__class__ and type()

[mattgreen]

I'm working on a Python binding to a DSL. It exposes proxies to "objects" in Rust code, which are structures with an ObjectKind field denoting the type (Array, Int, etc). They are exposed to Python as subclasses of an Object class, so Python code sees an Int with a value property that queries the underlying Rust value.

The Rust-side objects can change their type during runtime. This causes their Python proxies to get out of sync: an Int class in Python could be referencing an ObjectKind::Array!

After too much head-scratching, I realized PyO3 lets you override __class__ as a getter, at which point I could return the class that the object should be. Pretty cool! However, I noticed that type(obj) still seems to return the class it was originally created with, instead of querying my implementation of __class__.

Is this a PyO3 thing? It smells like the Python interpreter querying private state on the object vs asking the object what type it is. It isn't a dealbreaker but I figured I'd ask.

[davidhewitt]

Is this a PyO3 thing? It smells like the Python interpreter querying private state on the object vs asking the object what type it is. It isn't a dealbreaker but I figured I'd ask.

Python interpreter querying private state on the object is exactly what it is. Here's the relevant line of C in the implementation of the type builtin:

https://github.com/python/cpython/blob/87af12bff33b3e7546fa26158b7d8680ecb6ecec/Objects/typeobject.c#L3343

[mattgreen]

Thank you! I see now, I need to also setattr(obj, '__class__', new) when I observe the type change.