| """Testing `tabnanny` module. |
| |
| Glossary: |
| * errored : Whitespace related problems present in file. |
| """ |
| from unittest import TestCase, mock |
| from unittest import mock |
| import errno |
| import os |
| import tabnanny |
| import tokenize |
| import tempfile |
| import textwrap |
| from test.support import (captured_stderr, captured_stdout, script_helper, |
| findfile, unlink) |
| |
| |
| SOURCE_CODES = { |
| "incomplete_expression": ( |
| 'fruits = [\n' |
| ' "Apple",\n' |
| ' "Orange",\n' |
| ' "Banana",\n' |
| '\n' |
| 'print(fruits)\n' |
| ), |
| "wrong_indented": ( |
| 'if True:\n' |
| ' print("hello")\n' |
| ' print("world")\n' |
| 'else:\n' |
| ' print("else called")\n' |
| ), |
| "nannynag_errored": ( |
| 'if True:\n' |
| ' \tprint("hello")\n' |
| '\tprint("world")\n' |
| 'else:\n' |
| ' print("else called")\n' |
| ), |
| "error_free": ( |
| 'if True:\n' |
| ' print("hello")\n' |
| ' print("world")\n' |
| 'else:\n' |
| ' print("else called")\n' |
| ), |
| "tab_space_errored_1": ( |
| 'def my_func():\n' |
| '\t print("hello world")\n' |
| '\t if True:\n' |
| '\t\tprint("If called")' |
| ), |
| "tab_space_errored_2": ( |
| 'def my_func():\n' |
| '\t\tprint("Hello world")\n' |
| '\t\tif True:\n' |
| '\t print("If called")' |
| ) |
| } |
| |
| |
| class TemporaryPyFile: |
| """Create a temporary python source code file.""" |
| |
| def __init__(self, source_code='', directory=None): |
| self.source_code = source_code |
| self.dir = directory |
| |
| def __enter__(self): |
| with tempfile.NamedTemporaryFile( |
| mode='w', dir=self.dir, suffix=".py", delete=False |
| ) as f: |
| f.write(self.source_code) |
| self.file_path = f.name |
| return self.file_path |
| |
| def __exit__(self, exc_type, exc_value, exc_traceback): |
| unlink(self.file_path) |
| |
| |
| class TestFormatWitnesses(TestCase): |
| """Testing `tabnanny.format_witnesses()`.""" |
| |
| def test_format_witnesses(self): |
| """Asserting formatter result by giving various input samples.""" |
| tests = [ |
| ('Test', 'at tab sizes T, e, s, t'), |
| ('', 'at tab size '), |
| ('t', 'at tab size t'), |
| (' t ', 'at tab sizes , , t, , '), |
| ] |
| |
| for words, expected in tests: |
| with self.subTest(words=words, expected=expected): |
| self.assertEqual(tabnanny.format_witnesses(words), expected) |
| |
| |
| class TestErrPrint(TestCase): |
| """Testing `tabnanny.errprint()`.""" |
| |
| def test_errprint(self): |
| """Asserting result of `tabnanny.errprint()` by giving sample inputs.""" |
| tests = [ |
| (['first', 'second'], 'first second\n'), |
| (['first'], 'first\n'), |
| ([1, 2, 3], '1 2 3\n'), |
| ([], '\n') |
| ] |
| |
| for args, expected in tests: |
| with self.subTest(arguments=args, expected=expected): |
| with captured_stderr() as stderr: |
| tabnanny.errprint(*args) |
| self.assertEqual(stderr.getvalue() , expected) |
| |
| |
| class TestNannyNag(TestCase): |
| def test_all_methods(self): |
| """Asserting behaviour of `tabnanny.NannyNag` exception.""" |
| tests = [ |
| ( |
| tabnanny.NannyNag(0, "foo", "bar"), |
| {'lineno': 0, 'msg': 'foo', 'line': 'bar'} |
| ), |
| ( |
| tabnanny.NannyNag(5, "testmsg", "testline"), |
| {'lineno': 5, 'msg': 'testmsg', 'line': 'testline'} |
| ) |
| ] |
| for nanny, expected in tests: |
| line_number = nanny.get_lineno() |
| msg = nanny.get_msg() |
| line = nanny.get_line() |
| with self.subTest( |
| line_number=line_number, expected=expected['lineno'] |
| ): |
| self.assertEqual(expected['lineno'], line_number) |
| with self.subTest(msg=msg, expected=expected['msg']): |
| self.assertEqual(expected['msg'], msg) |
| with self.subTest(line=line, expected=expected['line']): |
| self.assertEqual(expected['line'], line) |
| |
| |
| class TestCheck(TestCase): |
| """Testing tabnanny.check().""" |
| |
| def setUp(self): |
| self.addCleanup(setattr, tabnanny, 'verbose', tabnanny.verbose) |
| tabnanny.verbose = 0 # Forcefully deactivating verbose mode. |
| |
| def verify_tabnanny_check(self, dir_or_file, out="", err=""): |
| """Common verification for tabnanny.check(). |
| |
