"static type has not been initialized" when type lives in another module

[rockstar]

Summary

Let's say we have two separate native modules, things_types which defines a Thing class, and things which has a get_all() -> List[Thing] function. Then I write the following code.

import things

my_things = things.get_all()
for thing in my_things:
    thing.do_stuff()

Expected

I would expect this to work just fine, for Thing instances to be returned from things.get_all() and for the pymethod do_stuff to be called on each instance.

Actual

An error occurs when calling things.get_all(), which is static type is not initialized.

Workaround

A workaround here is to import things_types as well, but I don't think that's very ergonomic.

Context

I'm creating a set of native modules to be provided to a rustpython interpreter as a sandbox for users. Asking users to import everything, even if they don't need it, is a little awkward. If there was a way I could, in my pymodule itself, intialize the static type easily, I would be happy to do that; I just couldn't find a way in the docs or examples or my journey through the source code (admittedly, my journey through the source code was to figure out the static type is not initialized meaning.)

[youknowone]

static types are initializable in Rust side code. So python code doesn't provide that much information.

Technically it means StaticType::init_builtin_types is not called for the type. It is usually called by <module_name>::make_module() if you created a module named <module_name> with #[pymodule].

One of the possible reason can be separated but not initialized python modules. If you created modules named A and B, referencing A::SomeType in B, running code including B but didn't run A::make_module() yet.

[youknowone]

Do you have any advice to make this less confusing?

[dchiquito]

FWIW I ran into this same problem trying to add typing to the stdlib. After several hours of digging I found the essential_init phase of the vm initialization process.

I think the only real way to improve this discovery process is documentation. Either an example which covers this use case (with useful comments), or some documentation on all the magical [py*] macros.