cpython/Lib/test/test_argparse.py

5362 lines
175 KiB
Python
Raw Normal View History

# Author: Steven J. Bethard <steven.bethard@gmail.com>.
import inspect
import os
import shutil
import stat
import sys
import textwrap
import tempfile
import unittest
import argparse
from io import StringIO
from test import support
from unittest import mock
Merged revisions 79464,79471,79623,79626,79630,79632,79643,79648-79649,79679,79685,79711,79761,79774,79777,79792-79794,79877,79898-79900 via svnmerge from svn+ssh://pythondev@svn.python.org/python/trunk ........ r79464 | michael.foord | 2010-03-27 07:55:19 -0500 (Sat, 27 Mar 2010) | 1 line A fix for running unittest tests on platforms without the audioop module (e.g. jython and IronPython) ........ r79471 | michael.foord | 2010-03-27 14:10:11 -0500 (Sat, 27 Mar 2010) | 4 lines Addition of delta keyword argument to unittest.TestCase.assertAlmostEquals and assertNotAlmostEquals This allows the comparison of objects by specifying a maximum difference; this includes the comparing of non-numeric objects that don't support rounding. ........ r79623 | michael.foord | 2010-04-02 16:42:47 -0500 (Fri, 02 Apr 2010) | 1 line Addition of -b command line option to unittest for buffering stdout and stderr during test runs. ........ r79626 | michael.foord | 2010-04-02 17:08:29 -0500 (Fri, 02 Apr 2010) | 1 line TestResult stores original sys.stdout and tests no longer use sys.__stdout__ (etc) in tests for unittest -b command line option ........ r79630 | michael.foord | 2010-04-02 17:30:56 -0500 (Fri, 02 Apr 2010) | 1 line unittest tests no longer replace the sys.stdout put in place by regrtest ........ r79632 | michael.foord | 2010-04-02 17:55:59 -0500 (Fri, 02 Apr 2010) | 1 line Issue #8038: Addition of unittest.TestCase.assertNotRegexpMatches ........ r79643 | michael.foord | 2010-04-02 20:15:21 -0500 (Fri, 02 Apr 2010) | 1 line Support dotted module names for test discovery paths in unittest. Issue 8038. ........ r79648 | michael.foord | 2010-04-02 21:21:39 -0500 (Fri, 02 Apr 2010) | 1 line Cross platform unittest.TestResult newline handling when buffering stdout / stderr. ........ r79649 | michael.foord | 2010-04-02 21:33:55 -0500 (Fri, 02 Apr 2010) | 1 line Another attempt at a fix for unittest.test.test_result for windows line endings ........ r79679 | michael.foord | 2010-04-03 09:52:18 -0500 (Sat, 03 Apr 2010) | 1 line Adding -b command line option to the unittest usage message. ........ r79685 | michael.foord | 2010-04-03 10:20:00 -0500 (Sat, 03 Apr 2010) | 1 line Minor tweak to unittest command line usage message ........ r79711 | michael.foord | 2010-04-03 12:03:11 -0500 (Sat, 03 Apr 2010) | 1 line Documenting new features in unittest ........ r79761 | michael.foord | 2010-04-04 17:41:54 -0500 (Sun, 04 Apr 2010) | 1 line unittest documentation formatting changes ........ r79774 | michael.foord | 2010-04-04 18:28:44 -0500 (Sun, 04 Apr 2010) | 1 line Adding documentation for new unittest.main() parameters ........ r79777 | michael.foord | 2010-04-04 19:39:50 -0500 (Sun, 04 Apr 2010) | 1 line Document signal handling functions in unittest.rst ........ r79792 | michael.foord | 2010-04-05 05:26:26 -0500 (Mon, 05 Apr 2010) | 1 line Documentation fixes for unittest ........ r79793 | michael.foord | 2010-04-05 05:28:27 -0500 (Mon, 05 Apr 2010) | 1 line Furterh documentation fix for unittest.rst ........ r79794 | michael.foord | 2010-04-05 05:30:14 -0500 (Mon, 05 Apr 2010) | 1 line Further documentation fix for unittest.rst ........ r79877 | michael.foord | 2010-04-06 18:18:16 -0500 (Tue, 06 Apr 2010) | 1 line Fix module directory finding logic for dotted paths in unittest test discovery. ........ r79898 | michael.foord | 2010-04-07 18:04:22 -0500 (Wed, 07 Apr 2010) | 1 line unittest.result.TestResult does not create its buffers until they're used. It uses StringIO not cStringIO. Issue 8333. ........ r79899 | michael.foord | 2010-04-07 19:04:24 -0500 (Wed, 07 Apr 2010) | 1 line Switch regrtest to use StringIO instead of cStringIO for test_multiprocessing on Windows. Issue 8333. ........ r79900 | michael.foord | 2010-04-07 23:33:20 -0500 (Wed, 07 Apr 2010) | 1 line Correction of unittest documentation typos and omissions ........
2010-04-11 17:43:16 -03:00
class StdIOBuffer(StringIO):
pass
class TestCase(unittest.TestCase):
def setUp(self):
