Normalize types before collecting parts determining implicit scope

This is necessary to ensure the implicit scope is consistent when involving match types, since they may or may not have been reduced before implicit search.
We can for example get different results when loading from tasty than when in the same run.

Fixes #20071

compiler/src/dotty/tools/dotc/typer/Implicits.scala

@@ -636,7 +636,7 @@ trait ImplicitRunInfo:
         else if implicitScopeCache.contains(t) then parts += t
         else
           partSeen += t
-          t.dealias match
+          t.dealias.normalized match
             case t: TypeRef =>
               if isAnchor(t.symbol) then
                 parts += t
@@ -663,7 +663,6 @@ trait ImplicitRunInfo:
               traverseChildren(t)
             case t =>
               traverseChildren(t)
-              traverse(t.normalized)
       catch case ex: Throwable => handleRecursive("collectParts of", t.show, ex)
 
       def apply(tp: Type): collection.Set[Type] =
@@ -775,6 +774,7 @@ trait ImplicitRunInfo:
    *     if `T` is of the form `(P#x).type`, the anchors of `P`.
    *   - If `T` is the this-type of a static object, the anchors of a term reference to that object.
    *   - If `T` is some other this-type `P.this.type`, the anchors of `P`.
+   *   - If `T` is match type or an applied match alias, the anchors of the normalization of `T`.
    *   - If `T` is some other type, the union of the anchors of each constituent type of `T`.
    *
    *  The _implicit scope_ of a type `tp` is the smallest set S of term references (i.e. TermRefs)
@@ -787,7 +787,7 @@ trait ImplicitRunInfo:
    *   - If `T` is a reference to an opaque type alias named `A`, S includes
    *     a reference to an object `A` defined in the same scope as the type, if it exists,
    *     as well as the implicit scope of `T`'s underlying type or bounds.
-   *   - If `T` is a reference to an an abstract type or match type alias named `A`,
+   *   - If `T` is a reference to an an abstract type or unreducible match type alias named `A`,
    *     S includes a reference to an object `A` defined in the same scope as the type,
    *     if it exists, as well as the implicit scopes of `T`'s lower and upper bound,
    *     if present.

tests/neg/i20071.scala

@@ -0,0 +1,28 @@
+
+trait Scope
+object Scope:
+  given i: Int = ???
+
+type ReferencesScope[S] >: Int <: Int
+
+type ScopeToInt[Why] = Why match
+  case Scope => Int
+
+def foo[T](using d: ReferencesScope[T]): Any = ???
+
+def bar[T](using d: ScopeToInt[T]): Any = ???
+
+def test: Unit =
+  foo[Scope] // ok
+  bar[Scope] // error
+
+  import Scope.i
+  bar[Scope] // ok
+
+  /*
+  Before the changes:
+  `ScopeToInt[Scope]` may or may not be reduced before implicit search,
+  thereby impacting the scope considered for the search. `Scope.i` is included
+  iff `Scope` still appears in the type, which is the case only before reduction.
+  In contrast, `ReferencesScope[Scope]` is ok since it will never lose the anchor.
+  */

tests/pos/i15183/test_2.scala

@@ -1,4 +1,8 @@
 // Fails in each cases below
+import Decoder.{derived as _, given}
+// NOTE Decoder.derived is already in the implicit scope
+//  but the others require an import as they depend on match type reduction
+
 enum Env derives Decoder:
   case Local,Sit,Prod