mrdziuban
commented
1 year ago I was able to minimize this a bit:
https://scastie.scala-lang.org/mrdziuban/IHV2q8voTuSKVOSw3WP0Ug/19
expand code
```scala trait LiftAll[F[_], T] { type Out def instances: Out } object LiftAll { inline def apply[F[_], T <: Tuple]: LiftAll[F, T] { type Out = Tuple.Map[T, F] } = new LiftAll[F, T] { type Out = Tuple.Map[T, F] lazy val instances = compiletime.summonAll[Tuple.Map[T, F]] } } sealed trait Test[A, B] object Test { given inst[A, B]: Test[A, B] = new Test[A, B] {} def test = ((): Matchable) match { case _: Int => LiftAll[Test[Int, *], String *: Boolean *: EmptyTuple] case _: String => LiftAll[Test[String, *], Int *: Boolean *: EmptyTuple] case _: Boolean => LiftAll[Test[Boolean, *], Int *: String *: EmptyTuple] } } ```A few other things that may be helpful for debugging:
- The error doesn't happen when using
compiletime.summonAlldirectly instead ofLiftAll - The error doesn't happen when
trait Testonly has one type parameter and theLiftAllcalls look likeLiftAll[Test, String *: Boolean *: EmptyTuple] - The error doesn't happen if I add a type annotation of
: Anytodef test
WojciechMazur
Compiler version
3.3.0-RC3, also happens on 3.2.2
Minimized code
Please excuse the fact that this code is fairly meaningless, it's a very trimmed down version of a pattern (that actually has meaning) in my codebase.
https://scastie.scala-lang.org/mrdziuban/IHV2q8voTuSKVOSw3WP0Ug/1
Output
I've tried increasing the stack size as suggested, all the way up to
-Xss50M.When I compile with
-Yno-decode-stacktracesI get a stack trace. GitHub tells me the full stack trace is too long to include in the description here but you can view it here: https://gist.github.com/mrdziuban/9d60341f5c2173bd768996da4589af12. Here's a condensed version:Expectation
The code would compile successfully as it does in scala 2 -- https://scastie.scala-lang.org/mrdziuban/WlK9X64oSJOFNyzrSP5FBA/3
Strangely, if I comment out just one of the
case objects inobject Reland the correspondingcasestatement indef testthen everything compiles just fine. Not sure if this is actually relevant to the underlying issue but wanted to mention it in case it's helpful for debugging.