Didn't quite follow on the reasoning behind removing a lot of the logic for checking assignability. I don't really want to trigger a diagnostic if the user supplies an illegal value to the attribute, e.g. multidimensional arrays string[][], nullable struct arrays int?[] or reference types that aren't null or string or System.Type.

Didn't quite follow on the reasoning behind removing a lot of the logic for checking assignability. I don't really want to trigger a diagnostic if the user supplies an illegal value to the attribute, e.g. multidimensional arrays string[][], nullable struct arrays int?[] or reference types that aren't null or string or System.Type.

Much of the logic became unnecessary under the semantic model. The new implementation doesn't check the type of the array passed to the attribute (other than null), it only checks the type of each element.