# coding: utf8 # Licensed under the Apache License: http://www.apache.org/licenses/LICENSE-2.0 # For details: https://bitbucket.org/ned/coveragepy/src/default/NOTICE.txt """Tests for process behavior of coverage.py.""" import glob import os import os.path import re import sys import textwrap import coverage from coverage import env, CoverageData from tests.coveragetest import CoverageTest HERE = os.path.dirname(__file__) class ProcessTest(CoverageTest): """Tests of the per-process behavior of coverage.py.""" def number_of_data_files(self): """Return the number of coverage data files in this directory.""" num = 0 for f in os.listdir('.'): if f.startswith('.coverage.') or f == '.coverage': num += 1 return num def test_save_on_exit(self): self.make_file("mycode.py", """\ h = "Hello" w = "world" """) self.assert_doesnt_exist(".coverage") self.run_command("coverage run mycode.py") self.assert_exists(".coverage") def test_environment(self): # Checks that we can import modules from the test directory at all! self.make_file("mycode.py", """\ import covmod1 import covmodzip1 a = 1 print('done') """) self.assert_doesnt_exist(".coverage") out = self.run_command("coverage run mycode.py") self.assert_exists(".coverage") self.assertEqual(out, 'done\n') def make_b_or_c_py(self): """Create b_or_c.py, used in a few of these tests.""" self.make_file("b_or_c.py", """\ import sys a = 1 if sys.argv[1] == 'b': b = 1 else: c = 1 d = 1 print('done') """) def test_combine_parallel_data(self): self.make_b_or_c_py() out = self.run_command("coverage run -p b_or_c.py b") self.assertEqual(out, 'done\n') self.assert_doesnt_exist(".coverage") self.assertEqual(self.number_of_data_files(), 1) out = self.run_command("coverage run -p b_or_c.py c") self.assertEqual(out, 'done\n') self.assert_doesnt_exist(".coverage") # After two -p runs, there should be two .coverage.machine.123 files. self.assertEqual(self.number_of_data_files(), 2) # Combine the parallel coverage data files into .coverage . self.run_command("coverage combine") self.assert_exists(".coverage") # After combining, there should be only the .coverage file. self.assertEqual(self.number_of_data_files(), 1) # Read the coverage file and see that b_or_c.py has all 7 lines # executed. data = coverage.CoverageData() data.read_file(".coverage") self.assertEqual(data.line_counts()['b_or_c.py'], 7) def test_combine_parallel_data_in_two_steps(self): self.make_b_or_c_py() out = self.run_command("coverage run -p b_or_c.py b") self.assertEqual(out, 'done\n') self.assert_doesnt_exist(".coverage") self.assertEqual(self.number_of_data_files(), 1) # Combine the (one) parallel coverage data file into .coverage . self.run_command("coverage combine") self.assert_exists(".coverage") self.assertEqual(self.number_of_data_files(), 1) out = self.run_command("coverage run -p b_or_c.py c") self.assertEqual(out, 'done\n') self.assert_exists(".coverage") self.assertEqual(self.number_of_data_files(), 2) # Combine the parallel coverage data files into .coverage . self.run_command("coverage combine") self.assert_exists(".coverage") # After combining, there should be only the .coverage file. self.assertEqual(self.number_of_data_files(), 1) # Read the coverage file and see that b_or_c.py has all 7 lines # executed. data = coverage.CoverageData() data.read_file(".coverage") self.assertEqual(data.line_counts()['b_or_c.py'], 7) def test_append_data(self): self.make_b_or_c_py() out = self.run_command("coverage run b_or_c.py b") self.assertEqual(out, 'done\n') self.assert_exists(".coverage") self.assertEqual(self.number_of_data_files(), 1) out = self.run_command("coverage run --append b_or_c.py c") self.assertEqual(out, 'done\n') self.assert_exists(".coverage") self.assertEqual(self.number_of_data_files(), 1) # Read the coverage file and see that b_or_c.py has all 7 lines # executed. data = coverage.CoverageData() data.read_file(".coverage") self.assertEqual(data.line_counts()['b_or_c.py'], 7) def test_append_data_with_different_file(self): self.make_b_or_c_py() self.make_file(".coveragerc", """\ [run] data_file = .mycovdata """) out = self.run_command("coverage run b_or_c.py b") self.assertEqual(out, 'done\n') self.assert_doesnt_exist(".coverage") self.assert_exists(".mycovdata") out = self.run_command("coverage run --append b_or_c.py c") self.assertEqual(out, 'done\n') self.assert_doesnt_exist(".coverage") self.assert_exists(".mycovdata") # Read the coverage file and see that b_or_c.py has all 7 lines # executed. data = coverage.CoverageData() data.read_file(".mycovdata") self.assertEqual(data.line_counts()['b_or_c.py'], 7) def test_append_can_create_a_data_file(self): self.make_b_or_c_py() out = self.run_command("coverage run --append b_or_c.py b") self.assertEqual(out, 'done\n') self.assert_exists(".coverage") self.assertEqual(self.number_of_data_files(), 1) # Read the coverage file and see that b_or_c.py has only 6 lines # executed. data = coverage.CoverageData() data.read_file(".coverage") self.assertEqual(data.line_counts()['b_or_c.py'], 6) def test_combine_with_rc(self): self.make_b_or_c_py() self.make_file(".coveragerc", """\ [run] parallel = true """) out = self.run_command("coverage run b_or_c.py b") self.assertEqual(out, 'done\n') self.assert_doesnt_exist(".coverage") out = self.run_command("coverage run b_or_c.py c") self.assertEqual(out, 'done\n') self.assert_doesnt_exist(".coverage") # After two runs, there should be two .coverage.machine.123 files. self.assertEqual(self.number_of_data_files(), 2) # Combine the parallel coverage data files into .coverage . self.run_command("coverage combine") self.assert_exists(".coverage") self.assert_exists(".coveragerc") # After combining, there should be only the .coverage file. self.assertEqual(self.number_of_data_files(), 1) # Read the coverage file and see that b_or_c.py has all 7 lines # executed. data = coverage.CoverageData() data.read_file(".coverage") self.assertEqual(data.line_counts()['b_or_c.py'], 7) # Reporting should still work even with the .rc file out = self.run_command("coverage report") self.assertMultiLineEqual(out, textwrap.dedent("""\ Name Stmts Miss Cover ------------------------------- b_or_c.py 7 0 100% """)) def test_combine_with_aliases(self): self.make_file("d1/x.py", """\ a = 1 b = 2 print("%s %s" % (a, b)) """) self.make_file("d2/x.py", """\ # 1 # 2 # 3 c = 4 d = 5 print("%s %s" % (c, d)) """) self.make_file(".coveragerc", """\ [run] parallel = True [paths] source = src */d1 */d2 """) out = self.run_command("coverage run " + os.path.normpath("d1/x.py")) self.assertEqual(out, '1 2\n') out = self.run_command("coverage run " + os.path.normpath("d2/x.py")) self.assertEqual(out, '4 5\n') self.assertEqual(self.number_of_data_files(), 2) self.run_command("coverage combine") self.assert_exists(".coverage") # After combining, there should be only the .coverage file. self.assertEqual(self.number_of_data_files(), 1) # Read the coverage data file and see that the two different x.py # files have been combined together. data = coverage.CoverageData() data.read_file(".coverage") summary = data.line_counts(fullpath=True) self.assertEqual(len(summary), 1) actual = os.path.normcase(os.path.abspath(list(summary.keys())[0])) expected = os.path.normcase(os.path.abspath('src/x.py')) self.assertEqual(actual, expected) self.assertEqual(list(summary.values())[0], 6) def test_erase_parallel(self): self.make_file(".coveragerc", """\ [run] data_file = data.dat parallel = True """) self.make_file("data.dat") self.make_file("data.dat.fooey") self.make_file("data.dat.gooey") self.make_file(".coverage") self.run_command("coverage erase") self.assert_doesnt_exist("data.dat") self.assert_doesnt_exist("data.dat.fooey") self.assert_doesnt_exist("data.dat.gooey") self.assert_exists(".coverage") def test_missing_source_file(self): # Check what happens if the source is missing when reporting happens. self.make_file("fleeting.py", """\ s = 'goodbye, cruel world!' """) self.run_command("coverage run fleeting.py") os.remove("fleeting.py") out = self.run_command("coverage html -d htmlcov") self.assertRegex(out, "No source for code: '.*fleeting.py'") self.assertNotIn("Traceback", out) # It happens that the code paths are different for *.py and other # files, so try again with no extension. self.make_file("fleeting", """\ s = 'goodbye, cruel world!' """) self.run_command("coverage run fleeting") os.remove("fleeting") status, out = self.run_command_status("coverage html -d htmlcov") self.assertRegex(out, "No source for code: '.