| Use this method to assert expected values of `stdout` and `stderr` after |
| running tabnanny.check() on given `dir` or `file` path. Because |
| tabnanny.check() captures exceptions and writes to `stdout` and |
| `stderr`, asserting standard outputs is the only way. |
| """ |
| with captured_stdout() as stdout, captured_stderr() as stderr: |
| tabnanny.check(dir_or_file) |
| self.assertEqual(stdout.getvalue(), out) |
| self.assertEqual(stderr.getvalue(), err) |
| |
| def test_correct_file(self): |
| """A python source code file without any errors.""" |
| with TemporaryPyFile(SOURCE_CODES["error_free"]) as file_path: |
| self.verify_tabnanny_check(file_path) |
| |
| def test_correct_directory_verbose(self): |
| """Directory containing few error free python source code files. |
| |
| Because order of files returned by `os.lsdir()` is not fixed, verify the |
| existence of each output lines at `stdout` using `in` operator. |
| `verbose` mode of `tabnanny.verbose` asserts `stdout`. |
| """ |
| with tempfile.TemporaryDirectory() as tmp_dir: |
| lines = [f"{tmp_dir!r}: listing directory\n",] |
| file1 = TemporaryPyFile(SOURCE_CODES["error_free"], directory=tmp_dir) |
| file2 = TemporaryPyFile(SOURCE_CODES["error_free"], directory=tmp_dir) |
| with file1 as file1_path, file2 as file2_path: |
| for file_path in (file1_path, file2_path): |
| lines.append(f"{file_path!r}: Clean bill of health.\n") |
| |
| tabnanny.verbose = 1 |
| with captured_stdout() as stdout, captured_stderr() as stderr: |
| tabnanny.check(tmp_dir) |
| stdout = stdout.getvalue() |
| for line in lines: |
| with self.subTest(line=line): |
| self.assertIn(line, stdout) |
| self.assertEqual(stderr.getvalue(), "") |
| |
| def test_correct_directory(self): |
| """Directory which contains few error free python source code files.""" |
| with tempfile.TemporaryDirectory() as tmp_dir: |
| with TemporaryPyFile(SOURCE_CODES["error_free"], directory=tmp_dir): |
| self.verify_tabnanny_check(tmp_dir) |
| |
| def test_when_wrong_indented(self): |
| """A python source code file eligible for raising `IndentationError`.""" |
| with TemporaryPyFile(SOURCE_CODES["wrong_indented"]) as file_path: |
| err = ('unindent does not match any outer indentation level' |
| ' (<tokenize>, line 3)\n') |
| err = f"{file_path!r}: Indentation Error: {err}" |
| self.verify_tabnanny_check(file_path, err=err) |
| |
| def test_when_tokenize_tokenerror(self): |
| """A python source code file eligible for raising 'tokenize.TokenError'.""" |
| with TemporaryPyFile(SOURCE_CODES["incomplete_expression"]) as file_path: |
| err = "('EOF in multi-line statement', (7, 0))\n" |
| err = f"{file_path!r}: Token Error: {err}" |
| self.verify_tabnanny_check(file_path, err=err) |
| |
| def test_when_nannynag_error_verbose(self): |
| """A python source code file eligible for raising `tabnanny.NannyNag`. |
| |
| Tests will assert `stdout` after activating `tabnanny.verbose` mode. |
| """ |
| with TemporaryPyFile(SOURCE_CODES["nannynag_errored"]) as file_path: |
| out = f"{file_path!r}: *** Line 3: trouble in tab city! ***\n" |
| out += "offending line: '\\tprint(\"world\")\\n'\n" |
| out += "indent not equal e.g. at tab size 1\n" |
| |
| tabnanny.verbose = 1 |
| self.verify_tabnanny_check(file_path, out=out) |
| |
| def test_when_nannynag_error(self): |
| """A python source code file eligible for raising `tabnanny.NannyNag`.""" |
| with TemporaryPyFile(SOURCE_CODES["nannynag_errored"]) as file_path: |
| out = f"{file_path} 3 '\\tprint(\"world\")\\n'\n" |
| self.verify_tabnanny_check(file_path, out=out) |
| |
| def test_when_no_file(self): |
| """A python file which does not exist actually in system.""" |
| path = 'no_file.py' |
| err = (f"{path!r}: I/O Error: [Errno {errno.ENOENT}] " |
| f"{os.strerror(errno.ENOENT)}: {path!r}\n") |
| self.verify_tabnanny_check(path, err=err) |
| |
| def test_errored_directory(self): |
| """Directory containing wrongly indented python source code files.""" |
| with tempfile.TemporaryDirectory() as tmp_dir: |
| error_file = TemporaryPyFile( |
| SOURCE_CODES["wrong_indented"], directory=tmp_dir |
