Python Unit Testing

Note

Changes to this document must be approved by the System Architect (RFC-24). To request changes to these policies, please file an RFC.

This page provides technical guidance to developers writing unit tests for DM’s Python code base. See Software Unit Test Policy for an overview of LSST Stack testing. If you have legacy DM unittest suite-based code (code that sets up a unittest.TestSuite object by listing specific test classes and that uses lsst.utils.tests.run rather than unittest.main()), please refer to tech note SQR-012 for porting instructions. LSST tests should be written using the unittest framework, with default test discovery, and should support being run using the pytest test runner. If you want to jump straight to a full example of the standard LSST Python testing boilerplate without reading the background, read the section on memory testing later in this document.

Introduction to unittest

This document will not attempt to explain full details of how to use unittest but instead shows common scenarios encountered in the LSST codebase.

A simple unittest example is shown below:

import unittest
import math


class DemoTestCase1(unittest.TestCase):
    """Demo test case 1."""

    def testDemo(self):
        self.assertGreater(10, 5)
        with self.assertRaises(TypeError):
            1 + "2"


class DemoTestCase2(unittest.TestCase):
    """Demo test case 2."""

    def testDemo1(self):
        self.assertNotEqual("string1", "string2")

    def testDemo2(self):
        self.assertAlmostEqual(3.14, math.pi, places=2)


if __name__ == "__main__":
    unittest.main()

The important things to note in this example are:

• If the test is being executed using python from the command line the unittest.main() call performs the test discovery and executes the tests, setting exit status to non-zero if any of the tests fail.
• Test classes are executed in the order in which they appear in the test file. In this case the tests in DemoTestCase1 will be executed before those in DemoTestCase2.
• Test classes must, ultimately, inherit from unittest.TestCase in order to be discovered, and it is recommended that lsst.utils.tests.TestCase be used as the base class when afw objects are involved. The tests themselves must be methods of the test class with names that begin with test. All other methods and classes will be ignored by the test system but can be used by tests.
• Specific test asserts, such as assertGreater(), assertIsNone() or assertIn(), should be used wherever possible. It is always better to use a specific assert because the error message will contain more useful detail and the intent is more obvious to someone reading the code. Only use assertTrue() or assertFalse() if you are checking a boolean value, or a complex statement that is unsupported by other asserts.
• When testing that an exception is raised always use assertRaises() as a context manager, as shown in line 10 of the above example.
• If a test method completes, the test passes; if it throws an uncaught exception the test has failed.

Using a more advanced test runner

Note

pytest is not currently installed as part of a LSST stack installation. It can be installed using conda install pytest or pip install pytest.

All tests should be written such that they are runnable using pytest, a policy adopted in RFC-69. pytest provides a much richer execution and reporting environment for tests and can be used to run multiple tests files together. pytest test discovery is much more flexible than that provided by unittest, but LSST test files should not take advantage of that flexibility as it can lead to inconsistency in test reports that depend on the specific test runner. In particular, care must be taken not to have free functions that use a test prefix or non-TestCase test classes that are named with a Test prefix in the test files.

Note

The normal way to run tests is to use scons or scons tests. scons will then execute each of the tests and report any failures. Currently tests are executed one at a time using python and checking the command exit status. Eventually sconsUtils will be modified to call py.test directly with all Python test files (possibly with a different order each invocation). It is therefore important that during this transition period developers check that tests run correctly with pytest by explicitly running py.test when creating or updating test files.

pytest can run tests from more than one file in a single invocation and this can be used to verify that there is no state contaminating later tests. To run pytest use the py.test executable:

$ py.test tests/*.py

and to ensure that the order of test execution does not matter it is useful to sometimes run the tests in reverse order:

$ py.test `ls -r tests/*.py`

When writing tests it is important that tests are skipped using the proper unittest skipping framework rather than returning from the test early. unittest supports skipping of individual tests and entire classes using decorators or skip exceptions. LSST code sometimes raises skip exceptions in setUp() or setUpClass() class methods. It is also possible to indicate that a particular test is expected to fail, being reported as an error if the test unexpectedly passes. Expected failures can be used to write test code that triggers a reported bug before the fix to the bug has been implemented and without causing the continuous integration system to die. One of the primary advantages of using a modern test runner such as pytest is that it is very easy to generate machine-readable pass/fail/skip/xfail statistics to see how the system is evolving over time.