# The tests assume that line wrapping occurs at 80 columns, but this
# behaviour can be overridden by setting the COLUMNS environment
# variable. To ensure that this width is used, set COLUMNS to 80.
env = support.EnvironmentVarGuard()
env['COLUMNS'] = '80'
self.addCleanup(env.__exit__)
Merged revisions 79464,79471,79623,79626,79630,79632,79643,79648-79649,79679,79685,79711,79761,79774,79777,79792-79794,79877,79898-79900 via svnmerge from svn+ssh://pythondev@svn.python.org/python/trunk ........ r79464 | michael.foord | 2010-03-27 07:55:19 -0500 (Sat, 27 Mar 2010) | 1 line A fix for running unittest tests on platforms without the audioop module (e.g. jython and IronPython) ........ r79471 | michael.foord | 2010-03-27 14:10:11 -0500 (Sat, 27 Mar 2010) | 4 lines Addition of delta keyword argument to unittest.TestCase.assertAlmostEquals and assertNotAlmostEquals This allows the comparison of objects by specifying a maximum difference; this includes the comparing of non-numeric objects that don't support rounding. ........ r79623 | michael.foord | 2010-04-02 16:42:47 -0500 (Fri, 02 Apr 2010) | 1 line Addition of -b command line option to unittest for buffering stdout and stderr during test runs. ........ r79626 | michael.foord | 2010-04-02 17:08:29 -0500 (Fri, 02 Apr 2010) | 1 line TestResult stores original sys.stdout and tests no longer use sys.__stdout__ (etc) in tests for unittest -b command line option ........ r79630 | michael.foord | 2010-04-02 17:30:56 -0500 (Fri, 02 Apr 2010) | 1 line unittest tests no longer replace the sys.stdout put in place by regrtest ........ r79632 | michael.foord | 2010-04-02 17:55:59 -0500 (Fri, 02 Apr 2010) | 1 line Issue #8038: Addition of unittest.TestCase.assertNotRegexpMatches ........ r79643 | michael.foord | 2010-04-02 20:15:21 -0500 (Fri, 02 Apr 2010) | 1 line Support dotted module names for test discovery paths in unittest. Issue 8038. ........ r79648 | michael.foord | 2010-04-02 21:21:39 -0500 (Fri, 02 Apr 2010) | 1 line Cross platform unittest.TestResult newline handling when buffering stdout / stderr. ........ r79649 | michael.foord | 2010-04-02 21:33:55 -0500 (Fri, 02 Apr 2010) | 1 line Another attempt at a fix for unittest.test.test_result for windows line endings ........ r79679 | michael.foord | 2010-04-03 09:52:18 -0500 (Sat, 03 Apr 2010) | 1 line Adding -b command line option to the unittest usage message. ........ r79685 | michael.foord | 2010-04-03 10:20:00 -0500 (Sat, 03 Apr 2010) | 1 line Minor tweak to unittest command line usage message ........ r79711 | michael.foord | 2010-04-03 12:03:11 -0500 (Sat, 03 Apr 2010) | 1 line Documenting new features in unittest ........ r79761 | michael.foord | 2010-04-04 17:41:54 -0500 (Sun, 04 Apr 2010) | 1 line unittest documentation formatting changes ........ r79774 | michael.foord | 2010-04-04 18:28:44 -0500 (Sun, 04 Apr 2010) | 1 line Adding documentation for new unittest.main() parameters ........ r79777 | michael.foord | 2010-04-04 19:39:50 -0500 (Sun, 04 Apr 2010) | 1 line Document signal handling functions in unittest.rst ........ r79792 | michael.foord | 2010-04-05 05:26:26 -0500 (Mon, 05 Apr 2010) | 1 line Documentation fixes for unittest ........ r79793 | michael.foord | 2010-04-05 05:28:27 -0500 (Mon, 05 Apr 2010) | 1 line Furterh documentation fix for unittest.rst ........ r79794 | michael.foord | 2010-04-05 05:30:14 -0500 (Mon, 05 Apr 2010) | 1 line Further documentation fix for unittest.rst ........ r79877 | michael.foord | 2010-04-06 18:18:16 -0500 (Tue, 06 Apr 2010) | 1 line Fix module directory finding logic for dotted paths in unittest test discovery. ........ r79898 | michael.foord | 2010-04-07 18:04:22 -0500 (Wed, 07 Apr 2010) | 1 line unittest.result.TestResult does not create its buffers until they're used. It uses StringIO not cStringIO. Issue 8333. ........ r79899 | michael.foord | 2010-04-07 19:04:24 -0500 (Wed, 07 Apr 2010) | 1 line Switch regrtest to use StringIO instead of cStringIO for test_multiprocessing on Windows. Issue 8333. ........ r79900 | michael.foord | 2010-04-07 23:33:20 -0500 (Wed, 07 Apr 2010) | 1 line Correction of unittest documentation typos and omissions ........
2010-04-11 17:43:16 -03:00
class TempDirMixin(object):
def setUp(self):
self.temp_dir = tempfile.mkdtemp()
self.old_dir = os.getcwd()
os.chdir(self.temp_dir)
def tearDown(self):
os.chdir(self.old_dir)
for root, dirs, files in os.walk(self.temp_dir, topdown=False):
for name in files:
os.chmod(os.path.join(self.temp_dir, name), stat.S_IWRITE)
shutil.rmtree(self.temp_dir, True)
def create_readonly_file(self, filename):
file_path = os.path.join(self.temp_dir, filename)
with open(file_path, 'w') as file:
file.write(filename)
os.chmod(file_path, stat.S_IREAD)
class Sig(object):
def __init__(self, *args, **kwargs):
self.args = args
self.kwargs = kwargs
class NS(object):
def __init__(self, **kwargs):
self.__dict__.update(kwargs)
def __repr__(self):
sorted_items = sorted(self.__dict__.items())
kwarg_str = ', '.join(['%s=%r' % tup for tup in sorted_items])
return '%s(%s)' % (type(self).__name__, kwarg_str)
def __eq__(self, other):
return vars(self) == vars(other)
class ArgumentParserError(Exception):
def __init__(self, message, stdout=None, stderr=None, error_code=None):
Exception.__init__(self, message, stdout, stderr)
self.message = message
self.stdout = stdout
