pyteastrainer

A pytest plugin that fails the run when tests add no coverage, are too slow, or should be one parametrized test, and prints the exact changes that make it green. The minimal suite is solved exactly as an integer program (HiGHS).

pip install pyteastrainer
pytest --strain

Output

============================== pyteastrainer: RED ==============================
25 passed tests cover 41 branches; 9 of them (0.01 s of 0.02 s) cover all of it. To make teastrainer green:
  delete tests/test_cart.py::test_bad_coupon_rejected (duplicate of tests/test_cart.py::test_unknown_coupon)
  delete tests/test_cart.py::test_coupon_is_stored: every branch it reaches is reached by tests/test_cart.py::test_pct_coupon
  drop cases [0-1.0], [1--1.0] from the parametrize of tests/test_cart.py::test_add_rejects: ...
  parametrize: merge tests/test_cart.py::test_abs_coupon_floors_at_zero, test_pct_coupon into one test:
      @pytest.mark.parametrize('p0, p1, p2', [(2.0, 'FIVEOFF', 5.0), (200.0, 'TEN', 180.0)], ids=[...])
      def test_abs_coupon_floors_at_zero(p0, p1, p2):
          ...
Removing these leaves a suite with the same coverage, 9 tests, 0.01 s, that teastrainer passes.

Apply every action and rerun. A test that asserts something coverage cannot see is marked @pytest.mark.strain_exempt(reason="...") and always kept. cd demo && pytest --strain reproduces the full output.

Options

pytest-cov and xdist

pytest -n auto --cov=src --cov-branch --cov-context=test --strain

Per-test contexts are required. On Python 3.14 set COVERAGE_CORE=ctrace.

Docs

• docs/algorithm.md: formulations, convergence, parametrize detection, benchmark
• docs/releasing.md

Development

pip install -e .[test]
OMP_NUM_THREADS=2 MKL_NUM_THREADS=2 OPENBLAS_NUM_THREADS=2 pytest -n auto --cov
black --check . && pylint pyteastrainer tests