In my app, I have a sub-package having sub-packages that looks like the following:

a
├── ba
│   ├── __init__.py
│   └── resources.py
├── bb
│   ├── __init__.py
│   └── resources.py
├── bc
│   ├── __init__.py
│   └── resources.py
├── bd
│   ├── __init__.py
│   └── resources.py
├── be
│   ├── __init__.py
│   └── resources.py
├── bf
│   ├── __init__.py
│   └── resources.py
└── __init__.py

a/__init__.py

uses

importlib.import_module

to load all the

resources.py

files under it (done to do auto-registration on a factory). This isn't picked up by the pants' inference system. I thought to have

a/BUILD

with something like

python_sources(
  overrides={
    "__init__.py": {
        "dependencies": ["*/**/*.py"]
    }
  }
)

But pants is complaining. I'm not sure I can use glob on

dependencies

anyway. Therefore, I'm not sure how to handle that situation. Can you guys tell me if my approach is ok and what is missing to make it works?

Yeah, globs don't work with

dependencies

, unfortunately. So you have three options: 1) Rewrite the code to use something Pants can infer 2) Hardcode each dependency 3) Automate with a light weight dependency inference plugin, which will auto-detect

resource.py

in subtrees Happy to help w/ #3 if you're interested

ok, that's what I thought. No big deal for now but I'll probably go back to 3) soon. Meanwhile, I think I'm starting to get a grip on how the dependencies and BUILD files work: All my pytest tests are successful :-)