Replacement for pkg_resources.require()?

[sinoroc]

Is your feature request related to a problem? Please describe.

With setuptools' pkg_resources being deprecated, some are looking for replacement solutions.

Describe the solution you'd like

It would be nice if distlib offered an alternative to pkg_resources.require().

Describe alternatives you've considered

The next best thing right now is probably a mix of importlib.metadata and packaging, but as far as I understood it does not get all the way there.

Additional context

• https://setuptools.pypa.io/en/latest/pkg_resources.html#basic-workingset-methods
• Solution to check if a requirement is met without pkg_resources packaging-problems#664
• [WIP] Add guide on how to handle missing extras at runtime packaging.python.org#1606
• No package inspection routines outside pkg_resources packaging-problems#317

---

CC: @smheidrich

[vsajip]

I made a small change to the repo to refactor a function. With a couple of undocumented functions, you can (I think) do what you want. I did a simple smoke test:

$ python3 -m venv venv1
$ python3 -m venv venv2
$ venv1/bin/pip install textual 2>&1 > /dev/null
$ venv2/bin/pip install textual[syntax] 2>&1 > /dev/null

The textual package has only one extra, syntax.
Now, the following script test_script6.py:

#!/usr/bin/env python3
# -*- coding: utf-8 -*-
#
# Copyright (C) 2024 Vinay Sajip. MIT-licensed.
#
import argparse
import os
import sys

DEBUGGING = 'PY_DEBUG' in os.environ

from distlib.database import DistributionPath
from distlib.locators import DistPathLocator
# Undocumented functions imported here
from distlib.markers import interpret_parsed
from distlib.util import parse_requirement

def main():
    adhf = argparse.ArgumentDefaultsHelpFormatter
    ap = argparse.ArgumentParser(formatter_class=adhf)
    aa = ap.add_argument
    aa('reqt', metavar='REQT', help='Requirement')
    options = ap.parse_args()
    distpath = DistributionPath(include_egg=True)
    locator = DistPathLocator(distpath)
    sreq = options.reqt
    r = parse_requirement(sreq)
    failures = []
    d = locator.locate(sreq)
    if d is None:
        failures.append(r.name)
    else:
        print(d)
        for s in d.run_requires:
            dr = parse_requirement(s)
            if dr.marker:
                if r.extras:
                    for e in r.extras:
                        c = {'extra': e}
                        try:
                            v = interpret_parsed(dr.marker, c)
                        except Exception:
                            v = False
                        if v:
                            dd = locator.locate(dr.name)
                            if dd:
                                print(dd)
                            else:
                                failures.append(dr.name)
    if failures:
        print('Not found: %s' % failures)

if __name__ == '__main__':
    try:
        rc = main()
    except KeyboardInterrupt:
        rc = 2
    except Exception as e:
        if DEBUGGING:
            s = ' %s:' % type(e).__name__
        else:
            s = ''
        sys.stderr.write('Failed:%s %s\n' % (s, e))
        if DEBUGGING: import traceback; traceback.print_exc()
        rc = 1
    sys.exit(rc)

Seems to give the expected results:

$ venv1/bin/python test_script6.py foo
Not found: ['foo']
$ venv1/bin/python test_script6.py textual
textual 0.83.0
$ venv1/bin/python test_script6.py textual[syntax]
textual 0.83.0
Not found: ['tree-sitter-languages', 'tree-sitter']
$ venv2/bin/python test_script6.py foo
Not found: ['foo']
$ venv2/bin/python test_script6.py textual
textual 0.83.0
$ venv2/bin/python test_script6.py textual[syntax]
textual 0.83.0
tree-sitter 0.20.4
tree-sitter-languages 1.10.2

You can try out this logic with the latest repo and give me feedback. I can document the undocumented functions and cut a release in due course.

[smheidrich]

As far as I can tell, this takes pretty much the same amount of effort as it does currently with packaging and importlib.metadata, doesn't it? The snippet seems of a similar length as the one from pypa/packaging.python.org#1606 (comment) and the primitive operations used seem very similar (unless I misunderstood any of them):

Task	importlib.metadata / packaging	distlib
Parse a requirement string into its components	packging.requirements.Reqirement	parse_requirement
Find out if a package is installed in the current env, allow retrieving its version	importlib.metadata.distribution	Locator.locate
Evaluate a marker	Requirement.marker.evaluate	markers.interpret_parsed

Meanwhile, with the old pkg_resources.require it's just a single line that takes care of all of that lower-level stuff behind the scenes, and I guess what this issue was meant to be about is whether a replacement for that should make it into this project.

But as I wrote in pypa/packaging-problems#664 (comment), I personally don't think distlib is the right place for this in any case.

[vsajip]

I personally don't think distlib is the right place for this

I agree - in general, the things one might want to do in different branches of the above logic will depend on the exact use case. I put together the above to see how straightforward it would be to do using distlib only. And pkg_resources.require, while indeed simple, doesn't seem to allow quite as much flexibility as might be needed in some situations. Still, if that's all that's needed, fair enough.

[sinoroc]

I only wanted to get the discussion started, I do not particularly feel one way or another regarding this topic. Feel free to close whenever you feel like the discussion has run its course. :)