Nested inline calls - unrelated code affects compilation

[ncreep]

Compiler version

3.3.0

Minimized code

While trying to work around #8739 (nested transparent calls) I stumbled upon some spooky compilation side-effects:

//> using scala "3.3.0"
import scala.compiletime.*

class Typed[A]

transparent inline def computeType[T <: Tuple]: Typed[_ <: Tuple] =
  inline erasedValue[T] match
    case _: EmptyTuple => Typed[EmptyTuple]
    case _: (h *: t) =>
      inline computeType[t] match
        case _: Typed[ts] =>
          Typed[h *: ts]

val x: Typed[("a", "b", "c")] =
  computeType[("a", "b", "c")]

This fails to compile with the following message:

Found:    Typed[?1.CAP]
Required: Typed[(("a" : String), ("b" : String), ("c" : String))]

where:    ?1 is an unknown value of type scala.runtime.TypeBox[("a" : String) *: Nothing, ("a" : String) *: Tuple]

But if I add any other code in the second branch everything compiles just fine.
So this compiles:

case _: Typed[ts] =>
  println("")
  Typed[h *: ts]

And this:

case _: Typed[ts] =>
  123
  Typed[h *: ts]

And even this:

case _: Typed[ts] =>
  val x = Typed[h *: ts]
  x

Expectation

The code should have a consistent type irrespective of code that is unrelated to the final result type.

Thanks

[odersky]

That's indeed a mystery. Maybe for experienced people in the Spree to find out?

[SethTisue]

minimized somewhat:

//> using scala 3.3.1-RC3

import scala.compiletime.erasedValue

class Typed[A]

transparent inline def computeType: Any =
  inline erasedValue[Tuple1["a"]] match
    case _ =>
      inline (Typed[EmptyTuple]: Typed[_ <: Tuple]) match
        case _: Typed[ts] =>
//          println("")   // compiles if uncommented
          Typed["a" *: ts]

val x: Typed[Tuple1["a"]] = computeType

not sure if actually-helpful minimization, or just minimization golf

also not positive it can't be reduced further

[ncreep]

@SethTisue, thanks for the minimization. It does seem actually helpful as it shows that the nested transparent call is not part of the issue (as my title implied).

[scala-center-bot]

This issue was picked for the Issue Spree No. 40 of December 5th, 2023 which takes place in 6 days. @mbovel, @EugeneFlesselle and @entangled90 will be working on it. If you have any insight into the issue or guidance on how to fix it, please leave it here.

[entangled90]

I tested both snippets on last commit on master (7a56da7) and both snippets provided compile with or without a println statement.

Version 3.3.1 manifest the same issue as 3.3.0.

[som-snytt]

I did not do a bisect, but a quick probe says it progressed in August, which makes sense. Either someone had a restful Aug 1 holiday, then came back and saw the problem immediately, or someone was restless during their extended summer break, uninterested in further conversations with family, and decided to tackle some bugs in their spare time.

[entangled90]

Run a bisect: commit that fixed the issue was 2b9a62b

[mbovel]

Excellent, thanks @entangled90! Let's close this issue then 🥳

[ncreep]

Thanks for resolving this.

Can anyone please briefly explain what was the root cause of the issue?
Seeing how strange the failure was, I'm very curious to know what triggered it.