Make refined type printing more source compatible

[jan-pieter]

Fixes #12939
Supersedes #13698

[SethTisue]

on the previous PR, @tgodzik wondered if the formatting could be improved. I think it's out of scope for this PR. Maybe Dale's recent pretty-printing PR (not merged yet) will improve the situation

[SethTisue]

we only made two further small tweaks:

• added a test that checks that the type printing also works as expected in the REPL when the refined type is written directly, rather than originating from an anonymous subclass
• add a catchall case that prints something if we see something unexpected in the refinedInfo — hopefully this case will never be hit. as I read the Scala 2 spec, the only things that are syntactically legal in a refined type are val, def, and type; but I'm a little less certain that we've covered all the cases of how those might be represented as trees

if this passes review, we can squash the commits before merge

[SethTisue]

marking as draft for now since there are some test failures. probably just need to update some expected outputs to match the new format?

[jan-pieter]

probably just need to update some expected outputs to match the new format?

Most of the failures were indeed due to some unexpected output. I've updated the expected output for those.

The catchall case was actually being hit by the tests for capture checking. I lack knowledge about the implementation details to explain that. For now I've updated this case to return an empty string which makes the tests pass. Some example output with the previous implementation:

-- [E007] Type Mismatch Error: tests/neg-custom-args/captures/i15772.scala:20:49 ---------------------------------------
20 |    val boxed1 : (({*} C) => Unit) -> Unit = box1(c)  // error
   |                                             ^^^^^^^
   |Found:    {c} 
   |  ({*} 
   |    ({c} 
   |      C{
   |        AnnotatedType(TypeRef(ThisType(TypeRef(NoPrefix,module class <empty>)),class C),CaptureAnnotation(BiMapped3802(Var3741{TermRef(ThisType(TypeRef(ThisType(TypeRef(NoPrefix,module class scala)),module class caps$)),val *)}, elems = {TermRef(ThisType(TypeRef(ThisType(TypeRef(NoPrefix,module class scala)),module class caps$)),val *)}),false)) 
   |          arg
   |        : {*} C
   |      }
   |    )
   |   -> Unit)
   | -> Unit
   |Required: (({*} C) => Unit) -> Unit
   |
   | longer explanation available when compiling with `-explain`

[bishabosha]

I suggest this case to be made below TypeBounds and turn it to TypeProxy, so it can work with e.g. val x: List[Int], which is AppliedType, type proxy should cover most cases - such as annotated types Foo { val x: String @unchecked }.

[bishabosha]

there could also be PolyType, e.g. def bar: Foo { def foo[T <: Any](t: T): String } = ???

[bishabosha]

for this case, Foo must be a class that already defines foo as a polymorphic method

[jan-pieter]

Thanks for your comments and suggestions @bishabosha. I've updated the code & tests accordingly. This also solves the issue with AnnotatedTypes in the capture checking tests.

[jan-pieter]

@SethTisue Shall I rebase on the main branch, squash the commits and force push this branch?

[dwijnand]

Is it just me that's bothered that C1{val x: Int} isn't spaced out as C1 { val x: Int }?

[jan-pieter]

No, I prefer the latter one as well. However we previously said it to be out of scope for this PR: #16303 (comment) . This PR only adds keywords. Perhaps #16277 did bring some improvements already? Otherwise I propose to improve that in a separate PR.

[SethTisue]

Shall I rebase on the main branch, squash the commits and force push this branch?

Yes please!

[jan-pieter]

@SethTisue It took a few tries, but now it's ready to be merged I think 😉

[SethTisue]

Yay!