Also exclude when param is alias type #21191
Conversation
| @@ -1148,6 +1148,9 @@ object RefChecks { | |||
| def strippedResultType = Applications.stripImplicit(tp.stripPoly, wildcardOnly = true).resultType | |||
| def firstExplicitParamTypes = Applications.stripImplicit(tp.stripPoly, wildcardOnly = true).firstParamTypes | |||
| def hasImplicitParams = tp.stripPoly match { case mt: MethodType => mt.isImplicitMethod case _ => false } | |||
| def isAnyAlias = | |||
| val denot = tp.typeSymbol.denot | |||
| denot.isAliasType || denot.isOpaqueAlias | |||
There was a problem hiding this comment.
I'm not sure this is right. type Foo = Int is an alias type, isn't it?
But if you omit the opaque, the warning is correct: you can't ever call the extension method as an extension method.
I guess it's okay to not warn when there should be a warning, but the tricky part here is that opaque types look different depending on context.
Since I don't understand how much work is being done here and how much is being done by the later .isOpaqueAlias bit, I'm not sure this under-warns, but it seems like it might.
There was a problem hiding this comment.
Where the "receiver" is an alias, https://github.com/scala/scala3/blob/main/tests/warn/i16743.scala#L56
The current diff is at L1165 where the params are checked; the frozen_<:< is not sufficient.
For non-opaque aliases, I think the screws could be tightened where there can be no override (of the alias member).
The goal is indeed to prefer underwarn to overwarn.
Fixes #21190
Adjust test that params must not be opaque.
Other aliases are permitted, but could check if they are effectively final. String alias can't be overridden.