Fix #6588: synthesis of Type[T] should mapOver the type T #6589
Conversation
| def apply(t: Type) = t match { | ||
| case t @ TypeRef(NoPrefix, _) => | ||
| def apply(t: Type) = t.dealias match { | ||
| case t @ TypeRef(NoPrefix, _) if !ctx.inInlineMethod => |
There was a problem hiding this comment.
I did not understand the qualification !ctx.inInlineMethod. This looks fishy where it is. The other synthesized arguments do not depend on this test, so why would quoted.Type be different?
There was a problem hiding this comment.
In the case below in the outer match, we have the following logic:
I conjecture the motivation is that in inline methods, type parameters are treated as constants so that they are not restricted by phase consistency law, just like inline term parameters. /cc : @nicolasstucki
More concretely, without the test, the following test case i4514.scala fail:
object Foo {
inline def foo[X](x: X): Unit = ${fooImpl('x)}
def fooImpl[X: quoted.Type](x: X): quoted.Expr[Unit] = '{}
}The implicit search first looks for Type[Expr[X]], the mapOver will reach X, and the implicit search for Type[X] fails.
| inferImplicit(defn.QuotedTypeType.appliedTo(t), EmptyTree, span) match { | ||
| case SearchSuccess(tag, _, _) if tag.tpe.isStable => | ||
| tag.tpe.select(defn.QuotedType_splice) | ||
| case _ => | ||
| ok = false | ||
| t | ||
| } | ||
| case _ => t | ||
| case tp => mapOver(tp) |
There was a problem hiding this comment.
what is the motivation to mapOver? This is suspicious. Is it to dealias further term as you have in the tests? Mapping over may also trigger another implicit search for parts of the type which should only be done in the PCP checks.
There was a problem hiding this comment.
If we do not mapOver, for the following code:
import scala.quoted._
inline def foo[T:Type]: Int = 10
def main = {
type S = Int
type V = List[S]
foo[V]
}It will generate an evidence in staging:
type V = List[S]
{
val evidence$1: quoted.Type[List[S]] =
'[List[quoted.Type[Int]#$splice]]
10:Int
}The code will then crash ReifyQuotes.scala:153, as it assumes the prefixes of evidences are TermRefs.
|
@liufengyun as we discussed I replaced this PR with #6599 which contains an improved fix and an additional assertion. |
Fix #6588: synthesis of Type[T] should mapOver the type T