Fix flake8 per-file settings, the fallback interpreter, and the undocumented flake8 settings

[ZayanKhan-12]

Description

Refs #190.

The headline ask on #190 — run flake8 itself, so flake8 plugins work — was resolved by @youben11's #656. pyls/plugins/flake8_lint.py shells out to the real flake8, so flake8-isort and friends load normally. What is left is the second half of the title, configuration, and it is genuinely broken in three ways. All three are things people in this thread hit and worked around by guesswork.

1. per-file-ignores and exclude silently never apply

The document is piped to flake8 over stdin so unsaved changes get linted, but nothing tells flake8 what the file is called, so it sees the name stdin. Every setting flake8 resolves per filename then cannot match.

Verified against flake8 3.8.4 with a setup.cfg containing per-file-ignores = pkg/legacy.py: E501,F401:

invocation	result
flake8 - (what pyls does today)	F401 and E501 both reported
flake8 --stdin-display-name=pkg/legacy.py -	clean — ignores applied

exclude behaves the same way: a file the user excluded is still linted today, and is correctly skipped once flake8 knows its name.

The plugin now passes --stdin-display-name, and omits it when the document has no path (unsaved, or a non-file URI).

To be precise about the limit of this fix: it does not change flake8's config discovery, which flake8 3.8 still does from the working directory. Passing an absolute display name from an unrelated cwd still failed to find the project config in my testing, so that is a separate problem and I have not claimed to fix it.

2. The fallback interpreter does not exist on Python 3 only systems

When the configured executable is not on PATH, the plugin fell back to python -m flake8. Bare python does not exist on a modern macOS or on most current Linux distributions. That second Popen was not guarded, so it raised too — and the error the user saw was FileNotFoundError: [Errno 2] No such file or directory: 'flake8', which names the wrong thing and suggests nothing.

This matters more than it looks, because installing pyls and flake8 into different environments is the normal case: pipx install python-language-server plus a project virtualenv. The fallback exists precisely for that, and it never worked.

It now falls back to sys.executable -m flake8, which is guaranteed to exist and is the interpreter that pip install python-language-server[flake8] put flake8 into. If that fails too it logs what to do rather than raising.

This fixes two tests that currently fail on develop: test_flake8_lint and test_flake8_unsaved fail on any machine without flake8 on PATH, for exactly this reason.

3. None of the flake8 settings were discoverable

The README says to see vscode-client/package.json for "the full set of supported configuration options". The flake8 plugin has been in the tree since #656 and had zero entries there, which is why this thread is mostly people guessing at names — and why @horseinthesky's max-line-length was ignored while maxLineLength worked.

The second commit adds the nine settings the plugin actually reads, recording that it is disabled by default.

It also corrects pyls.configurationSources, whose enum was wrong in both directions: it omitted flake8, the value the README tells people to use, and listed pyflakes, which is not a config source at all. pyls/config/config.py registers only flake8 and pycodestyle, and there is no pyflakes_conf.py.

Tests

Five new tests: the display name is passed with the document path; it is omitted when there is no path; the fallback uses sys.executable and not bare python; both attempts failing logs something actionable instead of raising; and an end-to-end test running the real flake8 against a project whose per-file-ignores should suppress a code.

I checked they hold the line by re-breaking each fix: removing --stdin-display-name fails 2 (including the end-to-end one), restoring the bare python fallback fails 3 — including the two pre-existing failures, which is direct evidence that bug caused them — and always sending a display name fails the no-path test.

Verification

Python 3.8 with jedi 0.17.2, matching the CI matrix.

	baseline on develop	with this change
pytest test/	12 failed, 99 passed, 8 skipped	10 failed, 106 passed, 8 skipped
difference in failing set	—	only test_flake8_lint and test_flake8_unsaved, both now passing
pycodestyle pyls test	clean	clean
pyflakes pyls test	1 pre-existing (_utils.py)	unchanged
pylint pyls test	84 messages	+1 consider-using-f-string, no new message types

That one added message is a .format() call, which Python 2.7 support requires; the baseline already has eleven. I fixed the one genuinely new message type I had introduced.

The package.json change is insert-only apart from the single enum line — I reverted a first attempt that round-tripped the JSON and reformatted unrelated entries.

Notes

I have not touched the README here. It already points at package.json, and that pointer becomes accurate with this change; my other open PR (#984) edits the README, and I did not want the two to conflict.

CLAUDE.md lives in #982 and is not duplicated here.

[palantirtech]

Thanks for your interest in palantir/python-language-server, @ZayanKhan-12! Before we can accept your pull request, you need to sign our contributor license agreement - just visit https://cla.palantir.com/ and follow the instructions. Once you sign, I'll automatically update this pull request.