*fleeting'") self.assertNotIn("Traceback", out) self.assertEqual(status, 1) def test_running_missing_file(self): status, out = self.run_command_status("coverage run xyzzy.py") self.assertRegex(out, "No file to run: .*xyzzy.py") self.assertNotIn("raceback", out) self.assertNotIn("rror", out) self.assertEqual(status, 1) def test_code_throws(self): self.make_file("throw.py", """\ def f1(): raise Exception("hey!") def f2(): f1() f2() """) # The important thing is for "coverage run" and "python" to report the # same traceback. status, out = self.run_command_status("coverage run throw.py") out2 = self.run_command("python throw.py") if env.PYPY: # Pypy has an extra frame in the traceback for some reason lines2 = out2.splitlines() out2 = "".join(l+"\n" for l in lines2 if "toplevel" not in l) self.assertMultiLineEqual(out, out2) # But also make sure that the output is what we expect. self.assertIn('File "throw.py", line 5, in f2', out) self.assertIn('raise Exception("hey!")', out) self.assertNotIn('coverage', out) self.assertEqual(status, 1) def test_code_exits(self): self.make_file("exit.py", """\ import sys def f1(): print("about to exit..") sys.exit(17) def f2(): f1() f2() """) # The important thing is for "coverage run" and "python" to have the # same output. No traceback. status, out = self.run_command_status("coverage run exit.py") status2, out2 = self.run_command_status("python exit.py") self.assertMultiLineEqual(out, out2) self.assertMultiLineEqual(out, "about to exit..\n") self.assertEqual(status, status2) self.assertEqual(status, 17) def test_code_exits_no_arg(self): self.make_file("exit_none.py", """\ import sys def f1(): print("about to exit quietly..") sys.exit() f1() """) status, out = self.run_command_status("coverage run exit_none.py") status2, out2 = self.run_command_status("python exit_none.py") self.assertMultiLineEqual(out, out2) self.assertMultiLineEqual(out, "about to exit quietly..\n") self.assertEqual(status, status2) self.assertEqual(status, 0) def test_coverage_run_is_like_python(self): tryfile = os.path.join(HERE, "try_execfile.py") with open(tryfile) as f: self.make_file("run_me.py", f.read()) out_cov = self.run_command("coverage run run_me.py") out_py = self.run_command("python run_me.py") self.assertMultiLineEqual(out_cov, out_py) def test_coverage_run_dashm_is_like_python_dashm(self): # These -m commands assume the coverage tree is on the path. out_cov = self.run_command("coverage run -m tests.try_execfile") out_py = self.run_command("python -m tests.try_execfile") self.assertMultiLineEqual(out_cov, out_py) def test_coverage_run_dir_is_like_python_dir(self): tryfile = os.path.join(HERE, "try_execfile.py") with open(tryfile) as f: self.make_file("with_main/__main__.py", f.read()) out_cov = self.run_command("coverage run with_main") out_py = self.run_command("python with_main") # The coverage.py results are not identical to the Python results, and # I don't know why. For now, ignore those failures. If someone finds # a real problem with the discrepancies, we can work on it some more. ignored = r"__file__|__loader__|__package__" # PyPy includes the current directory in the path when running a # directory, while CPython and coverage.py do not. Exclude that from # the comparison also... if env.PYPY: ignored += "|"+re.escape(os.getcwd()) out_cov = remove_matching_lines(out_cov, ignored) out_py = remove_matching_lines(out_py, ignored) self.assertMultiLineEqual(out_cov, out_py) def test_coverage_run_dashm_equal_to_doubledashsource(self): """regression test for #328 When imported by -m, a module's __name__ is __main__, but we need the --source machinery to know and respect the original name. """ # These -m commands assume the coverage tree is on the path. out_cov = self.run_command( "coverage run --source tests.try_execfile -m tests.try_execfile" ) out_py = self.run_command("python -m tests.try_execfile") self.assertMultiLineEqual(out_cov, out_py) def test_coverage_run_dashm_superset_of_doubledashsource(self): """Edge case: --source foo -m foo.bar""" # These -m commands assume the coverage tree is on the path. out_cov = self.run_command( "coverage run --source tests -m