self.stderr = stderr
self.error_code = error_code
def stderr_to_parser_error(parse_args, *args, **kwargs):
# if this is being called recursively and stderr or stdout is already being
# redirected, simply call the function and let the enclosing function
# catch the exception
Merged revisions 79464,79471,79623,79626,79630,79632,79643,79648-79649,79679,79685,79711,79761,79774,79777,79792-79794,79877,79898-79900 via svnmerge from svn+ssh://pythondev@svn.python.org/python/trunk ........ r79464 | michael.foord | 2010-03-27 07:55:19 -0500 (Sat, 27 Mar 2010) | 1 line A fix for running unittest tests on platforms without the audioop module (e.g. jython and IronPython) ........ r79471 | michael.foord | 2010-03-27 14:10:11 -0500 (Sat, 27 Mar 2010) | 4 lines Addition of delta keyword argument to unittest.TestCase.assertAlmostEquals and assertNotAlmostEquals This allows the comparison of objects by specifying a maximum difference; this includes the comparing of non-numeric objects that don't support rounding. ........ r79623 | michael.foord | 2010-04-02 16:42:47 -0500 (Fri, 02 Apr 2010) | 1 line Addition of -b command line option to unittest for buffering stdout and stderr during test runs. ........ r79626 | michael.foord | 2010-04-02 17:08:29 -0500 (Fri, 02 Apr 2010) | 1 line TestResult stores original sys.stdout and tests no longer use sys.__stdout__ (etc) in tests for unittest -b command line option ........ r79630 | michael.foord | 2010-04-02 17:30:56 -0500 (Fri, 02 Apr 2010) | 1 line unittest tests no longer replace the sys.stdout put in place by regrtest ........ r79632 | michael.foord | 2010-04-02 17:55:59 -0500 (Fri, 02 Apr 2010) | 1 line Issue #8038: Addition of unittest.TestCase.assertNotRegexpMatches ........ r79643 | michael.foord | 2010-04-02 20:15:21 -0500 (Fri, 02 Apr 2010) | 1 line Support dotted module names for test discovery paths in unittest. Issue 8038. ........ r79648 | michael.foord | 2010-04-02 21:21:39 -0500 (Fri, 02 Apr 2010) | 1 line Cross platform unittest.TestResult newline handling when buffering stdout / stderr. ........ r79649 | michael.foord | 2010-04-02 21:33:55 -0500 (Fri, 02 Apr 2010) | 1 line Another attempt at a fix for unittest.test.test_result for windows line endings ........ r79679 | michael.foord | 2010-04-03 09:52:18 -0500 (Sat, 03 Apr 2010) | 1 line Adding -b command line option to the unittest usage message. ........ r79685 | michael.foord | 2010-04-03 10:20:00 -0500 (Sat, 03 Apr 2010) | 1 line Minor tweak to unittest command line usage message ........ r79711 | michael.foord | 2010-04-03 12:03:11 -0500 (Sat, 03 Apr 2010) | 1 line Documenting new features in unittest ........ r79761 | michael.foord | 2010-04-04 17:41:54 -0500 (Sun, 04 Apr 2010) | 1 line unittest documentation formatting changes ........ r79774 | michael.foord | 2010-04-04 18:28:44 -0500 (Sun, 04 Apr 2010) | 1 line Adding documentation for new unittest.main() parameters ........ r79777 | michael.foord | 2010-04-04 19:39:50 -0500 (Sun, 04 Apr 2010) | 1 line Document signal handling functions in unittest.rst ........ r79792 | michael.foord | 2010-04-05 05:26:26 -0500 (Mon, 05 Apr 2010) | 1 line Documentation fixes for unittest ........ r79793 | michael.foord | 2010-04-05 05:28:27 -0500 (Mon, 05 Apr 2010) | 1 line Furterh documentation fix for unittest.rst ........ r79794 | michael.foord | 2010-04-05 05:30:14 -0500 (Mon, 05 Apr 2010) | 1 line Further documentation fix for unittest.rst ........ r79877 | michael.foord | 2010-04-06 18:18:16 -0500 (Tue, 06 Apr 2010) | 1 line Fix module directory finding logic for dotted paths in unittest test discovery. ........ r79898 | michael.foord | 2010-04-07 18:04:22 -0500 (Wed, 07 Apr 2010) | 1 line unittest.result.TestResult does not create its buffers until they're used. It uses StringIO not cStringIO. Issue 8333. ........ r79899 | michael.foord | 2010-04-07 19:04:24 -0500 (Wed, 07 Apr 2010) | 1 line Switch regrtest to use StringIO instead of cStringIO for test_multiprocessing on Windows. Issue 8333. ........ r79900 | michael.foord | 2010-04-07 23:33:20 -0500 (Wed, 07 Apr 2010) | 1 line Correction of unittest documentation typos and omissions ........
2010-04-11 17:43:16 -03:00
if isinstance(sys.stderr, StdIOBuffer) or isinstance(sys.stdout, StdIOBuffer):
return parse_args(*args, **kwargs)
# if this is not being called recursively, redirect stderr and
# use it as the ArgumentParserError message
old_stdout = sys.stdout
old_stderr = sys.stderr
Merged revisions 79464,79471,79623,79626,79630,79632,79643,79648-79649,79679,79685,79711,79761,79774,79777,79792-79794,79877,79898-79900 via svnmerge from svn+ssh://pythondev@svn.python.org/python/trunk ........ r79464 | michael.foord | 2010-03-27 07:55:19 -0500 (Sat, 27 Mar 2010) | 1 line A fix for running unittest tests on platforms without the audioop module (e.g. jython and IronPython) ........ r79471 | michael.foord | 2010-03-27 14:10:11 -0500 (Sat, 27 Mar 2010) | 4 lines Addition of delta keyword argument to unittest.TestCase.assertAlmostEquals and assertNotAlmostEquals This allows the comparison of objects by specifying a maximum difference; this includes the comparing of non-numeric objects that don't support rounding. ........ r79623 | michael.foord | 2010-04-02 16:42:47 -0500 (Fri, 02 Apr 2010) | 1 line Addition of -b command line option to unittest for buffering stdout and stderr during test runs. ........ r79626 | michael.foord | 2010-04-02 17:08:29 -0500 (Fri, 02 Apr 2010) | 1 line TestResult stores original sys.stdout and tests no longer use sys.__stdout__ (etc) in tests for unittest -b command line option ........ r79630 | michael.foord | 2010-04-02 17:30:56 -0500 (Fri, 02 Apr 2010) | 1 line unittest tests no longer replace the sys.stdout put in place by regrtest ........ r79632 | michael.foord | 2010-04-02 17:55:59 -0500 (Fri, 02 Apr 2010) | 1 line Issue #8038: Addition of unittest.TestCase.assertNotRegexpMatches ........ r79643 | michael.foord | 2010-04-02 20:15:21 -0500 (Fri, 02 Apr 2010) | 1 line Support dotted module names for test discovery paths in unittest. Issue 8038. ........ r79648 | michael.foord | 2010-04-02 21:21:39 -0500 (Fri, 02 Apr 2010) | 1 line Cross platform unittest.TestResult newline handling when buffering stdout / stderr. ........ r79649 | michael.foord | 2010-04-02 21:33:55 -0500 (Fri, 02 Apr 2010) | 1 line Another attempt at a fix for unittest.test.test_result for windows line endings ........ r79679 | michael.foord | 2010-04-03 09:52:18 -0500 (Sat, 03 Apr 2010) | 1 line Adding -b command line option to the unittest usage message. ........ r79685 | michael.foord | 2010-04-03 10:20:00 -0500 (Sat, 03 Apr 2010) | 1 line Minor tweak to unittest command line usage message ........ r79711 | michael.foord | 2010-04-03 12:03:11 -0500 (Sat, 03 Apr 2010) | 1 line Documenting new features in unittest ........ r79761 | michael.foord | 2010-04-04 17:41:54 -0500 (Sun, 04 Apr 2010) | 1 line unittest documentation formatting changes ........ r79774 | michael.foord | 2010-04-04 18:28:44 -0500 (Sun, 04 Apr 2010) | 1 line Adding documentation for new unittest.main() parameters ........ r79777 | michael.foord | 2010-04-04 19:39:50 -0500 (Sun, 04 Apr 2010) | 1 line Document signal handling functions in unittest.rst ........ r79792 | michael.foord | 2010-04-05 05:26:26 -0500 (Mon, 05 Apr 2010) | 1 line Documentation fixes for unittest ........ r79793 | michael.foord | 2010-04-05 05:28:27 -0500 (Mon, 05 Apr 2010) | 1 line Furterh documentation fix for unittest.rst ........ r79794 | michael.foord | 2010-04-05 05:30:14 -0500 (Mon, 05 Apr 2010) | 1 line Further documentation fix for unittest.rst ........ r79877 | michael.foord | 2010-04-06 18:18:16 -0500 (Tue, 06 Apr 2010) | 1 line Fix module directory finding logic for dotted paths in unittest test discovery. ........ r79898 | michael.foord | 2010-04-07 18:04:22 -0500 (Wed, 07 Apr 2010) | 1 line unittest.result.TestResult does not create its buffers until they're used. It uses StringIO not cStringIO. Issue 8333. ........ r79899 | michael.foord | 2010-04-07 19:04:24 -0500 (Wed, 07 Apr 2010) | 1 line Switch regrtest to use StringIO instead of cStringIO for test_multiprocessing on Windows. Issue 8333. ........ r79900 | michael.foord | 2010-04-07 23:33:20 -0500 (Wed, 07 Apr 2010) | 1 line Correction of unittest documentation typos and omissions ........
2010-04-11 17:43:16 -03:00
sys.stdout = StdIOBuffer()
sys.stderr = StdIOBuffer()
try:
try:
result = parse_args(*args, **kwargs)
for key in list(vars(result)):
if getattr(result, key) is sys.stdout:
setattr(result, key, old_stdout)
if getattr(result, key) is sys.stderr:
setattr(result, key, old_stderr)
return result
except SystemExit:
code = sys.exc_info()[1].code
stdout = sys.stdout.getvalue()
stderr = sys.stderr.getvalue()
raise ArgumentParserError(
"SystemExit", stdout, stderr, code) from None
finally:
sys.stdout = old_stdout
sys.stderr = old_stderr
class ErrorRaisingArgumentParser(argparse.ArgumentParser):
def parse_args(self, *args, **kwargs):
parse_args = super(ErrorRaisingArgumentParser, self).parse_args
return stderr_to_parser_error(parse_args, *args, **kwargs)
def exit(self, *args, **kwargs):
exit = super(ErrorRaisingArgumentParser, self).exit
return stderr_to_parser_error(exit, *args, **kwargs)
def error(self, *args, **kwargs):
error = super(ErrorRaisingArgumentParser, self).error
return stderr_to_parser_error(error, *args, **kwargs)
class ParserTesterMetaclass(type):
"""Adds parser tests using the class attributes.
Classes of this type should specify the following attributes:
argument_signatures -- a list of Sig objects which specify
the signatures of Argument objects to be created
failures -- a list of args lists that should cause the parser
to fail
successes -- a list of (initial_args, options, remaining_args) tuples
where initial_args specifies the string args to be parsed,
options is a dict that should match the vars() of the options
parsed out of initial_args, and remaining_args should be any
remaining unparsed arguments
"""
def __init__(cls, name, bases, bodydict):
if name == 'ParserTestCase':
return
# default parser signature is empty
if not hasattr(cls, 'parser_signature'):
cls.parser_signature = Sig()
if not hasattr(cls, 'parser_class'):
cls.parser_class = ErrorRaisingArgumentParser
# ---------------------------------------
# functions for adding optional arguments
# ---------------------------------------
def no_groups(parser, argument_signatures):
"""Add all arguments directly to the parser"""
for sig in argument_signatures:
parser.add_argument(*sig.args, **sig.kwargs)
def one_group(parser, argument_signatures):
"""Add all arguments under a single group in the parser"""
