🚀 feature request

related: #341

Relevant Rules

entry_point

Description

Currently, the entry_point is a pre-generated py_binary to expose the console_scripts. It works for tools that takes file as input and with limited reliance on module resolution e.g. static analysis tools, flake8, black.

However, it doesn't quite work with things that expect dependencies on the import path, which is common for running dev server for web development frameworks, e.g flask, uvicorn, strawberry. In order for these use cases to work, we need a mechanism to properly propagate the import paths when executing the console scripts target.

Describe the solution you'd like

One solution would be to convert entry_point into a macro that dynamically generate py_binary on the fly, so that it can take additional deps attribute to construct the console scripts py_binary.

Roughly can be implemented as below?

In the generated @pip//:requirements.bzl

load("@rules_python//python:defs.bzl", "py_binary")
load("@pip//:requirements.bzl", "requirement")

def new_entry_point(name, pkg, script = None, **kwargs):
    if not script:
        script = pkg
    entry_point_py = "@{repo_prefix}" + _clean_name(pkg) + "//:{entry_point_prefix}_" + script + ".py"

    py_binary(
        name = name,
        srcs = kwargs.pop("srcs", []) + [entry_point_py],
        main = entry_point_py,
        deps = kwargs.pop("deps", []) + [requirement(pkg)],
        **kwargs
    )

In BUILD.bazel

new_entry_porint(
    name = "strawberry",
    pkg = "strawberry-graphql",
    script = "strawberry",
    deps = [
        ":schema",
    ],
)

Describe alternatives you've considered

use environment variables PYTHONPATH when calling PYTHONPATH=... bazel run ... . but it's hard and not trivial to figure out all import paths and add to it.