tests.try_execfile" ) out_py = self.run_command("python -m tests.try_execfile") self.assertMultiLineEqual(out_cov, out_py) st, out = self.run_command_status("coverage report") self.assertEqual(st, 0) self.assertEqual(self.line_count(out), 6, out) def test_coverage_run_script_imports_doubledashsource(self): # This file imports try_execfile, which compiles it to .pyc, so the # first run will have __file__ == "try_execfile.py" and the second will # have __file__ == "try_execfile.pyc", which throws off the comparison. # Setting dont_write_bytecode True stops the compilation to .pyc and # keeps the test working. self.make_file("myscript", """\ import sys; sys.dont_write_bytecode = True import tests.try_execfile """) # These -m commands assume the coverage tree is on the path. out_cov = self.run_command( "coverage run --source tests myscript" ) out_py = self.run_command("python myscript") self.assertMultiLineEqual(out_cov, out_py) st, out = self.run_command_status("coverage report") self.assertEqual(st, 0) self.assertEqual(self.line_count(out), 6, out) def test_coverage_run_dashm_is_like_python_dashm_off_path(self): # https://bitbucket.org/ned/coveragepy/issue/242 tryfile = os.path.join(HERE, "try_execfile.py") self.make_file("sub/__init__.py", "") with open(tryfile) as f: self.make_file("sub/run_me.py", f.read()) out_cov = self.run_command("coverage run -m sub.run_me") out_py = self.run_command("python -m sub.run_me") self.assertMultiLineEqual(out_cov, out_py) def test_coverage_run_dashm_is_like_python_dashm_with__main__207(self): if sys.version_info < (2, 7): # Coverage.py isn't bug-for-bug compatible in the behavior of -m for # Pythons < 2.7 self.skip("-m doesn't work the same < Python 2.7") # https://bitbucket.org/ned/coveragepy/issue/207 self.make_file("package/__init__.py", "print('init')") self.make_file("package/__main__.py", "print('main')") out_cov = self.run_command("coverage run -m package") out_py = self.run_command("python -m package") self.assertMultiLineEqual(out_cov, out_py) def test_fork(self): if not hasattr(os, 'fork'): self.skip("Can't test os.fork since it doesn't exist.") self.make_file("fork.py", """\ import os def child(): print('Child!') def main(): ret = os.fork() if ret == 0: child() else: os.waitpid(ret, 0) main() """) out = self.run_command("coverage run -p fork.py") self.assertEqual(out, 'Child!\n') self.assert_doesnt_exist(".coverage") # After running the forking program, there should be two # .coverage.machine.123 files. self.assertEqual(self.number_of_data_files(), 2) # Combine the parallel coverage data files into .coverage . self.run_command("coverage combine") self.assert_exists(".coverage") # After combining, there should be only the .coverage file. self.assertEqual(self.number_of_data_files(), 1) # Read the coverage file and see that b_or_c.py has all 7 lines # executed. data = coverage.CoverageData() data.read_file(".coverage") self.assertEqual(data.line_counts()['fork.py'], 9) def test_warnings(self): self.make_file("hello.py", """\ import sys, os print("Hello") """) out = self.run_command("coverage run --source=sys,xyzzy,quux hello.py") self.assertIn("Hello\n", out) self.assertIn(textwrap.dedent("""\ Coverage.py warning: Module sys has no Python source. Coverage.py warning: Module xyzzy was never imported. Coverage.py warning: Module quux was never imported. Coverage.py warning: No data was collected. """), out) def test_warnings_during_reporting(self): # While fixing issue #224, the warnings were being printed far too # often. Make sure they're not any more. self.make_file("hello.py", """\ import sys, os, the_other print("Hello") """) self.make_file("the_other.py", """\ print("What?") """) self.make_file(".coveragerc", """\ [run] source = . xyzzy """) self.run_command("coverage run hello.py") out = self.run_command("coverage html") self.assertEqual(out.count("Module xyzzy was never imported."), 0) def test_warnings_if_never_run(self): out = self.run_command("coverage run i_dont_exist.py") self.assertIn("No file to run: 'i_dont_exist.py'", out) self.assertNotIn("warning", out) self.assertNotIn("Exception", out) out = self.run_command("coverage run -m no_such_module") self.assertTrue( ("No module named no_such_module" in