gh-143636: fix a crash when calling __replace__ on invalid SimpleNamespace instances

[picnixz]

• Issue: Type confusion in SimpleNamespace.__replace__ via re-entrant __new__ #143636

[serhiy-storchaka]

There are many ways to clone an object, you should make a decision whether call __init__ and/or __new__bor just create an instance of the base class. This is done differently for other types.

The following code follows the way of __reduce__() (which is also used in copy.copy()). It is equivalent to:

new_ns = type(ns)()
new_dict = dict(ns.__dict__)
for k, v in kwargs.items():
    new_dict[k] = v
for k, v in new_dict.items():
    new_ns.__dict__[k] = v

but more efficient.

This is what copy.copy() will execute.

We can implement this for the case when the result of type(ns)() is not a SimpleNamespace, but this will add an amount of code, and all this for the case which perhaps never occurs in real code. So, I am fine with simply raising an error.

[picnixz]

I didn't want to change the existing behavior. Maybe we could decide in a follow-up PR how to change this?

[serhiy-storchaka]

I think about something like "%T() returned %T object". __new__() is rarely called directly.

[picnixz]

I had this locally but I wondered whether this could be confused with __init__ or a simple function call.