Erase constant value type

[som-snytt]

Mixin was confused when String("hello, world") overrides String.

Fixes scala/bug#12301

[som-snytt]

Quick lunchtime fix. This probably breaks overriding a method of literal type. (Edit: that works after all. I was wondering about the asymmetry of the fix.)

[lrytz]

I'm not convinced this is the right fix. IIUC, matches is used for override checking.

---

I really don't understand why alwaysMatchSimple is !phase.erasedTypes, an explanation of this would be very welcome. The result is (without your patch):

scala> typeOf[Int] matches typeOf[String]
val res3: Boolean = true

scala> exitingErasure(typeOf[Int] matches typeOf[String])
val res4: Boolean = false

---

The bug is that Mixin creates a mixin forwarder in X for P.p, it shouldn't do that. Your patch fixes that, so it does have an effect in Mixin, but I fear that it has more (potentially undesired) effects. With your patch:

scala> exitingErasure(typeOf[""] matches typeOf[String])
val res2: Boolean = false

scala> typeOf[Int] matches typeOf[String]
val res3: Boolean = true

---

In Mixin, there is a call to matchesType which explicitly sets alwaysMatchSimple = true. So maybe the bug is that through some different patch we get from Mixin to matchesType with alwaysMatchSimple = false?

Or maybe the bug is that the ConstantType is not erased?

[lrytz]

I really don't understand why alwaysMatchSimple is !phase.erasedTypes, an explanation of this would be very welcome

Maybe the idea is "let's assume things override" before erasure. RefChecks will issue errors when overridding is not correct.

Or maybe the bug is that the ConstantType is not erased?

scala> trait P { def p: String = "p" }
     | class X extends P { override final val p      = "x" }
     | class Y extends P { override final val p: "x" = "x" }

scala> exitingErasure(typeOf[X].member(TermName("p")).tpe)
val res1: $r.intp.global.Type = String("x") <and> String

scala> exitingErasure(typeOf[Y].member(TermName("p")).tpe)
val res2: $r.intp.global.Type = (): String

scala> (new X).p
java.lang.ClassFormatError: Duplicate method name "p" with signature "()Ljava.lang.String;" in class file X
  at java.lang.ClassLoader.defineClass1(Native Method)
  at java.lang.ClassLoader.defineClass(ClassLoader.java:756)
  at scala.reflect.internal.util.AbstractFileClassLoader.findClass(AbstractFileClassLoader.scala:77)
  at java.lang.ClassLoader.loadClass(ClassLoader.java:418)
  at java.lang.ClassLoader.loadClass(ClassLoader.java:351)
  ... 36 elided

scala> (new Y).p
val res4: String = x

[som-snytt]

I'll convert to draft and give it some thought.

I moved the scouting commits to a separate PR.

[som-snytt]

@lrytz I'll rework this. Also Jean Rondeau will be in Zürich per the website

[SethTisue]

@som-snytt is this one you're thinking of reviving...?

[som-snytt]

@SethTisue yes, although like all old relationships, I've forgotten what was so attractive about it.

[som-snytt]

For my stage name, I may take "Johnny Rondo". TIL "Ryan Starr" is just a stage name.

This is the "quick lunchtime fix" I must have intended two years ago. It is also the last of Lukas's semi-rhetorical questions.

In the test, explicitly annotated override def x: "x" = "x" was erased correctly. Unannotated x(): String("x") was untouched.

I did not answer Lukas's other question about matches during this visit; I left that red herring for another meal.

[som-snytt]

The answer is

[error]  * method BLACK()java.lang.String in interface scala.io.AnsiColor does not have a correspondent in current version
[error]    filter with: ProblemFilters.exclude[DirectMissingMethodProblem]("scala.io.AnsiColor.BLACK")

and various osgi

[error] Test tools.test.osgi.reflection.toolbox.ReflectionToolBoxTest.basicMirrorThroughOsgi failed: java.lang.ClassFormatError: Method LogPendingSubTypesThreshold in class scala/reflect/internal/tpe/TypeComparers has illegal modifiers: 0x402, took 0.212 sec