group = parser.add_argument_group('foo')
for sig in argument_signatures:
group.add_argument(*sig.args, **sig.kwargs)
def many_groups(parser, argument_signatures):
"""Add each argument in its own group to the parser"""
for i, sig in enumerate(argument_signatures):
group = parser.add_argument_group('foo:%i' % i)
group.add_argument(*sig.args, **sig.kwargs)
# --------------------------
# functions for parsing args
# --------------------------
def listargs(parser, args):
"""Parse the args by passing in a list"""
return parser.parse_args(args)
def sysargs(parser, args):
"""Parse the args by defaulting to sys.argv"""
old_sys_argv = sys.argv
sys.argv = [old_sys_argv[0]] + args
try:
return parser.parse_args()
finally:
sys.argv = old_sys_argv
# class that holds the combination of one optional argument
# addition method and one arg parsing method
class AddTests(object):
def __init__(self, tester_cls, add_arguments, parse_args):
self._add_arguments = add_arguments
self._parse_args = parse_args
add_arguments_name = self._add_arguments.__name__
parse_args_name = self._parse_args.__name__
for test_func in [self.test_failures, self.test_successes]:
func_name = test_func.__name__
names = func_name, add_arguments_name, parse_args_name
test_name = '_'.join(names)
def wrapper(self, test_func=test_func):
test_func(self)
try:
wrapper.__name__ = test_name
except TypeError:
pass
setattr(tester_cls, test_name, wrapper)
def _get_parser(self, tester):
args = tester.parser_signature.args
kwargs = tester.parser_signature.kwargs
parser = tester.parser_class(*args, **kwargs)
self._add_arguments(parser, tester.argument_signatures)
return parser
def test_failures(self, tester):
parser = self._get_parser(tester)
for args_str in tester.failures:
args = args_str.split()
with tester.assertRaises(ArgumentParserError, msg=args):
parser.parse_args(args)
def test_successes(self, tester):
parser = self._get_parser(tester)
for args, expected_ns in tester.successes:
if isinstance(args, str):
args = args.split()
result_ns = self._parse_args(parser, args)
tester.assertEqual(expected_ns, result_ns)
# add tests for each combination of an optionals adding method
# and an arg parsing method
for add_arguments in [no_groups, one_group, many_groups]:
for parse_args in [listargs, sysargs]:
AddTests(cls, add_arguments, parse_args)
bases = TestCase,
ParserTestCase = ParserTesterMetaclass('ParserTestCase', bases, {})
# ===============
# Optionals tests
# ===============
class TestOptionalsSingleDash(ParserTestCase):
"""Test an Optional with a single-dash option string"""
argument_signatures = [Sig('-x')]
failures = ['-x', 'a', '--foo', '-x --foo', '-x -y']
successes = [
('', NS(x=None)),
('-x a', NS(x='a')),
('-xa', NS(x='a')),
('-x -1', NS(x='-1')),
('-x-1', NS(x='-1')),
]
class TestOptionalsSingleDashCombined(ParserTestCase):
"""Test an Optional with a single-dash option string"""
argument_signatures = [
Sig('-x', action='store_true'),
Sig('-yyy', action='store_const', const=42),
Sig('-z'),
]
failures = ['a', '--foo', '-xa', '-x --foo', '-x -z', '-z -x',
'-yx', '-yz a', '-yyyx', '-yyyza', '-xyza']
successes = [
('', NS(x=False, yyy=None, z=None)),
('-x', NS(x=True, yyy=None, z=None)),
('-za', NS(x=False, yyy=None, z='a')),
('-z a', NS(x=False, yyy=None, z='a')),
('-xza', NS(x=True, yyy=None, z='a')),
('-xz a', NS(x=True, yyy=None, z='a')),
('-x -za', NS(x=True, yyy=None, z='a')),
('-x -z a', NS(x=True, yyy=None, z='a')),
('-y', NS(x=False, yyy=42, z=None)),
('-yyy', NS(x=False, yyy=42, z=None)),
('-x -yyy -za', NS(x=True, yyy=42, z='a')),
('-x -yyy -z a', NS(x=True, yyy=42, z='a')),
]
class TestOptionalsSingleDashLong(ParserTestCase):
"""Test an Optional with a multi-character single-dash option string"""
argument_signatures = [Sig('-foo')]
failures = ['-foo', 'a', '--foo', '-foo --foo', '-foo -y', '-fooa']
successes = [
('', NS(foo=None)),
('-foo a', NS(foo='a')),
('-foo -1', NS(foo='-1')),
('-fo a', NS(foo='a')),
('-f a', NS(foo='a')),
]
class TestOptionalsSingleDashSubsetAmbiguous(ParserTestCase):
"""Test Optionals where option strings are subsets of each other"""
argument_signatures = [Sig('-f'), Sig('-foobar'), Sig('-foorab')]
failures = ['-f', '-foo', '-fo', '-foo b', '-foob', '-fooba', '-foora']
successes = [
('', NS(f=None, foobar=None, foorab=None)),
('-f a', NS(f='a', foobar=None, foorab=None)),
('-fa', NS(f='a', foobar=None, foorab=None)),
('-foa', NS(f='oa', foobar=None, foorab=None)),
('-fooa', NS(f='ooa', foobar=None, foorab=None)),
('-foobar a', NS(f=None, foobar='a', foorab=None)),
('-foorab a', NS(f=None, foobar=None, foorab='a')),
]
class TestOptionalsSingleDashAmbiguous(ParserTestCase):
"""Test Optionals that partially match but are not subsets"""
argument_signatures = [Sig('-foobar'), Sig('-foorab')]
failures = ['-f', '-f a', '-fa', '-foa', '-foo', '-fo', '-foo b']
successes = [
('', NS(foobar=None, foorab=None)),
('-foob a', NS(foobar='a', foorab=None)),
('-foor a', NS(foobar=None, foorab='a')),
('-fooba a', NS(foobar='a', foorab=None)),
('-foora a', NS(foobar=None, foorab='a')),
('-foobar a', NS(foobar='a', foorab=None)),
('-foorab a', NS(foobar=None, foorab='a')),
]
class TestOptionalsNumeric(ParserTestCase):
"""Test an Optional with a short opt string"""
argument_signatures = [Sig('-1', dest='one')]
failures = ['-1', 'a', '-1 --foo', '-1 -y', '-1 -1', '-1 -2']
successes = [
