Heyo! A question about dependency inference. I have a Dash app build sort of like this

my_app/
+-- app.py
+-- pages/
  +-- page1.py
  +-- page2.py

The modules

page1.py

/

page2.py

are not explicitly imported anywhere - they don't even appear as strings, since Dash does some dynamic loading shenanigans. This means I have to tell pants to include them anyways. If in my build file I do this

python_sources(
    name="app_src",
    sources=["my_app/**/*.py", "!my_app/pages/*.py"],
    dependencies=[":pages"],
)

python_sources(
    name="pages",
    sources=["my_app/pages/*.py"],
)

everything works. If instead I just do

python_sources(
    name="app_src",
    sources=["my_app/**/*.py"],
)

it doesn't (I can see the files are indeed not included by running with

pants run --keep-sandboxes=always my_app/app.py

). This doesn't either, so I don't think it's about the

**

vs

*

python_sources(
    name="app_src",
    sources=["my_app/**/*.py", "my_app/pages/*.py"],
)

I can go with the first version, I'm just trying to understand what's going on.

I think the missing link here is that

python_sources

isn't a "real" target. In Pants parlance, it's a target generator. When evaluated, it expands to one actual target per matched source. Then, Pants dependency inference will handle those individually -- and you end up with no dependency inference again. When you create an explicit dependency, that ensures that all those expanded per-page-sources are picked up as dependencies regardless of the app sources.