Yes. In a perfect world, it would be nice to have precise deps. But yes, I’m arguing that is it not worth the additional complexity, given how common of a pattern it already is to filter out irrelevant targets + the

for_sources_type=[PythonSources]

. It’s already fairly complex that plugin authors have to write

HydrateSourcesRequest(tgt[Sources], enable_codegen=True, for_sources_types=[PythonSources, ResourcesSources, FilesSources])

. Adding that they now have to say `DependenciesRequest(tgt[Dependencies], for_dependencies_type=PythonDependencies)`—which also means that we need to create a

PythonDependencies

field—is not great. (We only subclass

Dependencies

in exactly 1 place right now, and the sole reason we do it is to override docstring. It’d be great to avoid a proliferation of

Dependencies

subclasses. We will need subclasses for each codegen protocol target type, but that’s it, I think)