

//> using dep io.scalaland::chimney::0.8.1
import io.scalaland.chimney.dsl._

case class Foo(a: Option[String])
case class Bar(a: String)

Foo(Some("value")).into[Bar].enableUnsafeOption.transform // Bar("value")
Foo(None).into[Bar].enableUnsafeOption.transform // throws Exception

Throwing exceptions made sense as a workaround in simpler times, when Transformers were the only option. However, now we have PartialTransformers. They have a build-in ability to unwrap Option as failed result.

Example

//> using dep io.scalaland::chimney::0.8.1
import io.scalaland.chimney.dsl._

case class Foo(a: Option[String])
case class Bar(a: String)

Foo(Some("value")).transformIntoPartial[Bar](failFast = true).asOption // Some(Bar("value"))
Foo(None).transformIntoPartial[Bar](failFast = true).asOption // None

With failFast = true and .asOption Partial Transformers have similar semantics to Transformer with unsafe Option but the result is an explicit Option instead of implied exception handling.

Changes to automatic derivation logic

Types returned by automatic derivation got split from types that are used for user-provided transformations and configured (semiautomatic) derivation: Transformer got split into Transformer and Transformer.AutoDerived while PartialTransformer got split into PartialTransformer and PartialTransformer.AutoDerived.

It was caused by the change in the mechanism for recursive derivation: since Chimney avoid boxing and allocation where possible, it is used to check if summoned implicit was generated by automatic derivation. If implicit came from Transformer.derive or PartialTransformer.derive it was discarded and the macro attempted to derive it again without wrapping the result in a type class.

It both complicated code and increased compilation times, as each field or subtype would attempt to summon implicit (potentially triggering macro expansion) and then discard it if it didn't come from the user. Splitting types allows the compiler to not summon any implicit if the user hasn't provided any.

The consequence is only visible if there is some implicit def which takes another implicit Transformer.

Example

//> using dep io.scalaland::chimney::0.8.1
import io.scalaland.chimney._

class MyType[A](private val a: A) {
  def map[B](f: A => B): MyType[B] =
    new MyType(f(a))
}

implicit def provideMyType[A, B](
    implicit a2b: Transformer[A, B]
): Transformer[MyType[A], MyType[B]] =
  myA => myA.map(_.transformInto[B])

After changes in 0.8.x implicit Transformer[A, B] means "instance provided by user", either manually or through semiautomatic derivation. If the user wants to allow summoning there the automatic instances as well, they need to use Transformer.AutoDerived:

Example

//> using dep io.scalaland::chimney::0.8.1
import io.scalaland.chimney._

class MyOtherType[A](private val a: A) {
  def map[B](f: A => B): MyOtherType[B] =
    new MyOtherType(f(a))
}

implicit def provideMyOtherType[A, B](
    implicit a2b: Transformer.AutoDerived[A, B]
): Transformer[MyOtherType[A], MyOtherType[B]] =
  myA => myA.map(_.transformInto[B])

which would summon both automatically derived instances and manually provided ones. The difference is shown in this example:

Example

//> using dep io.scalaland::chimney::0.8.1
import io.scalaland.chimney.dsl._

// implicit provided by the user
implicit val int2str: Transformer[Int, String] = _.toString

val myType: MyType[Int] = new MyType(10)
val myOtherType: MyOtherType[Int] = new MyOtherType(10)

// uses provideMyType(int2str):
myType.transformInto[MyType[String]]

// uses provideMyOtherType(int2str):
myOtherType.transformInto[MyOtherType[String]

val myType2: MyType[Either[Int, Int]] = new MyType(Right(10))
val myOtherType2: MyOtherType[Either[Int, Int]] = new MyOtherType(Right(10)

// requires manually provided transformer e.g.
//   implicit val either2either =
//     Transformer.derive[Either[Int, Int], Either[String, String]]
// without it, the compilation fails
// myType2.transformInto[MyType[Either[String, String]]]

// uses provideMyOtherType(Transformer.derive):
myOtherType2.transformInto[Either[String, String]]

Default values no longer are used as fallback if the source field exists

If:

  • default values were enabled,
  • source and target had fields of the same name
  • this field had default value defined
  • macro couldn't derive transformation from source field type to target field type

Chimney used to use the default value.

