Make toplevel opaque types transparent to the whole file #21530
Conversation
| @@ -1620,6 +1620,7 @@ class TypeComparer(@constructorOnly initctx: Context) extends ConstraintHandling | |||
|
|
|||
| def findEnclosingThis(moduleClass: Symbol, from: Symbol): Type = | |||
| if ((from.owner eq moduleClass) && from.isPackageObject && from.is(Opaque)) from.thisType | |||
| else if (from eq moduleClass.owner) && moduleClass.isTopLevelDefinitionsObject && moduleClass.is(Opaque) then moduleClass.thisType | |||
There was a problem hiding this comment.
Isn't that going to create a C$.this reference outside of C$? If yes, I don't think that's supposed to be possible/valid.
There was a problem hiding this comment.
The reference wasn't outside when the user wrote it. The compiler moved it so it's nested in a package object, but to the user the call site isn't from "outside".
There was a problem hiding this comment.
I understand, but still ... from an elaborated, TASTy-level point of view, this is invalid. A C.this reference has no meaning outside of C. This is worse than being "not accessible" because of protected or even "not visible" because of private; here it does not even exist.
There was a problem hiding this comment.
I'd argue that C.this, where C is a singleton module is just an alternative to C.type. Which, btw, is what widenInferred does as a final cleanup.
There was a problem hiding this comment.
I'd argue that C.this, where C is a singleton module is just an alternative to C.type.
It represents the same set of values at run-time, sure, but it's not valid everywhere C.type is valid.
Which, btw, is what
widenInferreddoes as a final cleanup.
AFAICT, widentInferred turns pre.C$ into pre.C; not C$.this into pre.C nor the converse.
Fixes #15050