test: 100% coverage on src/ via real tests + fail_under gate (#37)

[cmeans-claude-dev]

Summary

Closes #37. Pushes statement coverage on src/pypi_winnow_downloads/ from 99% → 100%, then locks the gain with fail_under = 100 in pyproject.toml so future regressions trip CI immediately. No # pragma: no cover annotations, no coverage_exclude_lines patterns — every previously-uncovered defensive line is exercised by a real test.

Per-line coverage delta

File	Pre-PR	Post-PR	Lines added
__init__.py	60%	100%	covers 7-8 (PackageNotFoundError fallback)
__main__.py	97%	100%	covers 52 (if __name__ == "__main__": guard)
badge.py	100%	100%	(unchanged)
collector.py	99%	100%	covers 190 (non-int download_count raise)
config.py	96%	100%	covers 42 + 86 (defensive isinstance checks)
TOTAL	99%	100%	5 lines covered

New tests

Test	File	Mechanism
test_init_falls_back_to_dev_version_when_package_not_installed	tests/test_init.py (new file)	monkeypatch importlib.metadata.version to raise PackageNotFoundError, importlib.reload, assert __version__ == "0.0.0+dev". Restores in finally so other tests see the real version.
test_main_module_invokes_main_when_run_as_script	tests/test_main.py	runpy.run_module("pypi_winnow_downloads", run_name="__main__") triggers the guard at __main__.py:52 in-process. Stubs collector.collect to return an empty CollectorResult so the package runs without shelling out to pypinfo. Patches the stub BEFORE runpy so __main__.py's from .collector import collect reads the stub.
test_run_pypinfo_raises_on_non_integer_download_count	tests/test_collector.py	Feeds a row with download_count: "not-an-int"; asserts CollectorError("non-integer download_count").
test_load_config_rejects_non_mapping_service_value	tests/test_config.py	YAML with service: just-a-string; asserts ConfigError("'service' must be a mapping, got str").
test_load_config_rejects_non_list_packages_value	tests/test_config.py	YAML with packages: 42; asserts ConfigError("'packages' must be a list, got int"). Distinct from the existing null-packages test (separate branches).

Coverage config added to pyproject.toml

```toml
[tool.coverage.run]
source = ["src/pypi_winnow_downloads"]

[tool.coverage.report]
fail_under = 100
show_missing = true
```

uv run pytest --cov now exits non-zero if any line in src/ is uncovered. CI's test job will fail fast on a coverage regression rather than silently letting it land — same shape as the existing --strict-markers / --strict-config pytest gates.

Per the standing rule (feedback_no_pragma_no_cover.md): no exclusion mechanism is used to game the metric. If a future line is "hard to test", the answer is to write the test (as done here for __main__:52 via runpy) or to delete the line as dead code.

Test plan

• uv run pytest --cov — 71/71 passed (5 new tests, was 66), coverage gate green at 100.00%
• uv run ruff check src/ tests/ — clean
• uv run ruff format --check src/ tests/ — clean
• uv run mypy src/pypi_winnow_downloads/ — clean
• Per-line verification: --cov-report=term-missing shows 0 missing on every src/ file
• fail_under = 100 enforcement verified: temporarily commented one new test locally, pytest exited non-zero with "Required test coverage of 100.0% reached. Total coverage: <100" — gate works
• CI green on all jobs (lint, typecheck, test 3.11/3.12/3.13, deploy-smoke, codecov)

Out of scope

• Branch coverage (--cov-branch) — current setup is statement coverage only. Branch coverage would catch the implicit else arms of the new isinstance checks; reasonable follow-up if the noise stays manageable.
• Coverage of tests/ themselves — [tool.coverage.run] source deliberately excludes the test files, which is the standard practice. The previous report (without the source setting) was including them.

Closes #37

[codecov-commenter]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!

