Instantiate Kotlin data classes using the _primary_ constructor

[jawo-a]

A lot of Kotlin projects define DTOs using data classes with optional parameters like

data class Person(
    val name: String? = null,
    val email: String? = null
)

Unfortunately these data classes do not work with MapStruct because the Kotlin compiler generates multiple constructors under the hood.
Issues #2281, #2378 describe this problem.

This PR is an attempt to make MapStruct use the primary constructor of the data class to instantiate it, and ignore all other generated constructors.

How it works?

Recognizes data classes and the primary constructor by

• the annotation kotlin.Metadata, that is present on any class file produced by the Kotlin compiler
• and by the presence of componentN() methods, as described in the documentation about data classes.

Note: There might be false positives.

How to review?

• Start with the changes to BeanMappingMethod.
• Then take a look at KotlinType.
• The rest are tests.

TODO

• Update the reference documentation

Improvement ideas

• Instead of using the number of componentN() methods, use their exact return types to more accurately determine the primary constructor.

[sskrla]

You could also maybe use the generated copy method as a form of detection?

[jawo-a]

@sskrla Yes, I've taken a look at decompiled data classes and the generated copy() method has exactly the same parameters as the primary constructor, although I'm not sure if that is actually guaranteed by Kotlin.

[filiphr]

This is an interesting approach @jawo-a. However, I am not sure that I like the taken approach with using the component method count. I would prefer in using the kotlinx-metadata-jvm library as an optional dependency instead. Perhaps with a mapstruct-kotlin-extensions or some other extension that will pull the needed Kotlin dependencies and that will have the needed code to make Kotlin things work properly.

[DavidRomao]

Hi @filiphr , how would you suggest that mapstruct-kotlin-extensions extension be done ? Would it be okay through the creation of a Constructor Provider SPI, where that extenstion package would then implement ? Otherwise mapstruct would load the current implementation

[jawo-a]

@DavidRomao I've already prepared a change with a Constructor Provider SPI. Going to open a PR in the next few weeks.

[MiguelAngelLV]

Any change with this?