However, this was a buggy behavior, and currently, it only uses default values if there is no source field nor other fallback or override. Although it is a bugfix, it is also a breaking change so it has to be documented. The fix would be a manual resolution for all fields which now (correctly) fail due to the bugfix.

Recursive types fail to compile

Chimney attempts to avoid unnecessary memory allocations for good performance.

It means that the code foo.into[Bar].transform would try to avoid the creation of Transformer[Foo, Bar] - if the user provided one it would have to be used, but if the only available implicit would come from automatic derivation, it would be ignored so that macro would generate an inlined expression.

This isn't possible with recursive types, as you cannot inline potentially unbounded nesting of transformations. For them, it is suggested to derive the Transformer, assigning it to implicit val/implicit def so that recursive transformations would be handled by recursive calls. This can be done with:

Example

implicit val foo2bar: Transformer[Foo, Bar] = Transformer.derive[Foo, Bar]

// or

implicit val foo2bar: Transformer[Foo, Bar] = Transformer.define[Foo, Bar].buildTransformer

and then

Example

foo.transformInto[Bar] // uses implicit Transformer (with recursive transformation)

The same is true for partial transformers.

Recursive calls on implicits

Old versions of Chimney in situations like this:

Example

implicit val t: Transformer[Foo, Bar] = foo => foo.transformInto[Bar] // or
implicit val t: Transformer[Foo, Bar] = foo => foo.into[Bar].transform

would result in errors like:

Example

forward reference extends over definition of value t

In newer, it can result in would result in errors like:

Example

java.lang.StackOverflowError

It's a sign of recursion which has to be handled with semiautomatic derivation.

Example

implicit val t: Transformer[Foo, Bar] = Transformer.derive[Foo, Bar] // or
implicit val t: Transformer[Foo, Bar] = Transformer.define[Foo, Bar].buildTransformer

sealed traits fail to recompile

In the case of incremental compilation, the Zinc compiler sometimes has issues with caching certain kind of information and macros don't get proper information from knownDirectSubclasses method. It usually helps when you clean and compile again. It cannot be fixed in the library as it relies on the compiler to provide it with this data, and the compiler fails to do so.

On Scala 2.12.0 it failed in other cases as well, so it is recommended to update 2.12 to at least 2.12.1.

Scala 3 complains that implicit/given TransformerConfiguration needs an explicit return type

In Scala 2 syntax like

Example

//> using dep io.scalaland::chimney::0.8.1
import io.scalaland.chimney.dsl._

implicit def cfg = TransformerConfiguration.default.enableMacrosLogging

was perfectly OK. Using implicits without a type was a bad practice but not an error.

This changes in Scala 3 where you'll get an error:

Example

result type of implicit definition needs to be given explicitly

You can work around this by slightly longer incantation:

Example

//> using dep io.scalaland::chimney::0.8.1
import io.scalaland.chimney.dsl._

transparent inline given TransformerConfiguration[?] =
  TransformerConfiguration.default.enableMacrosLogging

Debugging macros

In some cases, it could be helpful to preview what is the expression generated by macros, which implicits were used in macro (or not) and what was the exact logic that lead to the final expression or compilation errors.