out) or ("No module named 'no_such_module'" in out) ) self.assertNotIn("warning", out) self.assertNotIn("Exception", out) def test_warnings_trace_function_changed_with_threads(self): # https://bitbucket.org/ned/coveragepy/issue/164 self.make_file("bug164.py", """\ import threading import time class MyThread (threading.Thread): def run(self): print("Hello") thr = MyThread() thr.start() thr.join() """) out = self.run_command("coverage run --timid bug164.py") self.assertIn("Hello\n", out) self.assertNotIn("warning", out) def test_warning_trace_function_changed(self): self.make_file("settrace.py", """\ import sys print("Hello") sys.settrace(None) print("Goodbye") """) out = self.run_command("coverage run --timid settrace.py") self.assertIn("Hello\n", out) self.assertIn("Goodbye\n", out) self.assertIn("Trace function changed", out) def test_note(self): self.make_file(".coveragerc", """\ [run] data_file = mydata.dat note = These are musical notes: ♫𝅗𝅥♩ """) self.make_file("simple.py", """print('hello')""") self.run_command("coverage run simple.py") data = CoverageData() data.read_file("mydata.dat") infos = data.run_infos() self.assertEqual(len(infos), 1) self.assertEqual(infos[0]['note'], u"These are musical notes: ♫𝅗𝅥♩") def test_fullcoverage(self): # pragma: not covered if env.PY2: # This doesn't work on Python 2. self.skip("fullcoverage doesn't work on Python 2.") # It only works with the C tracer, and if we aren't measuring ourselves. if not env.C_TRACER or env.METACOV: self.skip("fullcoverage only works with the C tracer.") # fullcoverage is a trick to get stdlib modules measured from # the very beginning of the process. Here we import os and # then check how many lines are measured. self.make_file("getenv.py", """\ import os print("FOOEY == %s" % os.getenv("FOOEY")) """) fullcov = os.path.join( os.path.dirname(coverage.__file__), "fullcoverage" ) self.set_environ("FOOEY", "BOO") self.set_environ("PYTHONPATH", fullcov) out = self.run_command("python -m coverage run -L getenv.py") self.assertEqual(out, "FOOEY == BOO\n") data = coverage.CoverageData() data.read_file(".coverage") # The actual number of executed lines in os.py when it's # imported is 120 or so. Just running os.getenv executes # about 5. self.assertGreater(data.line_counts()['os.py'], 50) def test_deprecation_warnings(self): # Test that coverage doesn't trigger deprecation warnings. # https://bitbucket.org/ned/coveragepy/issue/305/pendingdeprecationwarning-the-imp-module self.make_file("allok.py", """\ import warnings warnings.simplefilter('default') import coverage print("No warnings!") """) out = self.run_command("python allok.py") self.assertEqual(out, "No warnings!\n") def test_run_twice(self): # https://bitbucket.org/ned/coveragepy/issue/353/40a3-introduces-an-unexpected-third-case self.make_file("foo.py", """\ def foo(): pass """) self.make_file("run_twice.py", """\ import coverage for _ in [1, 2]: inst = coverage.Coverage(source=['foo']) inst.load() inst.start() import foo inst.stop() inst.combine() inst.save() """) out = self.run_command("python run_twice.py") self.assertEqual( out, "Coverage.py warning: Module foo was previously imported, but not measured.\n" ) class AliasedCommandTest(CoverageTest): """Tests of the version-specific command aliases.""" run_in_temp_dir = False def test_major_version_works(self): # "coverage2" works on py2 cmd = "coverage%d" % sys.version_info[0] out = self.run_command(cmd) self.assertIn("Code coverage for Python", out) def test_wrong_alias_doesnt_work(self): # "coverage3" doesn't work on py2 badcmd = "coverage%d" % (5 - sys.version_info[0]) out = self.run_command(badcmd) self.assertNotIn("Code coverage for Python", out) def test_specific_alias_works(self): # "coverage-2.7" works on py2.7 cmd = "coverage-%d.%d" % sys.version_info[:2] out = self.run_command(cmd) self.assertIn("Code coverage for Python", out) class PydocTest(CoverageTest): """Test that pydoc can get our information.""" run_in_temp_dir = False def assert_pydoc_ok(self, name, thing): """Check that pydoc of `name` finds the docstring from `thing`.""" # Run pydoc. out = self.run_command("python -m pydoc " + name) # It should say "Help on..", and not have a traceback self.assert_starts_with(out, "Help on ") self.assertNotIn("Traceback", out) # All of the lines in the docstring should be there somewhere. for line in thing.