| ) |
| code_file = TemporaryPyFile( |
| SOURCE_CODES["error_free"], directory=tmp_dir |
| ) |
| with error_file as e_file, code_file as c_file: |
| err = ('unindent does not match any outer indentation level' |
| ' (<tokenize>, line 3)\n') |
| err = f"{e_file!r}: Indentation Error: {err}" |
| self.verify_tabnanny_check(tmp_dir, err=err) |
| |
| |
| class TestProcessTokens(TestCase): |
| """Testing `tabnanny.process_tokens()`.""" |
| |
| @mock.patch('tabnanny.NannyNag') |
| def test_with_correct_code(self, MockNannyNag): |
| """A python source code without any whitespace related problems.""" |
| |
| with TemporaryPyFile(SOURCE_CODES["error_free"]) as file_path: |
| with open(file_path) as f: |
| tabnanny.process_tokens(tokenize.generate_tokens(f.readline)) |
| self.assertFalse(MockNannyNag.called) |
| |
| def test_with_errored_codes_samples(self): |
| """A python source code with whitespace related sampled problems.""" |
| |
| # "tab_space_errored_1": executes block under type == tokenize.INDENT |
| # at `tabnanny.process_tokens()`. |
| # "tab space_errored_2": executes block under |
| # `check_equal and type not in JUNK` condition at |
| # `tabnanny.process_tokens()`. |
| |
| for key in ["tab_space_errored_1", "tab_space_errored_2"]: |
| with self.subTest(key=key): |
| with TemporaryPyFile(SOURCE_CODES[key]) as file_path: |
| with open(file_path) as f: |
| tokens = tokenize.generate_tokens(f.readline) |
| with self.assertRaises(tabnanny.NannyNag): |
| tabnanny.process_tokens(tokens) |
| |
| |
| class TestCommandLine(TestCase): |
| """Tests command line interface of `tabnanny`.""" |
| |
| def validate_cmd(self, *args, stdout="", stderr="", partial=False): |
| """Common function to assert the behaviour of command line interface.""" |
| _, out, err = script_helper.assert_python_ok('-m', 'tabnanny', *args) |
| # Note: The `splitlines()` will solve the problem of CRLF(\r) added |
| # by OS Windows. |
| out = out.decode('ascii') |
| err = err.decode('ascii') |
| if partial: |
| for std, output in ((stdout, out), (stderr, err)): |
| _output = output.splitlines() |
| for _std in std.splitlines(): |
| with self.subTest(std=_std, output=_output): |
| self.assertIn(_std, _output) |
| else: |
| self.assertListEqual(out.splitlines(), stdout.splitlines()) |
| self.assertListEqual(err.splitlines(), stderr.splitlines()) |
| |
| def test_with_errored_file(self): |
| """Should displays error when errored python file is given.""" |
| with TemporaryPyFile(SOURCE_CODES["wrong_indented"]) as file_path: |
| stderr = f"{file_path!r}: Indentation Error: " |
| stderr += ('unindent does not match any outer indentation level' |
| ' (<tokenize>, line 3)') |
| self.validate_cmd(file_path, stderr=stderr) |
| |
| def test_with_error_free_file(self): |
| """Should not display anything if python file is correctly indented.""" |
| with TemporaryPyFile(SOURCE_CODES["error_free"]) as file_path: |
| self.validate_cmd(file_path) |
| |
| def test_command_usage(self): |
| """Should display usage on no arguments.""" |
| path = findfile('tabnanny.py') |
| stderr = f"Usage: {path} [-v] file_or_directory ..." |
| self.validate_cmd(stderr=stderr) |
| |
| def test_quiet_flag(self): |
| """Should display less when quite mode is on.""" |
| with TemporaryPyFile(SOURCE_CODES["nannynag_errored"]) as file_path: |
| stdout = f"{file_path}\n" |
| self.validate_cmd("-q", file_path, stdout=stdout) |
| |
| def test_verbose_mode(self): |
| """Should display more error information if verbose mode is on.""" |
| with TemporaryPyFile(SOURCE_CODES["nannynag_errored"]) as path: |
| stdout = textwrap.dedent( |
| "offending line: '\\tprint(\"world\")\\n'" |
| ).strip() |
| self.validate_cmd("-v", path, stdout=stdout, partial=True) |
| |
| def test_double_verbose_mode(self): |
| """Should display detailed error information if double verbose is on.""" |
| with TemporaryPyFile(SOURCE_CODES["nannynag_errored"]) as path: |
| stdout = textwrap.dedent( |
| "offending line: '\\tprint(\"world\")\\n'" |
| ).strip() |
| self.validate_cmd("-vv", path, stdout=stdout, partial=True) |