Coverage backlog (python/pydantable)

This document snapshots statement-line gaps by area and complements Testing. mypy_plugin.py is omitted from measurement (pyproject.toml [tool.coverage.run] omit).

Baseline (full suite, --cov-fail-under=0)

Last regeneration (full pytest with branch coverage): ~83.1% total (pytest-cov reports 83.15% on the same run; the terminal TOTAL line may show 83%). About 1393 statements still missed (of 10084; branch coverage enabled). CI --cov-fail-under is 83 (see Makefile / _shared-ci.yml). Regenerate:

.venv/bin/python -m pytest -q -n auto --cov=pydantable --cov-report=term-missing:skip-covered --cov-fail-under=0

Top files by missed statements (refresh from latest term-missing)

Rough ordering by Miss column (largest gaps first; numbers drift each run):

Rank	Module	Miss (stmts)	Notes
1	pandas.py	~346	Façade; extend tests/third_party/
2	dataframe/_impl.py	~219	Core engine paths + errors
3	schema/_impl.py	~163	Unions / narrowing edge cases
4	pyspark/dataframe.py	~139	Parity smoke tests
5	dataframe_model.py	~131	I/O and async helpers
6	io/__init__.py	~80	Materialize/export branches
7	pyspark/sql/functions.py	~53	Representative calls only
8	io/extras.py	~42	Optional SDKs + mocks
9	grouped.py	~23	Convenience sum/mean/… and streaming
10	expressions.py	~26	Public Expr surfaces

Also high value: awaitable_dataframe_model.py (~11 miss), selectors.py (~27), rust_engine.py (~19).

Per-package focus (highest missing lines first)

Area	Modules (examples)	Notes
Facades	pandas.py, pyspark/dataframe.py, pyspark/sql/functions.py	Large surface; extend tests/third_party/ and parametrized API smoke tests.
Core DataFrame	dataframe/_impl.py, dataframe_model.py	Regression tests per feature; error paths via stub engine / invalid inputs.
Schema	schema/_impl.py	Narrowing, unions, edge dtypes; pairs with typing tests where helpful.
I/O	io/__init__.py, io/extras.py	Extras need optional deps (Excel, Arrow, cloud mocks); use optional_cloud / env gates.
Async model	awaitable_dataframe_model.py	Async materialization routes; mirror sync tests where possible.
Expressions	expressions.py	Focus on uncovered branches tied to public Expr APIs.
Small modules	engine/_binding.py, types.py, redaction.py, …	Cheap wins; keep tests/dataframe/test_coverage_supplement.py style tests.

Policy

• Prefer # pragma: no cover only for truly unreachable defensive code.
• Raise --cov-fail-under in CI only after the full suite is green at the new floor (see .github/workflows/_shared-ci.yml).
• Next numeric gate: increase 83 → 84 once pytest-cov total coverage is ≥ 84.0% on the Ubuntu + Python 3.11 leg (local numbers may differ slightly).
• Large PRs: run make test-cov (or download coverage.xml from CI), then make diff-cover so new/changed lines stay covered vs origin/main (see Testing). The Makefile uses --fail-under=85 on touched lines; raise that gradually as overall coverage climbs so new code stays tighter than legacy gaps.