Scala 3.7 given priority change causing ambiguous givens

[yoohaemin]

Not an issue right now, but probably going to be in the near future...

https://www.scala-lang.org/2024/08/19/given-priority-change-3.7.html

I've turned on -source 3.6 and now I'm seeing these warnings:

-- Warning: /path/to/source.scala:39:33
39 |        _.transformInto[Bar]
   |                       ^
   |Given search preference for io.scalaland.chimney.Transformer.AutoDerived[Foo, Bar] between alternatives
   |  (Foo.given_Transformer_Foo_Bar : io.scalaland.chimney.Transformer[Foo, Bar])
   |and
   |  (io.scalaland.chimney.Transformer.AutoDerived.deriveAutomatic : [From, To]: io.scalaland.chimney.Transformer.AutoDerived[From, To])
   |will change.
   |Current choice           : the first alternative
   |New choice from Scala 3.7: none - it's ambiguous

Right now I have a transformer instance defined (unnamed given) in the companion object of Foo (different source) and I have dsl.* imported at the callsite.

[MateuszKubuszok]

I'll contact the compiler devs about it. For now I am not touching this - it could be solved with e.g. replacing each

final def transformInto[To](implicit transformer: Transformer.AutoDerived[From, To]): To =
  transformer.transform(source)

with

inline def transformInto[To]: To = summonFrom {
  case transformer: Transformer[From, To] => transformer.transform(source)
  case _                                  => Transformer.AutoDerived.derive
}

or similar, but that would break binary compatibility, bumping version to 2.0.0... I'm going to delay that until I get more info from the compiler team.

In the meantime - these warnings and priorities should only be for givens, not for implicits - if you rename your anonymous given to a named implicit it should disappear.

[yoohaemin]

Thanks for the reply. I am safe until 3.7 is released, unless I decide to use the new priorities before 3.7 release (unlikely). This is more of a heads-up 😄

There are plenty of workarounds, so no worries.

[MateuszKubuszok]

Yeah, there are, although I'd prefer to:

• focus on some remaining long pending issues in the next following months
• make sure that all "planned" features are there
• before I'll even consider some cleanup that would justify breaking bincompat and releasing 2.0.0
• so that people on 1.x would get as much as possible, hassle of moving to 2.x should be justified.

Ideally, I'd want out users to get the same source compatibility even after the major version bump, especially since we are trying to keep the code compiling in 2.12/2.13/3.

This change can make it tricky sometimes, e.g. when someone would use:

implicit def keyConverter[KeyA, KeyB](
  // allows using user provider Transformer or fallback on generated Transformer.AutoDerived
  implicit transformer: Transformer.AutoDerived[From, To]
): KeyConverter[KeyA, KeyB] = ...

then whatever we would use in core library would not automatically propagate to our users and that could badly reflect on us :/.

[MateuszKubuszok]

I tried to reproduce the issue and fix it. From what I see you have it when you store the givens in your companion objects. Unfortunately, the implicit priorities do not work. I pinned down the current behavior so that future Community Build runs would show the compiler team that the issue exists (IMHO the fact that so little projects had issues is a testament of not testing things that should obviously work rather than negligible impact of that change).

[MateuszKubuszok]

I found 1 way of working around the issue - https://scastie.scala-lang.org/eKrwKqN8S2uHp5pjr61uiw or https://scastie.scala-lang.org/KXeEJJxHTWSpZIM640FM6A - but I am not sure whether it going to be source-compatible change, while it very likely to be a binary incompatible change - something that would require releasing Chimney 2.0.0 much faster than I'd wanted to.

[MateuszKubuszok]

How we'll solve it depends on scala/scala3#21909 - if it's merged, it will let us simplify Chimney 2.0.0 for next S3 LTS. If not, it will be a PITA.

Currently, I am certain that Chimney 1.+ will never support semiautomatic derivation with new givens.