Presumed mutation occurring in Symbol processing (paramSymss function) causes crashes

[gzoller]

Minimized code

Please see minimal (2 file) project here: https://github.com/gzoller/dottybug1/tree/master/src

I am using Reflection to dive into a class and extract the type parameters of a class (AppliedType in my case). I get the primary constructor then the paramSymss like this:

symbol.primaryConstructor.paramSymss

You can see from my trivial project I can get quality output from this, as expected, on the first pass.
But...
If I turn around immediately and re-run the same function I get the error output below:

Output

=== Here in class HashSet
    Primary: method <init>
    Const: List(List(type A), List(val rootNode))
    Size: -1
    Nonempty? true
    Head: List(type A)
P: List(List(type A), List(val rootNode))
Syms : List(A)
=== Here in class HashSet
    Primary: method <init>
[error] -- Error: /Users/wmy965/git/dottybug/dottybug/src/main/scala/Main.scala:8:28 ---
[error] 8 |    println(LookSee.diveInto[scala.collection.immutable.HashSet[String]])
[error]   |            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
[error]   |Exception occurred while executing macro expansion.
[error]   |java.util.NoSuchElementException: head of empty list
[error]   |	at scala.collection.immutable.Nil$.head(List.scala:629)
[error]   |	at scala.collection.immutable.Nil$.head(List.scala:628)
[error]   |	at dotty.tools.dotc.core.SymDenotations$SymDenotation.recurWithParamss$1(SymDenotations.scala:403)
[error]   |	at dotty.tools.dotc.core.SymDenotations$SymDenotation.recurWithParamss$1(SymDenotations.scala:403)
[error]   |	at dotty.tools.dotc.core.SymDenotations$SymDenotation.paramSymss(SymDenotations.scala:419)
[error]   |	at dotty.tools.dotc.tastyreflect.ReflectionCompilerInterface.Symbol_paramSymss(ReflectionCompilerInterface.scala:1771)
[error]   |	at dotty.tools.dotc.tastyreflect.ReflectionCompilerInterface.Symbol_paramSymss(ReflectionCompilerInterface.scala:1770)
[error]   |	at scala.tasty.Reflection$SymbolOps$.paramSymss(Reflection.scala:2275)
[error]   |	at com.boom.LookSee$.getTypeParameters(LookSee.scala:33)
[error]   |	at com.boom.LookSee$.unwindType(LookSee.scala:24)
[error]   |	at com.boom.LookSee$.diveIntoImpl(LookSee.scala:15)
[error]   |
[error]   | This location contains code that was inlined from Main.scala:8

Expectation

The way paramSymss is presented, I would expect it is a "getter" and should be re-callable as many times as you wish on the same symbol and get the same results. It appears that this function, or its tributaries, is mutating something such that re-calling the function causes the repeatable crash above.

I'm running 0.25.0-RC2 for this example.

[gzoller]

I've further isolated/reduced the example further:

  inline def diveInto[T]: String = ${ diveIntoImpl[T]() }

  def diveIntoImpl[T]()(implicit qctx: QuoteContext, ttype: scala.quoted.Type[T]): Expr[String] = 
    import qctx.tasty.{_, given _}
    Expr( unwindType(qctx.tasty)(typeOf[T]) )

  def unwindType(reflect: Reflection)(aType: reflect.Type): String =
    import reflect.{_, given _}

    aType match {
      case AppliedType(t,tob) =>
        val cs = t.classSymbol.get.primaryConstructor  // this is shared
        val a = cs.paramSymss  // this call succeeds
        println("a: "+a)
        val b = cs.paramSymss // this call explodes
        println("b: "+b)
        
      case _ =>
    }
    "OK!"

You can see from this code I'm sharing a common class symbol and primary constructor symbol, so there's no monkey business going on there. Only the paramSymss call is duplicated. This code (for me anyway) produces a successful output for 'a' but the head of empty list error as above for 'b'.

The real mystery is what is changing? Looking at the code in SymDenotations, it is so far unclear what could be mutating anything. It seems kosher. Will keep looking...

[nicolasstucki]

This looks like what I saw when fixing #9206. There we had a bug when unpickling some symbols which lead to an inconsistent state.

[nicolasstucki]

@gzoller, now you can (and should) just import qctx.tasty._

[nicolasstucki]

This was fixed in #9221