Illegal MatchType Error reaches the backend

[odersky]

Compiler version

3.4.0

Minimized example

object Test:

  object Named:
    opaque type Named[name <: String & Singleton, A] >: A = A

  type DropNames[T <: Tuple] = T match
    case Named.Named[_, x] *: xs => x *: DropNames[xs]
    case _ => T

  def f[T <: Tuple]: DropNames[T] = ???

Output

an unexpected type representation reached the compiler backend while compiling named-tuples-strawman.scala: dotty.tools.dotc.core.Types$PreviousErrorType@2d9dff65. If possible, please file a bug on https://github.com/lampepfl/dotty/issues
scala.MatchError: dotty.tools.dotc.core.Types$PreviousErrorType@2d9dff65 (of class dotty.tools.dotc.core.Types$PreviousErrorType)
	at dotty.tools.backend.jvm.BCodeHelpers.dotty$tools$backend$jvm$BCodeHelpers$$typeToTypeKind(BCodeHelpers.scala:764)
	at dotty.tools.backend.jvm.BCodeHelpers$BCInnerClassGen.toTypeKind(BCodeHelpers.scala:202)
	at dotty.tools.backend.jvm.BCodeHelpers$BCInnerClassGen.toTypeKind$(BCodeHelpers.scala:130)
	at dotty.tools.backend.jvm.BCodeSkelBuilder$PlainSkelBuilder.toTypeKind(BCodeSkelBuilder.scala:134)
	at dotty.tools.backend.jvm.BCodeHelpers$BCInnerClassGen.asmMethodType(BCodeHelpers.scala:188)
	at dotty.tools.backend.jvm.BCodeHelpers$BCInnerClassGen.asmMethodType$(BCodeHelpers.scala:130)
	at dotty.tools.backend.jvm.BCodeSkelBuilder$PlainSkelBuilder.asmMethodType(BCodeSkelBuilder.scala:134)
	at dotty.tools.backend.jvm.BCodeSkelBuilder$PlainSkelBuilder.genDefDef(BCodeSkelBuilder.scala:815)
	at dotty.tools.backend.jvm.BCodeSkelBuilder$PlainSkelBuilder.gen(BCodeSkelBuilder.scala:693)
	at dotty.tools.backend.jvm.BCodeSkelBuilder$PlainSkelBuilder.gen$$anonfun$1(BCodeSkelBuilder.scala:699)
	at scala.runtime.function.JProcedure1.apply(JProcedure1.java:15)
	at scala.runtime.function.JProcedure1.apply(JProcedure1.java:10)
	at scala.collection.immutable.List.foreach(List.scala:333)
	at dotty.tools.backend.jvm.BCodeSkelBuilder$PlainSkelBuilder.gen(BCodeSkelBuilder.scala:699)
	at dotty.tools.backend.jvm.BCodeSkelBuilder$PlainSkelBuilder.genPlainClass(BCodeSkelBuilder.scala:293)
	at dotty.tools.backend.jvm.CodeGen.genClass(CodeGen.scala:154)
	at dotty.tools.backend.jvm.CodeGen.genClassDef$1(CodeGen.scala:62)
	at dotty.tools.backend.jvm.CodeGen.genClassDefs$1(CodeGen.scala:117)
	at dotty.tools.backend.jvm.CodeGen.genClassDefs$1$$anonfun$1(CodeGen.scala:115)
	at scala.runtime.function.JProcedure1.apply(JProcedure1.java:15)
	at scala.runtime.function.JProcedure1.apply(JProcedure1.java:10)
	at scala.collection.immutable.List.foreach(List.scala:333)
	at dotty.tools.backend.jvm.CodeGen.genClassDefs$1(CodeGen.scala:115)
	at dotty.tools.backend.jvm.CodeGen.genUnit(CodeGen.scala:120)
	at dotty.tools.backend.jvm.GenBCode.run(GenBCode.scala:83)
	at dotty.tools.dotc.core.Phases$Phase.runOn$$anonfun$1(Phases.scala:354)
	at scala.runtime.function.JProcedure1.apply(JProcedure1.java:15)
	at scala.runtime.function.JProcedure1.apply(JProcedure1.java:10)
	at scala.collection.immutable.List.foreach(List.scala:333)
	at dotty.tools.dotc.core.Phases$Phase.runOn(Phases.scala:360)
	at dotty.tools.backend.jvm.GenBCode.runOn(GenBCode.scala:91)
	at dotty.tools.dotc.Run.runPhases$1$$anonfun$1(Run.scala:315)
	at scala.runtime.function.JProcedure1.apply(JProcedure1.java:15)
	at scala.runtime.function.JProcedure1.apply(JProcedure1.java:10)
	at scala.collection.ArrayOps$.foreach$extension(ArrayOps.scala:1323)
	at dotty.tools.dotc.Run.runPhases$1(Run.scala:337)
	at dotty.tools.dotc.Run.compileUnits$$anonfun$1(Run.scala:348)
	at dotty.tools.dotc.Run.compileUnits$$anonfun$adapted$1(Run.scala:357)
	at dotty.tools.dotc.util.Stats$.maybeMonitored(Stats.scala:69)
	at dotty.tools.dotc.Run.compileUnits(Run.scala:357)
	at dotty.tools.dotc.Run.compileSources(Run.scala:261)
	at dotty.tools.dotc.Run.compile(Run.scala:246)
	at dotty.tools.dotc.Driver.doCompile(Driver.scala:37)
	at dotty.tools.dotc.Driver.process(Driver.scala:197)
	at dotty.tools.dotc.Driver.process(Driver.scala:165)
	at dotty.tools.dotc.Driver.process(Driver.scala:177)
	at dotty.tools.dotc.Driver.main(Driver.scala:207)
	at dotty.tools.dotc.Main.main(Main.scala)
Error while emitting named-tuples-strawman.scala
dotty.tools.dotc.core.Types$PreviousErrorType@2d9dff65 (of class dotty.tools.dotc.core.Types$PreviousErrorType)
1 warning found
1 error found

Expectation

A regular error, like this one:

1 |object Test:
  |       ^
  |       The match type contains an illegal case:
  |           case Test.Named.Named[_, x] *: xs => x *: Test.DropNames[xs]
  |       (this error can be ignored for now with `-source:3.3`)

I verified that that was the error that generated the illegal error type.