Using a shared base class

For some tests it is helpful to provide a base class and then share it amongst multiple test classes that are configured with different attributes. If this is required, be careful to not have helper functions prefixed with test. Do not have the base class named with a Test prefix and ensure it does not inherit from TestCase; if you do, pytest will attempt to find tests inside it and will issue a warning if none can be found. Historically LSST code has dealt with this by creating a test suite that only includes the classes to be tested, omitting the base class. This does not work in a pytest environment.

Consider the following test code:

import unittest


class BaseClass(object):
    def testParam(self):
        self.assertLess(self.PARAM, 5)


class ThisIsTest1(BaseClass, unittest.TestCase):
    PARAM = 3

if __name__ == "__main__":
    unittest.main()

which inherits from the helper class and unittest.TestCase and runs a single test without attempting to run any tests in BaseClass.

$ py.test -v coding/unit_test_snippets/unittest_baseclass.py
======================================= test session starts ========================================
platform darwin -- Python 3.4.3, pytest-2.9.1, py-1.4.30, pluggy-0.3.1 -- /usr/local/bin/python3.4
cachedir: coding/unit_test_snippets/.cache
rootdir: coding/unit_test_snippets, inifile:
collected 1 items

coding/unit_test_snippets/unittest_baseclass.py::ThisIsTest1::testParam PASSED

===================================== 1 passed in 0.02 seconds =====================================

LSST Utility Test Support Classes

lsst.utils.tests provides several helpful functions and classes for writing Python tests that developers should make use of.

Special Asserts

Inheriting from lsst.utils.tests.TestCase rather than unittest.TestCase enables new asserts that are useful for doing element-wise comparison of two floating-point numpy-like arrays or scalars.

lsst.utils.tests.TestCase.assertFloatsAlmostEqual

Asserts that floating point scalars and/or arrays are equal within the specified tolerance. The default tolerance is significantly tighter than the tolerance used by unittest.TestCase.assertAlmostEqual() or numpy.testing.assert_almost_equal(); if you are replacing either of those methods you may have to specify rtol and/or atol to prevent failing asserts.

lsst.utils.tests.TestCase.assertFloatsEqual

Asserts that floating point scalars and/or arrays are identically equal.

lsst.utils.tests.TestCase.assertFloatsNotEqual

Asserts that floating point scalars and/or arrays are not equal.

Note that assertClose and assertNotClose methods have been deprecated by the above methods.

Additionally, afw provides additional asserts that get loaded into lsst.utils.tests.TestCase when the associated module is loaded. These include methods for Coords, Geom (Angles, Pairs, Boxes), and Images, such as:

assertCoordsNearlyEqual

Assert that two coords represent nearly the same point on the sky (provided by lsst.afw.coord.utils).

assertAnglesNearlyEqual

Assert that two angles are nearly equal, ignoring wrap differences by default (provided by lsst.afw.geom.utils).

assertPairsNearlyEqual

Assert that two planar pairs (e.g. Point2D or Extent2D) are nearly equal (provided by lsst.afw.geom.utils).

assertBoxesNearlyEqual

Assert that two boxes (Box2D or Box2I) are nearly equal (provided by lsst.afw.geom.utils).

assertWcsNearlyEqualOverBBox

Compare pixelToSky and skyToPixel for two WCS over a rectangular grid of pixel positions (provided by lsst.afw.image.basicUtils).

assertImagesNearlyEqual

Assert that two images are nearly equal, including non-finite values (provided by lsst.afw.image.testUtils).

assertMasksEqual

Assert that two masks are equal (provided by lsst.afw.image.testUtils).

assertMaskedImagesNearlyEqual

Assert that two masked images are nearly equal, including non-finite values (provided by lsst.afw.image.testUtils).