[cmeans]

QA round 1 — clean, no follow-ups

Issue #37 fully delivered. Reproduced locally:

Pre-PR	Post-PR	File	New test
60% (lines 7-8)	100%	__init__.py	test_init.py::test_init_falls_back_to_dev_version_when_package_not_installed — monkeypatch importlib.metadata.version to raise PackageNotFoundError, importlib.reload, assert __version__ == "0.0.0+dev". finally block restores real version + reloads so the rest of the suite isn't poisoned.
97% (line 52)	100%	__main__.py	test_main.py::test_main_module_invokes_main_when_run_as_script — runpy.run_module(..., run_name="__main__") exercises the guard in-process. The stub-collect-then-runpy ordering (with the comment about why) is the correct fix to the runpy-vs-prior-import gotcha I called out in our earlier conversation.
99% (line 190)	100%	collector.py	test_collector.py::test_run_pypinfo_raises_on_non_integer_download_count — feeds a row with "download_count": "not-an-int", asserts the exact CollectorError("non-integer download_count") message.
96% (lines 42, 86)	100%	config.py	test_config.py::test_load_config_rejects_non_mapping_service_value — YAML with service: just-a-string, asserts ConfigError("'service' must be a mapping, got str"). test_load_config_rejects_non_list_packages_value — YAML with packages: 42, asserts ConfigError("'packages' must be a list, got int"). Distinct from the existing null-packages test (separate branch).
99%	100%	TOTAL (258 stmts, 0 miss)	5 new tests

fail_under = 100 gate verification:

Reproduced your local check from a different angle. Ran the suite with --deselect tests/test_init.py to drop two lines off the bottom; pytest reported:

TOTAL                                      258      2    99%
FAIL Required test coverage of 100.0% not reached. Total coverage: 99.22%

Gate fires at the right boundary — a coverage regression in CI will exit non-zero.

Standing-rule compliance:

Standing rule	State
feedback_use_intentions_for_scheduled_followups (no # pragma: no cover cheats)	confirmed — git grep -nE "pragma:.*no.cover|coverage_exclude" returns only the two prose mentions in CHANGELOG and pyproject.toml comments explicitly disclaiming any annotation use. No actual exclusion mechanism in the codebase.
Test count in PR body matches pytest output	matches — PR body says 71/71, I observed 71 passed locally.
Coverage config focused on production source, not tests	[tool.coverage.run] source = ["src/pypi_winnow_downloads"] — correct shape. PR body's reasoning ("tests don't have defensive branches") is accurate.

Local verification on head 0c3d2dc:

Check	Result
uv run pytest	71 passed, 0 deselected, 0.23s
All 5 new tests run individually	5 passed
uv run pytest --cov --cov-report=term-missing	0 missing on every src/ file; Required test coverage of 100.0% reached. Total coverage: 100.00%
uv run ruff check, ruff format --check, mypy src	all clean
CI on PR head	all SUCCESS (test 3.11/3.12/3.13, lint, typecheck, deploy-smoke, qa-approved, on-push)

Note (not a finding): the runpy test emits one RuntimeWarning: 'pypi_winnow_downloads.__main__' found in sys.modules after import of package …. This is a known runpy interaction with test_main.py's top-level from pypi_winnow_downloads.__main__ import main import — sys.modules is already populated, then runpy re-executes. The warning is harmless (the test passes deterministically across the CI matrix, and coverage observes line 52 firing) and is the standard cost of testing __main__ guards via runpy. Could be filtered with a one-line filterwarnings = ['ignore:.*found in sys.modules.*:RuntimeWarning'] in pyproject.toml if the noise stays bothersome, but leaving it visible preserves signal for any other RuntimeWarning that might surface later. Operator's call.

No findings. Transitioning label to Ready for QA Signoff.

[cmeans]

Applying Ready for QA Signoff — see review above. Issue #37 fully delivered: 100% coverage on src/ via real tests (zero pragma annotations confirmed via repo-wide grep), fail_under=100 gate verified to fire correctly when coverage regresses. All 5 previously-uncovered defensive lines exercised by tests targeting the exact code path (not incidental coverage). 71/71 tests, ruff/format/mypy clean, CI green. One harmless runpy RuntimeWarning noted as informational.

[cmeans]

LGTM