__doc__.splitlines(): self.assertIn(line.strip(), out) def test_pydoc_coverage(self): self.assert_pydoc_ok("coverage", coverage) def test_pydoc_coverage_coverage(self): self.assert_pydoc_ok("coverage.Coverage", coverage.Coverage) class FailUnderTest(CoverageTest): """Tests of the --fail-under switch.""" def setUp(self): super(FailUnderTest, self).setUp() self.make_file("forty_two_plus.py", """\ # I have 42.857% (3/7) coverage! a = 1 b = 2 if a > 3: b = 4 c = 5 d = 6 e = 7 """) st, _ = self.run_command_status("coverage run forty_two_plus.py") self.assertEqual(st, 0) st, out = self.run_command_status("coverage report") self.assertEqual(st, 0) self.assertEqual( self.last_line_squeezed(out), "forty_two_plus.py 7 4 43%" ) def test_report(self): st, _ = self.run_command_status("coverage report --fail-under=42") self.assertEqual(st, 0) st, _ = self.run_command_status("coverage report --fail-under=43") self.assertEqual(st, 0) st, _ = self.run_command_status("coverage report --fail-under=44") self.assertEqual(st, 2) def test_html_report(self): st, _ = self.run_command_status("coverage html --fail-under=42") self.assertEqual(st, 0) st, _ = self.run_command_status("coverage html --fail-under=43") self.assertEqual(st, 0) st, _ = self.run_command_status("coverage html --fail-under=44") self.assertEqual(st, 2) def test_xml_report(self): st, _ = self.run_command_status("coverage xml --fail-under=42") self.assertEqual(st, 0) st, _ = self.run_command_status("coverage xml --fail-under=43") self.assertEqual(st, 0) st, _ = self.run_command_status("coverage xml --fail-under=44") self.assertEqual(st, 2) def test_fail_under_in_config(self): self.make_file(".coveragerc", "[report]\nfail_under = 43\n") st, _ = self.run_command_status("coverage report") self.assertEqual(st, 0) self.make_file(".coveragerc", "[report]\nfail_under = 44\n") st, _ = self.run_command_status("coverage report") self.assertEqual(st, 2) class FailUnderNoFilesTest(CoverageTest): """Test that nothing to report results in an error exit status.""" def setUp(self): super(FailUnderNoFilesTest, self).setUp() self.make_file(".coveragerc", "[report]\nfail_under = 99\n") def test_report(self): st, out = self.run_command_status("coverage report") self.assertIn('No data to report.', out) self.assertEqual(st, 1) def test_xml(self): st, out = self.run_command_status("coverage xml") self.assertIn('No data to report.', out) self.assertEqual(st, 1) def test_html(self): st, out = self.run_command_status("coverage html") self.assertIn('No data to report.', out) self.assertEqual(st, 1) class FailUnderEmptyFilesTest(CoverageTest): """Test that empty files produce the proper fail_under exit status.""" def setUp(self): super(FailUnderEmptyFilesTest, self).setUp() self.make_file(".coveragerc", "[report]\nfail_under = 99\n") self.make_file("empty.py", "") st, _ = self.run_command_status("coverage run empty.py") self.assertEqual(st, 0) def test_report(self): st, _ = self.run_command_status("coverage report") self.assertEqual(st, 2) def test_xml(self): st, _ = self.run_command_status("coverage xml") self.assertEqual(st, 2) def test_html(self): st, _ = self.run_command_status("coverage html") self.assertEqual(st, 2) def possible_pth_dirs(): """Produce a sequence of directories for trying to write .pth files.""" # First look through sys.path, and we find a .pth file, then it's a good # place to put ours. for d in sys.path: g = glob.glob(os.path.join(d, "*.pth")) if g: yield d # If we're still looking, then try the Python library directory. # https://bitbucket.org/ned/coveragepy/issue/339/pth-test-malfunctions import distutils.sysconfig # pylint: disable=import-error yield distutils.sysconfig.get_python_lib() class ProcessCoverageMixin(object): """Set up a .pth file to coverage-measure all sub-processes.""" def setUp(self): super(ProcessCoverageMixin, self).setUp() # Find a place to put a .pth file. pth_contents = "import coverage; coverage.process_startup()\n" for pth_dir in possible_pth_dirs(): # pragma: part covered pth_path = os.path.join(pth_dir, "subcover.pth") with open(pth_path, "w") as pth: