Make check-haproxy-stats Python 2.7 / 3.7 dual-compatible

[jjpersch]

Summary

Makes this package Python 2.7 / 3.7 dual-compatible. No source code changes were needed: the package was already dual-compatible and only the packaging metadata claimed otherwise. This PR adds the 3.7 tox env and setup.py classifier, and drops the unused future dependency.

Background

All three entry-point modules already carry from __future__ import print_function, and there is no py2-only syntax anywhere in the package. The 2.7-only metadata dates to May 2017, when #3 (massage tox list -- our dep is unfortunately 27 only) cut envlist down to py27, flake8 and #5 removed the Programming Language :: Python :: 3* classifiers. Both cited the same reason: "one of our dependencies is python 2.7 [only]".

That reason does not hold up. The dependency in question is haproxy-stats. haproxystats is a single module, and upstream (bcicen/haproxy-stats) left haproxystats/__init__.py untouched from 2016-03-10 until 2018-03-01 -- so the file is identical across the entire 1.3 line. It contains no py2-only constructs, and it imports and runs on 3.7. The 2017 conclusion appears never to have been tested, and it has kept this package pinned to a long-dead interpreter ever since.

future is a separate matter: nothing in the package imports it. The modules use __future__, which is stdlib and needs no third-party package.

What changed

• setup.py: dropped future from install_requires; nothing imports it.
• setup.py: added the Programming Language :: Python :: 3.7 classifier alongside 2.7. Deliberately not adding a bare :: 3 -- only 3.7 has been verified.
• tox.ini: envlist = py27, py37, flake8, so both interpreters are exercised rather than just 2.7. skip_missing_interpreters is deliberately left at its default of false. The point of this change is to prove that both interpreters pass; letting tox exit 0 when one of them is absent would quietly turn that back into "whatever happened to be installed passed", which is how the stale 2.7-only claim went unchallenged for as long as it did. Anyone who lacks an interpreter locally can run a single env (tox -e py27) or pass --skip-missing-interpreters, which takes precedence over the ini and so does not weaken the committed config. Note this is not a new constraint: envlist already required 2.7, which is by now the harder of the two to have on hand.
• No changes to any module under check_haproxy_stats/. The source did not need porting.

Verification

Both interpreters were run against haproxy-stats 1.3, installed from the upstream git revision that carries that version:

• python setup.py test (the exact command the tox testenv runs) on CPython 2.7.18: 27 tests, OK (skipped=1).
• python setup.py test on CPython 3.7.17: 27 tests, OK (skipped=1).
• The single skip is intentional in both cases -- test_report_haproxy_rates, skipped as "gmetric reporting disabled".
• All three console scripts (check-haproxy-stats-up, check-haproxy-stats-5xx, metrics-haproxy-stats-5xx) start and serve --help on both interpreters, and check-haproxy-stats-up was additionally run against an absent stats endpoint on both to exercise the real haproxystats/requests network path; it degrades to the expected error on each.
• The same 27 tests pass identically on the unmodified tree, which is the evidence that this PR is metadata-only and changes no behavior.

Two pre-existing problems are NOT addressed here, because both are outside the scope of the interpreter work and one is in a module this PR deliberately does not touch:

• tox -e flake8 is already red on master, independent of this change: F401 for socket, subprocess, and .haproxy_util in metrics_haproxy_stats_5xx.py, plus one E501 at check_haproxy_stats_up.py:56. The three F401s are leftovers from 873632e/3a7a0c1, which commented out the gmetric body but left its imports behind.
• metrics_haproxy_stats_5xx.py is now entirely commented out -- _report_haproxy_rates just return 0s. The module and its metrics-haproxy-stats-5xx console script are a deletion candidate, to be handled in their own PR. Deleting them also clears three of the four flake8 findings above.

Deployment plan and risks

Low risk. This PR ships no behavior change: the installed Python is identical to 3.7.0, and the only packaging change is one fewer name in install_requires, for a package that was never imported.

Merging this alone deploys nothing. Follow-up, as separate work: cut a release, then update downstream consumers that pin this package by exact version. Consumers that install with --no-deps must continue to pin haproxy-stats explicitly themselves, since install_requires is not resolved in that mode.

Declaring 3.7 support does not change which interpreter any existing consumer runs; it only removes this package as a blocker for consumers that want to move.

[Copilot]

Pull request overview

Updates project packaging/testing metadata to reflect and exercise Python 2.7 + Python 3.7 compatibility (without changing runtime code).

Changes:

• Add py37 to tox envlist so tests run under Python 3.7 as well as 2.7.
• Remove unused future from install_requires.
• Add the Programming Language :: Python :: 3.7 classifier in setup.py.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated no comments.

File	Description
tox.ini	Expands tox envlist to include py37 for dual-interpreter test runs.
setup.py	Drops unused future dependency and adds a Python 3.7 trove classifier.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

Pull request overview

Copilot reviewed 2 out of 2 changed files in this pull request and generated no new comments.

Suppressed comments (1)

tox.ini:2

• Adding py37 to the default envlist makes tox fail with InterpreterNotFound on machines/CI images that don’t have an explicit Python 3.7 interpreter installed. If the goal is “run 3.7 when available”, consider enabling skip_missing_interpreters so contributors can still run tox without installing 3.7 locally.

[tox]
envlist = py27, py37, flake8