Regression 3.4.0+: scenario where Tuple.InverseMap does not compute the expected type

[kavedaa]

Compiler version

3.4.0 and higher

Minimized code

//> using scala 3.4.0

trait Column:
  type T
  type F[X]
  type Q = F[T]

class varchar extends Column:
  type T = String

trait notnull extends Column:
  type F[X] = X

object Error:

  val firstName = new varchar with notnull
  val lastName = new varchar with notnull

  val relation = (firstName, lastName)

  type RelationTypes = Tuple.InverseMap[relation.type, [X] =>> Column { type Q = X }]

  summon[RelationTypes =:= (String, String)]

object Works:

  object firstName extends varchar with notnull
  object lastName extends varchar with notnull

  val relation = (firstName, lastName)

  type RelationTypes = Tuple.InverseMap[relation.type, [X] =>> Column { type Q = X }]

  summon[RelationTypes =:= (String, String)]

Output

Compiling project (Scala 3.4.0, JVM (8))
[error] .\test.scala:23:45
[error] Cannot prove that Error.RelationTypes =:= (String, String).
[error]   summon[RelationTypes =:= (String, String)]
[error]                                             ^
Error compiling project (Scala 3.4.0, JVM (8))
Compilation failed

Expectation

Successful compilation.

[kavedaa]

Compiles in 3.3.3.

Could perhaps be related to #17222?

[kavedaa]

Regression happens in 3.4.0-RC1-bin-20231218-ec2b8bc-NIGHTLY:

Compiling project (Scala 3.4.0-RC1-bin-20231218-ec2b8bc-NIGHTLY, JVM (8))
[error] .\test.scala:23:45
[error] Cannot prove that Error.RelationTypes =:= (String, String).
[error]   summon[RelationTypes =:= (String, String)]
[error]                                             ^
Error compiling project (Scala 3.4.0-RC1-bin-20231218-ec2b8bc-NIGHTLY, JVM (8))
Compiling project (Scala 3.4.0-RC1-bin-20231214-423cd6c-NIGHTLY, JVM (8))
Compiled project (Scala 3.4.0-RC1-bin-20231214-423cd6c-NIGHTLY, JVM (8))

[EugeneFlesselle]

Minimized to:

trait A:
  type T
  type U = T

trait B extends A:
  type T = String

object C extends B


type F[X] = A { type U = X } // works when `U` is replaced with `T`

type InvF[Y] = Y match
  case F[x] => x


object Test:
  summon[InvF[C.type] =:= String] // ok
  summon[InvF[B] =:= String] // error: selector B does not uniquely determine parameter x

although I'm not 100% sure this an incorrect error.

[jchyb]

After investigating #17222, these two issues are probably unrelated

[mbovel]

This issue is fixed by #20161: adding import language.experimental.betterMatchTypeExtractors enabled to compile both the @EugeneFlesselle's minimization @kavedaa's original snippet.

Adding the import does not fix #17222.

@smarter would you have any insights on when the amendment will be stabilized?

Should I just add regression tests and mark this issue as solved?

[smarter]

It needs a minor release so it should be in 3.6.0

[mbovel]

Wonderful, so this means these examples will compile just fine in 3.6!

I suggest we add regressions tests with import language.experimental.betterMatchTypeExtractors and close this issue already. When 3.6 is released, we can remove the imports in the tests. Sound okay?

[EugeneFlesselle]

I suggest we add regressions tests with import language.experimental.betterMatchTypeExtractors

IIRC tests usually have scala-cli directives for experimental options but sounds fine to me either way