[odersky]

Also, I am not sure why this program would be illegal. I guess it should not be.

The problem seems to be that the match type is reduced only at erasure. At that time, opaque types are already eliminated, so Named looks like a normal type alias. So the type comparer complains.

[odersky]

Here's the stacktrace I see when I add a new Error().printStackTrace() to line 3573 of TypeComparer where the ErrorType gets generated:

	at dotty.tools.dotc.core.TrackingTypeComparer.recur$2(TypeComparer.scala:3574)
	at dotty.tools.dotc.core.TrackingTypeComparer.op$proxy82$1(TypeComparer.scala:3614)
	at dotty.tools.dotc.core.TrackingTypeComparer.matchCases(TypeComparer.scala:3615)
	at dotty.tools.dotc.core.Types$MatchType.matchCases$1(Types.scala:5041)
	at dotty.tools.dotc.core.Types$MatchType.reduced$$anonfun$1(Types.scala:5050)
	at dotty.tools.dotc.core.TypeComparer.inSubComparer(TypeComparer.scala:3054)
	at dotty.tools.dotc.core.TypeComparer.tracked(TypeComparer.scala:3064)
	at dotty.tools.dotc.core.TypeComparer$.tracked(TypeComparer.scala:3232)
	at dotty.tools.dotc.core.Types$MatchType.reduced(Types.scala:5050)
	at dotty.tools.dotc.core.Types$MatchType.tryNormalize(Types.scala:4999)
	at dotty.tools.dotc.core.Types$AppliedType.tryMatchAlias$1$$anonfun$1(Types.scala:4562)
	at dotty.tools.dotc.core.MatchTypeTrace$.recurseWith(MatchTypeTrace.scala:81)
	at dotty.tools.dotc.core.Types$AppliedType.tryMatchAlias$1(Types.scala:4563)
	at dotty.tools.dotc.core.Types$AppliedType.tryNormalize(Types.scala:4568)
	at dotty.tools.dotc.core.Types$AppliedType.translucentSuperType(Types.scala:4541)
	at dotty.tools.dotc.core.TypeErasure.checkedSuperType(TypeErasure.scala:752)
	at dotty.tools.dotc.core.TypeErasure.dotty$tools$dotc$core$TypeErasure$$apply(TypeErasure.scala:642)
	at dotty.tools.dotc.core.TypeErasure.eraseResult(TypeErasure.scala:879)
	at dotty.tools.dotc.core.TypeErasure.eraseInfo(TypeErasure.scala:815)
	at dotty.tools.dotc.core.TypeErasure$.transformInfo(TypeErasure.scala:273)
	at dotty.tools.dotc.transform.Erasure.transform(Erasure.scala:96)
	at dotty.tools.dotc.core.Denotations$SingleDenotation.goForward$1(Denotations.scala:831)
	at dotty.tools.dotc.core.Denotations$SingleDenotation.current(Denotations.scala:877)
	at dotty.tools.dotc.core.Symbols$Symbol.recomputeDenot(Symbols.scala:124)
	at dotty.tools.dotc.core.Symbols$Symbol.computeDenot(Symbols.scala:118)
	at dotty.tools.dotc.core.Symbols$Symbol.denot(Symbols.scala:111)
	at dotty.tools.dotc.core.Symbols$.toDenot(Symbols.scala:520)
	at dotty.tools.dotc.core.SymDenotations$ClassDenotation.paramAccessors$$anonfun$1(SymDenotations.scala:2386)
	at dotty.tools.dotc.core.Scopes$Scope.filter(Scopes.scala:103)
	at dotty.tools.dotc.core.SymDenotations$ClassDenotation.paramAccessors(SymDenotations.scala:2386)
	at dotty.tools.dotc.core.Contexts$Context.superCallContext(Contexts.scala:417)
	at dotty.tools.dotc.typer.Typer.typedClassDef(Typer.scala:2720)
	at dotty.tools.dotc.transform.Erasure$Typer.typedClassDef(Erasure.scala:1047)

[sjrd]

Definitely a problem in erasure. During erasure, when trying to reduce the match type before erasing its bound, I observe a weird behavior. It sees as tycon of the match case pattern Test.Named.Named -- so far so good. But tycon.info gives [name <: String & Singleton,A] = A. That does not seem correct to me. At erasure, the info should still be the public abstract bounds >: [name] =>> Tupe <: [name] =>> Any. Shouldn't it? ctx.phase is erasure when it computes that.

[odersky]

No, opaque type aliases are eliminated before erasure, at phase ElimOpaque.

[sjrd]

Oh. Well at least that explains it. It will make this tricky to fix, though. I'll have to think about it more.

[odersky]

I just hit this in the wild with some test code using NamedTuples.

[dwijnand]

I'm merging ElimOpaque into Erasure