import pytest @pytest.fixture def stepwise_testdir(testdir): # Rather than having to modify our testfile between tests, we introduce # a flag for whether or not the second test should fail. testdir.makeconftest( """ def pytest_addoption(parser): group = parser.getgroup('general') group.addoption('--fail', action='store_true', dest='fail') group.addoption('--fail-last', action='store_true', dest='fail_last') """ ) # Create a simple test suite. testdir.makepyfile( test_a=""" def test_success_before_fail(): assert 1 def test_fail_on_flag(request): assert not request.config.getvalue('fail') def test_success_after_fail(): assert 1 def test_fail_last_on_flag(request): assert not request.config.getvalue('fail_last') def test_success_after_last_fail(): assert 1 """ ) testdir.makepyfile( test_b=""" def test_success(): assert 1 """ ) # customize cache directory so we don't use the tox's cache directory, which makes tests in this module flaky testdir.makeini( """ [pytest] cache_dir = .cache """ ) return testdir @pytest.fixture def error_testdir(testdir): testdir.makepyfile( test_a=""" def test_error(nonexisting_fixture): assert 1 def test_success_after_fail(): assert 1 """ ) return testdir @pytest.fixture def broken_testdir(testdir): testdir.makepyfile( working_testfile="def test_proper(): assert 1", broken_testfile="foobar" ) return testdir def _strip_resource_warnings(lines): # Strip unreliable ResourceWarnings, so no-output assertions on stderr can work. # (https://github.com/pytest-dev/pytest/issues/5088) return [ x for x in lines if not x.startswith(("Exception ignored in:", "ResourceWarning")) ] def test_run_without_stepwise(stepwise_testdir): result = stepwise_testdir.runpytest("-v", "--strict-markers", "--fail") result.stdout.fnmatch_lines(["*test_success_before_fail PASSED*"]) result.stdout.fnmatch_lines(["*test_fail_on_flag FAILED*"]) result.stdout.fnmatch_lines(["*test_success_after_fail PASSED*"]) def test_fail_and_continue_with_stepwise(stepwise_testdir): # Run the tests with a failing second test. result = stepwise_testdir.runpytest( "-v", "--strict-markers", "--stepwise", "--fail" ) assert _strip_resource_warnings(result.stderr.lines) == [] stdout = result.stdout.str() # Make sure we stop after first failing test. assert "test_success_before_fail PASSED" in stdout assert "test_fail_on_flag FAILED" in stdout assert "test_success_after_fail" not in stdout # "Fix" the test that failed in the last run and run it again. result = stepwise_testdir.runpytest("-v", "--strict-markers", "--stepwise") assert _strip_resource_warnings(result.stderr.lines) == [] stdout = result.stdout.str() # Make sure the latest failing test runs and then continues. assert "test_success_before_fail" not in stdout assert "test_fail_on_flag PASSED" in stdout assert "test_success_after_fail PASSED" in stdout def test_run_with_skip_option(stepwise_testdir): result = stepwise_testdir.runpytest( "-v", "--strict-markers", "--stepwise", "--stepwise-skip", "--fail", "--fail-last", ) assert _strip_resource_warnings(result.stderr.lines) == [] stdout = result.stdout.str() # Make sure first fail is ignore and second fail stops the test run. assert "test_fail_on_flag FAILED" in stdout assert "test_success_after_fail PASSED" in stdout assert "test_fail_last_on_flag FAILED" in stdout assert "test_success_after_last_fail" not in stdout def test_fail_on_errors(error_testdir): result = error_testdir.runpytest("-v", "--strict-markers", "--stepwise") assert _strip_resource_warnings(result.stderr.lines) == [] stdout = result.stdout.str() assert "test_error ERROR" in stdout assert "test_success_after_fail" not in stdout def test_change_testfile(stepwise_testdir): result = stepwise_testdir.runpytest( "-v", "--strict-markers", "--stepwise", "--fail", "test_a.py" ) assert _strip_resource_warnings(result.stderr.lines) == [] stdout = result.stdout.str() assert "test_fail_on_flag FAILED" in stdout # Make sure the second test run starts from the beginning, since the # test to continue from does not exist in testfile_b. result = stepwise_testdir.runpytest( "-v", "--strict-markers", "--stepwise", "test_b.py" ) assert _strip_resource_warnings(result.stderr.lines) == [] stdout = result.stdout.str() assert "test_success PASSED" in stdout @pytest.mark.parametrize("broken_first", [True, False]) def test_stop_on_collection_errors(broken_testdir, broken_first): """Stop during collection errors. Broken test first or broken test last actually surfaced a bug (#5444), so we test both situations.""" files = ["working_testfile.py", "broken_testfile.py"] if broken_first: files.reverse() result = broken_testdir.runpytest("-v", "--strict-markers", "--stepwise", *files) result.stdout.fnmatch_lines("*error during collection*") def test_xfail_handling(testdir, monkeypatch): """Ensure normal xfail is ignored, and strict xfail interrupts the session in sw mode (#5547) """ monkeypatch.setattr("sys.dont_write_bytecode", True) contents = """ import pytest def test_a(): pass @pytest.mark.xfail(strict={strict}) def test_b(): assert {assert_value} def test_c(): pass def test_d(): pass """ testdir.makepyfile(contents.format(assert_value="0", strict="False")) result = testdir.runpytest("--sw", "-v") result.stdout.fnmatch_lines( [ "*::test_a PASSED *", "*::test_b XFAIL *", "*::test_c PASSED *", "*::test_d PASSED *", "* 3 passed, 1 xfailed in *", ] ) testdir.makepyfile(contents.format(assert_value="1", strict="True")) result = testdir.runpytest("--sw", "-v") result.stdout.fnmatch_lines( [ "*::test_a PASSED *", "*::test_b FAILED *", "* Interrupted*", "* 1 failed, 1 passed in *", ] ) testdir.makepyfile(contents.format(assert_value="0", strict="True")) result = testdir.runpytest("--sw", "-v") result.stdout.fnmatch_lines( [ "*::test_b XFAIL *", "*::test_c PASSED *", "*::test_d PASSED *", "* 2 passed, 1 deselected, 1 xfailed in *", ] )