('', NS(one=None)),
('-1 a', NS(one='a')),
('-1a', NS(one='a')),
('-1-2', NS(one='-2')),
]
class TestOptionalsDoubleDash(ParserTestCase):
"""Test an Optional with a double-dash option string"""
argument_signatures = [Sig('--foo')]
failures = ['--foo', '-f', '-f a', 'a', '--foo -x', '--foo --bar']
successes = [
('', NS(foo=None)),
('--foo a', NS(foo='a')),
('--foo=a', NS(foo='a')),
('--foo -2.5', NS(foo='-2.5')),
('--foo=-2.5', NS(foo='-2.5')),
]
class TestOptionalsDoubleDashPartialMatch(ParserTestCase):
"""Tests partial matching with a double-dash option string"""
argument_signatures = [
Sig('--badger', action='store_true'),
Sig('--bat'),
]
failures = ['--bar', '--b', '--ba', '--b=2', '--ba=4', '--badge 5']
successes = [
('', NS(badger=False, bat=None)),
('--bat X', NS(badger=False, bat='X')),
('--bad', NS(badger=True, bat=None)),
('--badg', NS(badger=True, bat=None)),
('--badge', NS(badger=True, bat=None)),
('--badger', NS(badger=True, bat=None)),
]
class TestOptionalsDoubleDashPrefixMatch(ParserTestCase):
"""Tests when one double-dash option string is a prefix of another"""
argument_signatures = [
Sig('--badger', action='store_true'),
Sig('--ba'),
]
failures = ['--bar', '--b', '--ba', '--b=2', '--badge 5']
successes = [
('', NS(badger=False, ba=None)),
('--ba X', NS(badger=False, ba='X')),
('--ba=X', NS(badger=False, ba='X')),
('--bad', NS(badger=True, ba=None)),
('--badg', NS(badger=True, ba=None)),
('--badge', NS(badger=True, ba=None)),
('--badger', NS(badger=True, ba=None)),
]
class TestOptionalsSingleDoubleDash(ParserTestCase):
"""Test an Optional with single- and double-dash option strings"""
argument_signatures = [
Sig('-f', action='store_true'),
Sig('--bar'),
Sig('-baz', action='store_const', const=42),
]
failures = ['--bar', '-fbar', '-fbaz', '-bazf', '-b B', 'B']
successes = [
('', NS(f=False, bar=None, baz=None)),
('-f', NS(f=True, bar=None, baz=None)),
('--ba B', NS(f=False, bar='B', baz=None)),
('-f --bar B', NS(f=True, bar='B', baz=None)),
('-f -b', NS(f=True, bar=None, baz=42)),
('-ba -f', NS(f=True, bar=None, baz=42)),
]
class TestOptionalsAlternatePrefixChars(ParserTestCase):
"""Test an Optional with option strings with custom prefixes"""
parser_signature = Sig(prefix_chars='+:/', add_help=False)
argument_signatures = [
Sig('+f', action='store_true'),
Sig('::bar'),
Sig('/baz', action='store_const', const=42),
]
failures = ['--bar', '-fbar', '-b B', 'B', '-f', '--bar B', '-baz', '-h', '--help', '+h', '::help', '/help']
successes = [
('', NS(f=False, bar=None, baz=None)),
('+f', NS(f=True, bar=None, baz=None)),
('::ba B', NS(f=False, bar='B', baz=None)),
('+f ::bar B', NS(f=True, bar='B', baz=None)),
('+f /b', NS(f=True, bar=None, baz=42)),
('/ba +f', NS(f=True, bar=None, baz=42)),
]
class TestOptionalsAlternatePrefixCharsAddedHelp(ParserTestCase):
"""When ``-`` not in prefix_chars, default operators created for help
should use the prefix_chars in use rather than - or --
http://bugs.python.org/issue9444"""
parser_signature = Sig(prefix_chars='+:/', add_help=True)
argument_signatures = [
Sig('+f', action='store_true'),
Sig('::bar'),
Sig('/baz', action='store_const', const=42),
]
failures = ['--bar', '-fbar', '-b B', 'B', '-f', '--bar B', '-baz']
successes = [
('', NS(f=False, bar=None, baz=None)),
('+f', NS(f=True, bar=None, baz=None)),
('::ba B', NS(f=False, bar='B', baz=None)),
('+f ::bar B', NS(f=True, bar='B', baz=None)),
('+f /b', NS(f=True, bar=None, baz=42)),
('/ba +f', NS(f=True, bar=None, baz=42))
]
class TestOptionalsAlternatePrefixCharsMultipleShortArgs(ParserTestCase):
"""Verify that Optionals must be called with their defined prefixes"""
parser_signature = Sig(prefix_chars='+-', add_help=False)
argument_signatures = [
Sig('-x', action='store_true'),
Sig('+y', action='store_true'),
Sig('+z', action='store_true'),
]
failures = ['-w',
'-xyz',
'+x',
'-y',
'+xyz',
]
successes = [
('', NS(x=False, y=False, z=False)),
('-x', NS(x=True, y=False, z=False)),
('+y -x', NS(x=True, y=True, z=False)),
('+yz -x', NS(x=True, y=True, z=True)),
]
class TestOptionalsShortLong(ParserTestCase):
"""Test a combination of single- and double-dash option strings"""
argument_signatures = [
Sig('-v', '--verbose', '-n', '--noisy', action='store_true'),
]
failures = ['--x --verbose', '-N', 'a', '-v x']
successes = [
('', NS(verbose=False)),
('-v', NS(verbose=True)),
('--verbose', NS(verbose=True)),
('-n', NS(verbose=True)),
('--noisy', NS(verbose=True)),
]
class TestOptionalsDest(ParserTestCase):
"""Tests various means of setting destination"""
argument_signatures = [Sig('--foo-bar'), Sig('--baz', dest='zabbaz')]
failures = ['a']
successes = [
('--foo-bar f', NS(foo_bar='f', zabbaz=None)),
('--baz g', NS(foo_bar=None, zabbaz='g')),
('--foo-bar h --baz i', NS(foo_bar='h', zabbaz='i')),
('--baz j --foo-bar k', NS(foo_bar='k', zabbaz='j')),
]
class TestOptionalsDefault(ParserTestCase):
"""Tests specifying a default for an Optional"""
argument_signatures = [Sig('-x'), Sig('-y', default=42)]
failures = ['a']
successes = [
('', NS(x=None, y=42)),
('-xx', NS(x='x', y=42)),
('-yy', NS(x=None, y='y')),
]
class TestOptionalsNargsDefault(ParserTestCase):
"""Tests not specifying the number of args for an Optional"""
argument_signatures = [Sig('-x')]
failures = ['a', '-x']
successes = [
('', NS(x=None)),
('-x a', NS(x='a')),
]
class TestOptionalsNargs1(ParserTestCase):
