How to exclude transitive sub dependencies?

[jm-nab]

Is there a way to explicitly exclude transitive sub dependencies?

Say for example I have in my dependencies:

• super-spam

And super-spam has a dependency:

• sub-spam-dont-use
• main-spam

If I do pip install -e . via the pyproject.toml in my own project, is there a way to exclude installing sub-spam-dont-use?

I've tried:

• sub-spam-dont-use != *
• a custom build hook, where i tried to override dependencies, but i didn't get it working
• a custom path based build with a stubbed repo which just fakes the transitive dependency

The hack that I have been using is:

pip install . && pip uninstall sub-spam-dont-use

But it seems off since its not part of pyproject.toml.

Reference attempt:

[tool.hatch.build.hooks.custom]
path = "hatch_hooks.py"

from typing import Any
from hatchling.builders.hooks.plugin.interface import BuildHookInterface
from hatchling.plugin import hookimpl
import logging
logging.basicConfig(level=logging.INFO)


class CustomBuildHook(BuildHookInterface):
    def dependencies(self, version: str, build_data: dict[str, Any]) -> None:
        # List of dependencies to remove (stub out or ignore)
        ignored_dependencies = [
            'sub-spam-dont-use',
        ]
        # Get the current list of dependencies from the build data
        print(build_data)
        print(build_data)
        print(build_data)
        print(build_data)
        current_dependencies = build_data.get("dependencies", [])
        logging.info(f"Current dependencies: {current_dependencies}")

        # Filter out dependencies that should be ignored
        modified_dependencies = [
            dep for dep in current_dependencies if dep.name not in ignored_dependencies
        ]

        # Update the build data with the modified dependencies
        build_data["dependencies"] = modified_dependencies
        return build_data

[DrPyser]

Pip's dependency resolver cannot resolve a set of constraints which are contradictory (one package requiring sub-spam-dont-use
and another requiring a different version, or blacklisting it). So what you want is to bypass the purpose of the resolver, which is to ensure that the set of dependencies is consistant.

Other than your "hack" of installing then removing, you could also avoid the resolver completely and install that dependency without its subdependencies.
Basically I would do a 3 step process:

• resolve the dependencies and generate a list
  • either install in a throwaway environment and use pip freeze, or use pip install --quiet --report - --dry-run --no-deps -I -r requirements.txt | jq -r '.install[] | "\(.metadata.name)==\(.metadata.version)"'' to generate the list without installing
• filter out the dependencies you don't want from that list
• install only the filtered list using pip install --no-deps -r filtered-requirements.txt

[DrPyser]

But it's heavy over just relying on pyproject.toml dependencies, obviously.