Fix ActiveRecordRelations Model.new typing

[bitwise-aiden]

Motivation

As pointed out in this issue: #1981

The typing for Model.new accepts an T::Array which it never actually does. The original discussion was around this being acceptable, but after trying to fix issues from this change to the compiler I'm now thinking that it is better to type this appropriately.

Consider:

def some_untyped_method
  {
    a: 1,
    b: 1,
    c: 1,
  }
end

SomeModel.new(some_untyped_method)
#             ^^^^^^^^^^^^^^^^^^^ T.untyped = return type becomes `T::Array`

Because of how Sorbet selects the signature in the case of T.untyped and overloaded signatures, it will tend to pick the first viable, which in the case of #new will be T::Array. This then implicitly means that any arguments to SomeModel.new will have to be typed otherwise it picks a type that is never possible for this method.

Implementation

I'm aware this may not be the correct approach, but I want to stat a conversation so we can best solve it. As for what I have here I've added a new with the removal of the T::Array signature to the model class after the extend. This will make it so that it overrides the signatures in CommonRelationMethods and avoid the T::Array issue.

Tests

Updated tests to reflect change.

[paracycle]

I don't think this is needed, since the attributes param is already optional.

[bitwise-aiden]

Yeah, you're right! It was originally a guard for the case where we had T::Array since it would default to that one in the case of SomeModel.new, but that doesn't apply here.