extending java.lang.Enum should be desugared before TASTy

[bishabosha]

minimized code

Compile with 0.27.0-RC1

enum A extends java.lang.Enum[A]
  case A1

In TASTy, the first parent of A is

TypeApply(Select(New(java.lang.Enum[A]),<init>[Signed (1)java.lang.Enum]), List(A))

with no value arguments passed.

expectation

The parent in TASTy should be

Apply(TypeApply(Select(New(java.lang.Enum[A]),<init>[Signed (1,java.lang.String,scala.Int)java.lang.Enum]), List(A)), List($name, _$ordinal))

with any other relevant desugarings applied, such as adding $name and _$ordinal to the constructor of A

Alternative

special-casing the synthetic no-arg java.lang.Enum constructor becomes part of the specification for TASTy

[bishabosha]

This one seems more tricky, if we move the phase CompleteJavaEnums before pickling, then the check fails for the @static annotated forwarders to enum values after moving them to the enum class, we can either keep the forwarder moving in a later phase, or special case the error for @static, or special case pickling for this signature, or do nothing.

@smarter @sjrd

[smarter]

if we move the phase CompleteJavaEnums before pickling

We can't have a phase that changes signature before pickling anyway: https://github.com/lampepfl/dotty/commit/1818b99f7cec44e70ea75eb41e63e4fd0b7d7647

[bishabosha]

we could chalk it up to another special case primitive of scala implementations? ala Nothing, &, |, AnyKind etc,

[bishabosha]

~I imagine this leaks into tasty reflect also so that could be a lot of special casing~ Edit: seems reflection doesn't reify apply nodes for parents

[bishabosha]

I am going to special case signatures in pickling and then benchmark it

[bishabosha]

On second thought, it just makes no sense to patch extends java.lang.Enum[MyEnum] with the correct signature without also expanding the tree to extends java.lang.Enum[MyEnum]($name, $ordinal), where $name and $ordinal do not exist yet. So again making this change seems pretty dubious without the complete desugaring of constructors appearing before tasty, which is more tricky because there is no precedent for patching a specific parent in desugaring