"""Tests specifying 1 arg for an Optional"""
argument_signatures = [Sig('-x', nargs=1)]
failures = ['a', '-x']
successes = [
('', NS(x=None)),
('-x a', NS(x=['a'])),
]
class TestOptionalsNargs3(ParserTestCase):
"""Tests specifying 3 args for an Optional"""
argument_signatures = [Sig('-x', nargs=3)]
failures = ['a', '-x', '-x a', '-x a b', 'a -x', 'a -x b']
successes = [
('', NS(x=None)),
('-x a b c', NS(x=['a', 'b', 'c'])),
]
class TestOptionalsNargsOptional(ParserTestCase):
"""Tests specifying an Optional arg for an Optional"""
argument_signatures = [
Sig('-w', nargs='?'),
Sig('-x', nargs='?', const=42),
Sig('-y', nargs='?', default='spam'),
Sig('-z', nargs='?', type=int, const='42', default='84'),
]
failures = ['2']
successes = [
('', NS(w=None, x=None, y='spam', z=84)),
('-w', NS(w=None, x=None, y='spam', z=84)),
('-w 2', NS(w='2', x=None, y='spam', z=84)),
('-x', NS(w=None, x=42, y='spam', z=84)),
('-x 2', NS(w=None, x='2', y='spam', z=84)),
('-y', NS(w=None, x=None, y=None, z=84)),
('-y 2', NS(w=None, x=None, y='2', z=84)),
('-z', NS(w=None, x=None, y='spam', z=42)),
('-z 2', NS(w=None, x=None, y='spam', z=2)),
]
class TestOptionalsNargsZeroOrMore(ParserTestCase):
"""Tests specifying args for an Optional that accepts zero or more"""
argument_signatures = [
Sig('-x', nargs='*'),
Sig('-y', nargs='*', default='spam'),
]
failures = ['a']
successes = [
('', NS(x=None, y='spam')),
('-x', NS(x=[], y='spam')),
('-x a', NS(x=['a'], y='spam')),
('-x a b', NS(x=['a', 'b'], y='spam')),
('-y', NS(x=None, y=[])),
('-y a', NS(x=None, y=['a'])),
('-y a b', NS(x=None, y=['a', 'b'])),
]
class TestOptionalsNargsOneOrMore(ParserTestCase):
"""Tests specifying args for an Optional that accepts one or more"""
argument_signatures = [
Sig('-x', nargs='+'),
Sig('-y', nargs='+', default='spam'),
]
failures = ['a', '-x', '-y', 'a -x', 'a -y b']
successes = [
('', NS(x=None, y='spam')),
('-x a', NS(x=['a'], y='spam')),
('-x a b', NS(x=['a', 'b'], y='spam')),
('-y a', NS(x=None, y=['a'])),
('-y a b', NS(x=None, y=['a', 'b'])),
]
class TestOptionalsChoices(ParserTestCase):
"""Tests specifying the choices for an Optional"""
argument_signatures = [
Sig('-f', choices='abc'),
Sig('-g', type=int, choices=range(5))]
failures = ['a', '-f d', '-fad', '-ga', '-g 6']
successes = [
('', NS(f=None, g=None)),
('-f a', NS(f='a', g=None)),
('-f c', NS(f='c', g=None)),
('-g 0', NS(f=None, g=0)),
('-g 03', NS(f=None, g=3)),
('-fb -g4', NS(f='b', g=4)),
]
class TestOptionalsRequired(ParserTestCase):
"""Tests an optional action that is required"""
argument_signatures = [
Sig('-x', type=int, required=True),
]
failures = ['a', '']
successes = [
('-x 1', NS(x=1)),
('-x42', NS(x=42)),
]
class TestOptionalsActionStore(ParserTestCase):
"""Tests the store action for an Optional"""
argument_signatures = [Sig('-x', action='store')]
failures = ['a', 'a -x']
successes = [
('', NS(x=None)),
('-xfoo', NS(x='foo')),
]
class TestOptionalsActionStoreConst(ParserTestCase):
"""Tests the store_const action for an Optional"""
argument_signatures = [Sig('-y', action='store_const', const=object)]
failures = ['a']
successes = [
('', NS(y=None)),
('-y', NS(y=object)),
]
class TestOptionalsActionStoreFalse(ParserTestCase):
"""Tests the store_false action for an Optional"""
argument_signatures = [Sig('-z', action='store_false')]
failures = ['a', '-za', '-z a']
successes = [
('', NS(z=True)),
('-z', NS(z=False)),
]
class TestOptionalsActionStoreTrue(ParserTestCase):
"""Tests the store_true action for an Optional"""
argument_signatures = [Sig('--apple', action='store_true')]
failures = ['a', '--apple=b', '--apple b']
successes = [
('', NS(apple=False)),
('--apple', NS(apple=True)),
]
class TestBooleanOptionalAction(ParserTestCase):
"""Tests BooleanOptionalAction"""
argument_signatures = [Sig('--foo', action=argparse.BooleanOptionalAction)]
failures = ['--foo bar', '--foo=bar']
successes = [
('', NS(foo=None)),
('--foo', NS(foo=True)),
('--no-foo', NS(foo=False)),
('--foo --no-foo', NS(foo=False)), # useful for aliases
('--no-foo --foo', NS(foo=True)),
]
class TestBooleanOptionalActionRequired(ParserTestCase):
"""Tests BooleanOptionalAction required"""
argument_signatures = [
Sig('--foo', required=True, action=argparse.BooleanOptionalAction)
]
failures = ['']
successes = [
('--foo', NS(foo=True)),
('--no-foo', NS(foo=False)),
]
class TestOptionalsActionAppend(ParserTestCase):
"""Tests the append action for an Optional"""
argument_signatures = [Sig('--baz', action='append')]
failures = ['a', '--baz', 'a --baz', '--baz a b']
successes = [
('', NS(baz=None)),
('--baz a', NS(baz=['a'])),
('--baz a --baz b', NS(baz=['a', 'b'])),
]
class TestOptionalsActionAppendWithDefault(ParserTestCase):
"""Tests the append action for an Optional"""
argument_signatures = [Sig('--baz', action='append', default=['X'])]
failures = ['a', '--baz', 'a --baz', '--baz a b']
successes = [
('', NS(baz=['X'])),
('--baz a', NS(baz=['X', 'a'])),
('--baz a --baz b', NS(baz=['X', 'a', 'b'])),
]
class TestOptionalsActionAppendConst(ParserTestCase):
"""Tests the append_const action for an Optional"""
argument_signatures = [
Sig('-b', action='append_const', const=Exception),
Sig('-c', action='append', dest='b'),
]
failures = ['a', '-c', 'a -c', '-bx', '-b x']
successes = [
('', NS(b=None)),
('-b', NS(b=[Exception])),