try: pth.write(pth_contents) self.pth_path = pth_path break except (IOError, OSError): # pragma: not covered pass else: # pragma: not covered raise Exception("Couldn't find a place for the .pth file") self.addCleanup(os.remove, self.pth_path) class ProcessStartupTest(ProcessCoverageMixin, CoverageTest): """Test that we can measure coverage in sub-processes.""" def test_subprocess_with_pth_files(self): # pragma: not covered if env.METACOV: self.skip( "Can't test sub-process pth file suppport during metacoverage" ) # Main will run sub.py self.make_file("main.py", """\ import os, os.path, sys ex = os.path.basename(sys.executable) os.system(ex + " sub.py") """) # sub.py will write a few lines. self.make_file("sub.py", """\ with open("out.txt", "w") as f: f.write("Hello, world!\\n") """) self.make_file("coverage.ini", """\ [run] data_file = .mycovdata """) self.set_environ("COVERAGE_PROCESS_START", "coverage.ini") import main # pylint: disable=import-error,unused-variable with open("out.txt") as f: self.assertEqual(f.read(), "Hello, world!\n") # Read the data from .coverage self.assert_exists(".mycovdata") data = coverage.CoverageData() data.read_file(".mycovdata") self.assertEqual(data.line_counts()['sub.py'], 2) class ProcessStartupWithSourceTest(ProcessCoverageMixin, CoverageTest): """Show that we can configure {[run]source} during process-level coverage. There are three interesting variables, for a total of eight tests: 1. -m versus a simple script argument (for example, `python myscript`), 2. filtering for the top-level (main.py) or second-level (sub.py) module, and 3. whether the files are in a package or not. """ def assert_pth_and_source_work_together( self, dashm, package, source ): # pragma: not covered """Run the test for a particular combination of factors. The arguments are all strings: * `dashm`: Either "" (run the program as a file) or "-m" (run the program as a module). * `package`: Either "" (put the source at the top level) or a package name to use to hold the source. * `source`: Either "main" or "sub", which file to use as the ``--source`` argument. """ if env.METACOV: self.skip( "Can't test sub-process pth file suppport during metacoverage" ) def fullname(modname): """What is the full module name for `modname` for this test?""" if package and dashm: return '.'.join((package, modname)) else: return modname def path(basename): """Where should `basename` be created for this test?""" return os.path.join(package, basename) # Main will run sub.py. self.make_file(path("main.py"), """\ import %s if True: pass """ % fullname('sub')) if package: self.make_file(path("__init__.py"), "") # sub.py will write a few lines. self.make_file(path("sub.py"), """\ with open("out.txt", "w") as f: f.write("Hello, world!") """) self.make_file("coverage.ini", """\ [run] source = %s """ % fullname(source)) self.set_environ("COVERAGE_PROCESS_START", "coverage.ini") if dashm: cmd = "python -m %s" % fullname('main') else: cmd = "python %s" % path('main.py') self.run_command(cmd) with open("out.txt") as f: self.assertEqual(f.read(), "Hello, world!") # Read the data from .coverage self.assert_exists(".coverage") data = coverage.CoverageData() data.read_file(".coverage") summary = data.line_counts() print(summary) self.assertEqual(summary[source + '.py'], 2) self.assertEqual(len(summary), 1) def test_dashm_main(self): self.assert_pth_and_source_work_together('-m', '', 'main') def test_script_main(self): self.assert_pth_and_source_work_together('', '', 'main') def test_dashm_sub(self): self.assert_pth_and_source_work_together('-m', '', 'sub') def test_script_sub(self): self.assert_pth_and_source_work_together('', '', 'sub') def test_dashm_pkg_main(self): self.assert_pth_and_source_work_together('-m', 'pkg', 'main') def test_script_pkg_main(self): self.assert_pth_and_source_work_together('', 'pkg', 'main') def test_dashm_pkg_sub(self): self.assert_pth_and_source_work_together('-m', 'pkg', 'sub') def test_script_pkg_sub(self): self.assert_pth_and_source_work_together('', 'pkg', 'sub') def remove_matching_lines(text, pat): """Return `text` with all lines matching `pat` removed.""" lines = [l for l in text.splitlines(True) if not re.search(pat, l)] return "".join(lines)