Fix incorrect wrong staging level error for Singleton type

[jchyb]

FIxes #15709.
Initially, when I started working on this I thought that it was interesting that 's was transformed into Apply(TypeApply(Ident(quote),List(TypeTree[TypeVar(TypeParamRef(T) -> TermRef(NoPrefix,val s))])),List(Ident(s)))during the Typer phase, with the interesting part being the typetree TypeTree[TypeVar(TypeParamRef(T) -> TermRef(NoPrefix,val s))]. Directly inside a TypeVar a TermRef can be found, which was the direct cause of errors thrown via PCPCheckAndHeal. Initially I wanted to see if it was possible to somehow painlessly wrap the TermRef into a TypeRef, however after spending too much time fruitlessly exploring Typer I decided to adjust checks in PCPCheckAndHeal instead.

There I changed it so that type arguments were healed via healType instead of healTypeOfTerm, as that seems more applicable there just by name alone. Next, after encountering an error with an annotation I changed things so that annotation contents are not recursively visited by PCPCheckAndHeal, like it was previously done in healTypeOfTerm. This lead to no failed tests, and I was not able to find any examples that would break the above changes.

[jchyb]

Hi @smarter, what do you think about this fix? It was difficult for me to judge from what direction to approach this problem, and I know you are well versed in all things Typer. I imagine that Typer may be more easy to break with small changes than PCPCheckAndHeal, but maybe the TypeTree with top level TermRef does not look right after all (again, it's difficult for me to judge these things still). If you see something off in the solution or it seems too hacky overall, please let me know

[smarter]

maybe the TypeTree with top level TermRef does not look right after all

It's not wrong technically, but I don't know if it's what should happen in this case. If I write s.type in source code, this will be represented by the typechecker as TermRef(..., s). The expression:

's

is desugared into

scala.quoted.runtime.Expr.quote(s)

when we typecheck this method, we get:

scala.quoted.runtime.Expr.quote[s.type](s)

Which is where the TermRef comes from. This only happens because fooImpl takes an Expr[Singleton], if we change it to take an Expr[Any] instead, then the type argument of quote is widened to a non-singleton. This is because type inference has a special-case where it doesn't widen type argument when the expected type is Singleton to give more precise types.

I would be a little bit hesitant to change the logic in PCPCheckAndHeal without having @nicolasstucki or @odersky double-check it first. Maybe instead we could change the behavior of type inference just for Expr.quote but I'm not sure how this could be done exactly.

[odersky]

I don't know enough about PCPCheckAndHeal to be able to say whether this is a fix or not. We have to wait for @nicolasstucki to come back.

[odersky]

What is the reasoning to allow us to ignore annotations here?

[jchyb]

I think it's wrong after all. Thank you for all the explanations and help. Back into the drafts it goes

[nicolasstucki]

It seems I fixed this issue with one of my recent PRs. I will add a regression test in #17143.