In such cases, we can use a dedicated flag, .enableMacrosLogging:

Example

//> using dep io.scalaland::chimney::0.8.1
import io.scalaland.chimney.dsl._

case class Foo(x: String, y: Int, z: Boolean = true)
case class Bar(x: String, y: Int)

Bar("abc", 10).into[Foo].enableDefaultValues.enableMacrosLogging.transform

For the snippet above, the macro could print this structured log:

Example

+ Start derivation with context: ForTotal[From = Bar, To = Foo](src = bar)(TransformerConfig(
|   flags = Flags(processDefaultValues, displayMacrosLogging),
|   fieldOverrides = Map(),
|   coproductOverrides = Map(),
|   preventResolutionForTypes = None
| ))
+ Deriving Total Transformer expression from Bar to Foo
  + Attempting expansion of rule Implicit
  + Rule Implicit decided to pass on to the next rule
  + Attempting expansion of rule Subtypes
  + Rule Subtypes decided to pass on to the next rule
  + Attempting expansion of rule OptionToOption
  + Rule OptionToOption decided to pass on to the next rule
  + Attempting expansion of rule PartialOptionToNonOption
  + Rule PartialOptionToNonOption decided to pass on to the next rule
  + Attempting expansion of rule ToOption
  + Rule ToOption decided to pass on to the next rule
  + Attempting expansion of rule ValueClassToValueClass
  + Rule ValueClassToValueClass decided to pass on to the next rule
  + Attempting expansion of rule ValueClassToType
  + Rule ValueClassToType decided to pass on to the next rule
  + Attempting expansion of rule TypeToValueClass
  + Rule TypeToValueClass decided to pass on to the next rule
  + Attempting expansion of rule EitherToEither
  + Rule EitherToEither decided to pass on to the next rule
  + Attempting expansion of rule MapToMap
  + Rule MapToMap decided to pass on to the next rule
  + Attempting expansion of rule IterableToIterable
  + Rule IterableToIterable decided to pass on to the next rule
  + Attempting expansion of rule ProductToProduct
    + Resolved Bar getters: (`x`: java.lang.String (ConstructorVal), `y`: scala.Int (ConstructorVal)) and Foo constructor (`x`: java.lang.String (ConstructorParameter, default = None), `y`: scala.Int (ConstructorParameter, default = None), `z`: scala.Boolean (ConstructorParameter, default = Some(Foo.apply$default)))
    + Recursive derivation for field `x`: java.lang.String into matched `x`: java.lang.String
      + Deriving Total Transformer expression from java.lang.String to java.lang.String
        + Attempting expansion of rule Implicit
        + Rule Implicit decided to pass on to the next rule
        + Attempting expansion of rule Subtypes
        + Rule Subtypes expanded successfully: bar.x
      + Derived recursively total expression bar.x
    + Resolved `x` field value to bar.x
    + Recursive derivation for field `y`: scala.Int into matched `y`: scala.Int
    + Deriving Total Transformer expression from scala.Int to scala.Int
      + Attempting expansion of rule Implicit
      + Rule Implicit decided to pass on to the next rule
      + Attempting expansion of rule Subtypes
      + Rule Subtypes expanded successfully: bar.y
    + Derived recursively total expression bar.y
    + Resolved `y` field value to bar.y
    + Resolved `z` field value to Foo.apply$default
    + Resolved 3 arguments, 3 as total and 0 as partial Expr
  + Rule ProductToProduct expanded successfully:
    | new Foo(bar.x, bar.y, Foo.apply$default)
+ Derived final expression is:
  | new Foo(bar.x, bar.y, Foo.apply$default)
+ Derivation took 0.109828000 s

With the structured log, the user could see e.g.:

  • that no implicit was summoned during the expansion
  • how Foo constructor was called
  • that default values were used and how they were obtained
  • what is the final expression and how long it took to compute it

Warning

Structured logs from macros are still logs - their role is to help with debugging, but their format changes over time and the log for one macro could look completely different from the log from another macro. Examples from this page should not be treated as any point of reference.

