Running SciPy Tests Locally#

Basic test writing and execution from within the Python interpreter is documented in the NumPy/SciPy testing guidelines. This page includes information about running tests from the command line using SciPy’s dev.py command line tool. Note: Before beginning, ensure that pytest is installed.

Note

The dev.py interface is self-documenting, in the sense that everything on this page and more (including usage examples for each command) can be accessed with python dev.py --help and for individual commands like python dev.py <command-name> --help. In this case, you can check python dev.py test --help.

To run all tests, navigate to the root SciPy directory at the command line and execute

python dev.py test

This builds SciPy (or updates an existing build) and runs the tests.

To run tests on a particular submodule, such as optimize, use the --submodule option:

python dev.py test -s optimize

To run a particular test module, use the Pytest syntax of --test (or -t):

python dev.py test -t scipy.<module>.tests.<test_file>

Example for scipy/optimize/tests/test_linprog.py file tests, run:

python dev.py test -t scipy.optimize.tests.test_linprog

To run a test class:

python dev.py test -t scipy.<module>.tests.<test_file>::<TestClass>

Example for TestLinprogRSCommon class from test_linprog.py:

python dev.py test -t scipy.optimize.tests.test_linprog::TestLinprogRSCommon

To run a particular test:

python dev.py test -t scipy.<module>.tests.<test_file>::<test_name>

Example for test_unknown_solvers_and_options from test_linprog.py:

python dev.py test -t scipy.optimize.tests.test_linprog::test_unknown_solvers_and_options

For tests within a class, you need to specify the class name and the test name:

python dev.py test -t scipy.<module>.tests.<test_file>::<TestClass>::<test_name>

Example:

python dev.py test -t scipy.optimize.tests.test_linprog::TestLinprogRSCommon::test_nontrivial_problem_with_guess

Other useful options include:

  • -v or --verbose, which activates the verbose option for more detailed output.

  • -b or --array-api-backend backend to include alternative array backends in array-api-compatible tests. See Support for the array API standard for details.

  • --coverage to generate a test coverage report in scipy/build/coverage/index.html. Note: pytest-cov must be installed.

  • -n or --no-build to prevent SciPy from updating the build before testing

  • -j or --parallel n to engage n cores when building SciPy; e.g. python dev.py test -j 4 engages four cores. As of #10172 this also runs the tests on four cores if pytest-xdist is installed.

  • -m full or --mode full to run the “full” test suite, including tests marked slow (e.g. with @pytest.mark.slow). Note that this does not run tests marked xslow; see Tips below.

  • -- to send remaining command line arguments to pytest instead of dev.py test. For instance, while -n sent to pytest.py activates the --no-build option, -n sent to pytest runs the tests on multiple cores; e.g. python dev.py test -- -n 4 runs tests using four cores. Note: pytest-xdist must be installed for testing on multiple cores. Common command line arguments for pytest include:

    • --durations=m to display durations of the slowest m tests. Use --durations=0 together with --durations-min=x to display durations of all tests with durations that exceed x seconds.

    • --fail-slow=x to cause test to fail if they exceed x seconds. (Note: pytest-fail-slow must be installed.)

    • --timeout=x to halt all test execution if any test time exceeds x seconds. (Note: pytest-timeout must be installed.)

For much more information about pytest, see the pytest documentation.

Tips:#

If you built SciPy from source but are having trouble running tests after a change to the codebase, try deleting the scipy/build directory. This forces dev.py to completely rebuild SciPy before performing tests.

There is an additional level of very slow tests (several minutes), which are disabled even when calling python dev.py test -m full. They can be enabled by setting the environment variable SCIPY_XSLOW=1 before running the test suite.

By default, tests that use Hypothesis run with the deterministic profile defined in scipy/scipy/conftest.py. This profile includes the Hypothesis setting derandomize=True so the same examples are used until Hypothesis, Python, or the test function are updated. To better use Hypothesis’ abilities to find counterexamples, select the nondeterministic profile by setting the environment variable SCIPY_HYPOTHESIS_PROFILE=nondeterministic before running the test suite. The number of examples that are run can be configured by editing the selected configuration, e.g. adding max_examples=100_000.