test: add HTML-fixture regression scaffolding for brittle compat parsing

[erseco]

Summary

Adds fast, network-free characterization tests under tests/unit/ that exercise the version-sensitive HTML parsers in py_moodle.compat against hand-authored HTML fixture files. Previously these fragile parsers (login token, sesskey, folder file listing, dashboard version) had no coverage in the fast make test-unit gate — the only compat tests live in tests/test_compat.py, which is auto-marked integration and skipped by default. Now an accidental selector/regex regression, or a future Moodle-markup change, is caught by the fast unit layer. Test-only: no source files changed.

Linked issue

Closes #68

Specification (SDD)

Add a fixtures directory tests/unit/fixtures/html/ with small, representative, fake-secret HTML documents (login page with a logintoken hidden input, a page embedding the M.cfg sesskey, a folder view listing pluginfile.php filenames, and modern/legacy/no-version dashboards), each with a provenance comment. Add tests/unit/test_html_fixtures.py loading those fixtures via pathlib relative to __file__ and asserting the expected extraction for extract_login_token, extract_sesskey (present and absent), extract_folder_filenames, and extract_version_from_dashboard (modern, legacy, and absent-marker), plus at least one legacy-vs-modern contrast. Uses the existing public compat surface read-only: DEFAULT_COMPATIBILITY / LEGACY_COMPATIBILITY strategy methods and the module-level extract_version_from_dashboard / get_strategy_for_version. No behavior change; this is regression/characterization coverage.

TDD evidence

Red-green with the source held read-only, so the natural "red" is the fixtures not existing yet.

1. Wrote tests/unit/test_html_fixtures.py first and ran it before creating any fixture:

$ .venv/bin/pytest tests/unit/test_html_fixtures.py -q
FileNotFoundError: [Errno 2] No such file or directory:
  '.../tests/unit/fixtures/html/dashboard_modern.html'
... 11 failed

2. Authored the six fixtures, re-ran, and one test surfaced a real defect: my provenance comment literally contained the sesskey regex "sesskey":"([^\"]+)", which the parser matched first (returning ([^ instead of the real value). Fixed the comment to not embed a matching literal:

$ .venv/bin/pytest tests/unit/test_html_fixtures.py -q
E  AssertionError: assert '([^' == 'fakeSesskey1a2b3c'

3. After the fix, focused tests are green and the full unit suite has no regressions:

$ .venv/bin/pytest tests/unit/test_html_fixtures.py
11 passed, 1 warning in 0.04s
$ .venv/bin/pytest tests/unit -q
192 passed

Non-vacuous proof (one-off, NOT committed): renaming the login-token attribute in the fixture (name="logintoken" -> name="logintoken_RENAMED", simulating a Moodle markup change) makes the parser return '' and the test fail, then reverting restores green:

$ .venv/bin/pytest tests/unit/test_html_fixtures.py::test_extract_login_token_from_modern_login_page -q
E  AssertionError: assert '' == 'fake-logintoken-9f8a7b6c5d4e3f2a'
... 1 failed
# (fixture reverted -> 11 passed again)

Test plan

• Added/authored failing test(s) first.
• Confirmed red before implementing (fixtures absent -> FileNotFoundError).
• Implemented minimal change (fixtures only; no source touched).
• Focused tests green (11 passed).
• Full pytest tests/unit green (192 passed).

Commands run:

.venv/bin/pytest tests/unit/test_html_fixtures.py -q          # red: 11 failed (FileNotFoundError)
.venv/bin/pytest tests/unit/test_html_fixtures.py -q          # after fixtures: 1 failed (comment matched regex)
.venv/bin/pytest tests/unit/test_html_fixtures.py             # 11 passed
.venv/bin/pytest tests/unit -q                                # 192 passed
.venv/bin/black --check src tests                             # 84 files unchanged
.venv/bin/isort --check-only src tests                        # clean
.venv/bin/flake8                                              # clean
# non-vacuous demo (not committed):
.venv/bin/pytest tests/unit/test_html_fixtures.py::test_extract_login_token_from_modern_login_page -q  # 1 failed after corrupting fixture

Backwards compatibility

Fully backwards compatible. Test-only change: adds new files under tests/unit/, touches no source and no existing tests, uses only the existing public/read-only compat surface.

Documentation

None. Per the issue, the contributor-facing "how brittle-fixture tests work" note is owned by the sibling test-layer-docs issue; docs/development.md is intentionally untouched. docs/cli.md is unaffected (no CLI options changed).

Notes for reviewers

• Stayed strictly within file-ownership boundaries. Files added (only these):
  • tests/unit/test_html_fixtures.py
  • tests/unit/fixtures/html/login_page_modern.html
  • tests/unit/fixtures/html/sesskey_config_modern.html
  • tests/unit/fixtures/html/folder_view_modern.html
  • tests/unit/fixtures/html/dashboard_modern.html
  • tests/unit/fixtures/html/dashboard_legacy.html
  • tests/unit/fixtures/html/dashboard_no_version.html
• src/py_moodle/compat.py, tests/test_compat.py, tests/conftest.py, and docs/* are all untouched.
• extract_folder_filenames takes a BeautifulSoup (not a raw string), so the folder test parses the fixture with the same lxml parser the production code uses.
• The legacy-vs-modern contrast is covered two ways: extract_folder_filenames is asserted identical across DEFAULT_COMPATIBILITY and LEGACY_COMPATIBILITY, and get_strategy_for_version is asserted to pick LegacyCompatibilityStrategy for the 3.x dashboard vs ModernCompatibilityStrategy for the 5.x dashboard.
• All embedded tokens/sesskeys/versions are obviously-fake placeholders; each fixture carries a provenance comment.