Enabling logs can be done both on an individual transformation level, like above, or with a shared implicit config:

Example

//> using dep io.scalaland::chimney::0.8.1
import io.scalaland.chimney.dsl._

 // All transformations derived in this scope will see these new flags (Scala 2-only syntax, see cookbook for Scala 3)
implicit val cfg = TransformerConfiguration.default.enableMacrosLogging

The flag is also available to Patchers, this code:

Example

//> using dep io.scalaland::chimney::0.8.1
import io.scalaland.chimney.dsl._

case class Email(address: String) extends AnyVal
case class Phone(number: Long) extends AnyVal

case class User(id: Int, email: Email, phone: Phone)
case class UserUpdateForm(email: String, phone: Long)

val user = User(10, Email("abc@@domain.com"), Phone(1234567890L))
val updateForm = UserUpdateForm("xyz@@domain.com", 123123123L)

user.using(updateForm).enableMacrosLogging.patch

would generate:

Example

+ Start derivation with context: PatcherContext[A = User, Patch = UserUpdateForm](obj = user, patch = userupdateform)(PatcherConfig(
|   flags = PatcherFlags(displayMacrosLogging),
|   preventImplicitSummoningForTypes = None
| ))
+ Deriving Patcher expression for User with patch UserUpdateForm
  + Deriving Total Transformer expression from java.lang.String to Email
    + Attempting expansion of rule Implicit
    + Rule Implicit decided to pass on to the next rule
    + Attempting expansion of rule Subtypes
    + Rule Subtypes decided to pass on to the next rule
    + Attempting expansion of rule OptionToOption
    + Rule OptionToOption decided to pass on to the next rule
    + Attempting expansion of rule PartialOptionToNonOption
    + Rule PartialOptionToNonOption decided to pass on to the next rule
    + Attempting expansion of rule ToOption
    + Rule ToOption decided to pass on to the next rule
    + Attempting expansion of rule ValueClassToValueClass
    + Rule ValueClassToValueClass decided to pass on to the next rule
    + Attempting expansion of rule ValueClassToType
    + Rule ValueClassToType decided to pass on to the next rule
    + Attempting expansion of rule TypeToValueClass
      + Deriving Total Transformer expression from java.lang.String to java.lang.String
        + Attempting expansion of rule Implicit
        + Rule Implicit decided to pass on to the next rule
        + Attempting expansion of rule Subtypes
        + Rule Subtypes expanded successfully: userupdateform.email
      + Derived recursively total expression userupdateform.email
    + Rule TypeToValueClass expanded successfully: new Email(userupdateform.email)
  + Deriving Total Transformer expression from scala.Long to Phone
    + Attempting expansion of rule Implicit
    + Rule Implicit decided to pass on to the next rule
    + Attempting expansion of rule Subtypes
    + Rule Subtypes decided to pass on to the next rule
    + Attempting expansion of rule OptionToOption
    + Rule OptionToOption decided to pass on to the next rule
    + Attempting expansion of rule PartialOptionToNonOption
    + Rule PartialOptionToNonOption decided to pass on to the next rule
    + Attempting expansion of rule ToOption
    + Rule ToOption decided to pass on to the next rule
    + Attempting expansion of rule ValueClassToValueClass
    + Rule ValueClassToValueClass decided to pass on to the next rule
    + Attempting expansion of rule ValueClassToType
    + Rule ValueClassToType decided to pass on to the next rule
    + Attempting expansion of rule TypeToValueClass
      + Deriving Total Transformer expression from scala.Long to scala.Long
        + Attempting expansion of rule Implicit
        + Rule Implicit decided to pass on to the next rule
        + Attempting expansion of rule Subtypes
        + Rule Subtypes expanded successfully: userupdateform.phone
      + Derived recursively total expression userupdateform.phone
    + Rule TypeToValueClass expanded successfully: new Phone(userupdateform.phone)
+ Derived final expression is:
| {
|   val user: User = new User(user.id, new Email(userupdateform.email), new Phone(userupdateform.phone));
|   user
| }
+ Derivation took 0.064756000 s

Ideas, questions or bug reports

If you haven't found a solution to your question try asking at GitHub discussions page. If your idea looks like a feature request or a bug, open an issue at GitHub issues page.