Fixed bug that leads to incorrect metatype for a class object that is created by calling the metaclass with a two-argument form.

packages/pyright-internal/src/analyzer/typeEvaluator.ts

@@ -12958,7 +12958,7 @@ export function createTypeEvaluator(
             fileInfo.fileUri,
             ClassTypeFlags.ValidTypeAliasClass,
             ParseTreeUtils.getTypeSourceId(errorNode),
-            /* declaredMetaclass */ undefined,
+            metaclass,
             arg1Type.details.effectiveMetaclass
         );
         arg1Type.tupleTypeArguments.forEach((typeArg) => {

packages/pyright-internal/src/tests/samples/newType2.py

@@ -7,12 +7,10 @@ X1 = type("X1", (object,), {})
 X2 = type("X2", (object,), {})
 
 
-class A(X1):
-    ...
+class A(X1): ...
 
 
-class B(X2, A):
-    ...
+class B(X2, A): ...
 
 
 # This should generate an error because the first arg is not a string.
@@ -27,3 +25,11 @@ X5 = type("X5", (3,))
 
 X6 = type("", tuple({str}), {})
 X7 = type("", (float, str), {})
+
+
+class Meta1(type): ...
+
+
+X8 = Meta1("X8", (list,), {})
+reveal_type(X8, expected_text="type[X8]")
+reveal_type(type(X8), expected_text="type[Meta1]")