Testing Executables

In some cases the test to be executed is a shell script or a compiled binary executable. In order for the test running environment to be aware of these tests, a Python test file must be present that can be run by pytest. If none of the tests require special arguments and all the files with the executable bit set are to be run, this can be achieved by copying the file $UTILS_DIR/tests/testExecutables.py to the relevant tests directory. The file is reproduced here:

import unittest
import lsst.utils.tests


class UtilsBinaryTester(lsst.utils.tests.ExecutablesTestCase):
    pass

EXECUTABLES = None
UtilsBinaryTester.create_executable_tests(__file__, EXECUTABLES)

if __name__ == "__main__":
    unittest.main()

The EXECUTABLES variable can be a tuple containing the names of the executables to be run (relative to the directory containing the test file). None indicates that the test script should discover the executables in the same directory as that containing the test file. The call to create_executable_tests initiates executable discovery and creates a test for each executable that is found.

In some cases an explicit test has to be written either because some precondition has to be met before the test will stand a chance of running or because some arguments have to be passed to the executable. To support this the assertExecutable method is available:

def testBinary(self):
    self.assertExecutable("binary1", args=None,
                          root_dir=os.path.dirname(__file__))

where binary1 is the name of the executable relative to the root directory specified in the root_dir optional argument. Arguments can be provided to the args keyword parameter in the form of a sequence of arguments in a list or tuple.

Note

The LSST codebase is currently in transition such that sconsUtils will run executables itself as well as running Python test scripts that run executables. Do not worry about this duplication of test running. When the codebase has migrated to consistently use the testing scheme described in this section sconsUtils will be modified to disable the duplicate testing.

Memory and file descriptor leak testing

lsst.utils.tests.MemoryTestCase is used to detect memory leaks in C++ objects and leaks in file descriptors. MemoryTestCase should be used in all test files where utils is in the dependency chain, even if C++ code is not explicitly referenced.

This example shows the basic structure of an LSST Python unit test module, including MemoryTestCase (the highlighted lines indicate the memory testing modifications):

import unittest
import lsst.utils.tests


class DemoTestCase(lsst.utils.tests.TestCase):
    """Demo test case."""

    def testDemo(self):
        self.assertNotIn("i", "team")


class MemoryTester(lsst.utils.tests.MemoryTestCase):
    pass


def setup_module(module):
    lsst.utils.tests.init()


if __name__ == "__main__":
    lsst.utils.tests.init()
    unittest.main()

which ends up running the single specified test plus the two running as part of the leak test:

$ py.test -v unittest_runner_example.py
============================= test session starts ==============================
platform darwin -- Python 2.7.11, pytest-2.8.5, py-1.4.31, pluggy-0.3.1 -- ~/lsstsw/miniconda/bin/python
cachedir: .cache
rootdir: .../coding/unit_test_snippets, inifile:
collected 3 items

unittest_runner_example.py::DemoTestCase::testDemo PASSED
unittest_runner_example.py::MemoryTester::testFileDescriptorLeaks <- .../lsstsw/stack/DarwinX86/utils/12.0.rc1+f79d1f7db4/python/lsst/utils/tests.py PASSED
unittest_runner_example.py::MemoryTester::testLeaks <- .../lsstsw/stack/DarwinX86/utils/12.0.rc1+f79d1f7db4/python/lsst/utils/tests.py PASSED

=========================== 3 passed in 0.28 seconds ===========================

Note that MemoryTestCase must always be the final test suite. For the memory test to function properly the lsst.utils.tests.init() function must be invoked before any of the tests in the class are executed. Since LSST test scripts are required to run properly from the command-line and when called from within pytest, the init() function has to be in the file twice: once in the setup_module function that is called by pytest whenever a test module is loaded (pytest will not use the __main__ code path), and also just before the call to unittest.main() call to handle being called with python.

Unicode

It is now commonplace for Unicode to be used in Python code and the LSST test cases should reflect this situation. In particular file paths, externally supplied strings and strings originating from third party software packages may well include code points outside of US-ASCII. LSST tests should ensure that these cases are handled by explicitly including strings that include code points outside of this range. For example,

• file paths should be generated that include spaces as well as international characters,
• accented characters should be included for name strings, and
• unit strings should include the µm if appropriate.