('-b -cx -b -cyz', NS(b=[Exception, 'x', Exception, 'yz'])),
]
class TestOptionalsActionAppendConstWithDefault(ParserTestCase):
"""Tests the append_const action for an Optional"""
argument_signatures = [
Sig('-b', action='append_const', const=Exception, default=['X']),
Sig('-c', action='append', dest='b'),
]
failures = ['a', '-c', 'a -c', '-bx', '-b x']
successes = [
('', NS(b=['X'])),
('-b', NS(b=['X', Exception])),
('-b -cx -b -cyz', NS(b=['X', Exception, 'x', Exception, 'yz'])),
]
class TestOptionalsActionCount(ParserTestCase):
"""Tests the count action for an Optional"""
argument_signatures = [Sig('-x', action='count')]
failures = ['a', '-x a', '-x b', '-x a -x b']
successes = [
('', NS(x=None)),
('-x', NS(x=1)),
]
class TestOptionalsAllowLongAbbreviation(ParserTestCase):
"""Allow long options to be abbreviated unambiguously"""
argument_signatures = [
Sig('--foo'),
Sig('--foobaz'),
Sig('--fooble', action='store_true'),
]
failures = ['--foob 5', '--foob']
successes = [
('', NS(foo=None, foobaz=None, fooble=False)),
('--foo 7', NS(foo='7', foobaz=None, fooble=False)),
('--fooba a', NS(foo=None, foobaz='a', fooble=False)),
('--foobl --foo g', NS(foo='g', foobaz=None, fooble=True)),
]
class TestOptionalsDisallowLongAbbreviation(ParserTestCase):
"""Do not allow abbreviations of long options at all"""
parser_signature = Sig(allow_abbrev=False)
argument_signatures = [
Sig('--foo'),
Sig('--foodle', action='store_true'),
Sig('--foonly'),
]
failures = ['-foon 3', '--foon 3', '--food', '--food --foo 2']
successes = [
('', NS(foo=None, foodle=False, foonly=None)),
('--foo 3', NS(foo='3', foodle=False, foonly=None)),
('--foonly 7 --foodle --foo 2', NS(foo='2', foodle=True, foonly='7')),
]
class TestOptionalsDisallowLongAbbreviationPrefixChars(ParserTestCase):
"""Disallowing abbreviations works with alternative prefix characters"""
parser_signature = Sig(prefix_chars='+', allow_abbrev=False)
argument_signatures = [
Sig('++foo'),
Sig('++foodle', action='store_true'),
Sig('++foonly'),
]
failures = ['+foon 3', '++foon 3', '++food', '++food ++foo 2']
successes = [
('', NS(foo=None, foodle=False, foonly=None)),
('++foo 3', NS(foo='3', foodle=False, foonly=None)),
('++foonly 7 ++foodle ++foo 2', NS(foo='2', foodle=True, foonly='7')),
]
class TestDisallowLongAbbreviationAllowsShortGrouping(ParserTestCase):
"""Do not allow abbreviations of long options at all"""
parser_signature = Sig(allow_abbrev=False)
argument_signatures = [
Sig('-r'),
Sig('-c', action='count'),
]
failures = ['-r', '-c -r']
successes = [
('', NS(r=None, c=None)),
('-ra', NS(r='a', c=None)),
('-rcc', NS(r='cc', c=None)),
('-cc', NS(r=None, c=2)),
('-cc -ra', NS(r='a', c=2)),
('-ccrcc', NS(r='cc', c=2)),
]
class TestDisallowLongAbbreviationAllowsShortGroupingPrefix(ParserTestCase):
"""Short option grouping works with custom prefix and allow_abbrev=False"""
parser_signature = Sig(prefix_chars='+', allow_abbrev=False)
argument_signatures = [
Sig('+r'),
Sig('+c', action='count'),
]
failures = ['+r', '+c +r']
successes = [
('', NS(r=None, c=None)),
('+ra', NS(r='a', c=None)),
('+rcc', NS(r='cc', c=None)),
('+cc', NS(r=None, c=2)),
('+cc +ra', NS(r='a', c=2)),
('+ccrcc', NS(r='cc', c=2)),
]
# ================
# Positional tests
# ================
class TestPositionalsNargsNone(ParserTestCase):
"""Test a Positional that doesn't specify nargs"""
argument_signatures = [Sig('foo')]
failures = ['', '-x', 'a b']
successes = [
('a', NS(foo='a')),
]
class TestPositionalsNargs1(ParserTestCase):
"""Test a Positional that specifies an nargs of 1"""
argument_signatures = [Sig('foo', nargs=1)]
failures = ['', '-x', 'a b']
successes = [
('a', NS(foo=['a'])),
]
class TestPositionalsNargs2(ParserTestCase):
"""Test a Positional that specifies an nargs of 2"""
argument_signatures = [Sig('foo', nargs=2)]
failures = ['', 'a', '-x', 'a b c']
successes = [
('a b', NS(foo=['a', 'b'])),
]
class TestPositionalsNargsZeroOrMore(ParserTestCase):
"""Test a Positional that specifies unlimited nargs"""
argument_signatures = [Sig('foo', nargs='*')]
failures = ['-x']
successes = [
('', NS(foo=[])),
('a', NS(foo=['a'])),
('a b', NS(foo=['a', 'b'])),
]
class TestPositionalsNargsZeroOrMoreDefault(ParserTestCase):
"""Test a Positional that specifies unlimited nargs and a default"""
argument_signatures = [Sig('foo', nargs='*', default='bar')]
failures = ['-x']
successes = [
('', NS(foo='bar')),
('a', NS(foo=['a'])),
('a b', NS(foo=['a', 'b'])),
]
class TestPositionalsNargsOneOrMore(ParserTestCase):
"""Test a Positional that specifies one or more nargs"""
argument_signatures = [Sig('foo', nargs='+')]
failures = ['', '-x']
successes = [
('a', NS(foo=['a'])),
('a b', NS(foo=['a', 'b'])),
]
class TestPositionalsNargsOptional(ParserTestCase):
"""Tests an Optional Positional"""
argument_signatures = [Sig('foo', nargs='?')]
failures = ['-x', 'a b']
successes = [
('', NS(foo=None)),
('a', NS(foo='a')),
]
class TestPositionalsNargsOptionalDefault(ParserTestCase):
"""Tests an Optional Positional with a default value"""
argument_signatures = [Sig('foo', nargs='?', default=42)]
failures = ['-x', 'a b']
successes = [
('', NS(foo=42)),
('a', NS(foo='a')),
]
class TestPositionalsNargsOptionalConvertedDefault(ParserTestCase):
"""Tests an Optional Positional with a default value
that needs to be converted to the appropriate type.
"""
argument_signatures = [
Sig('foo', nargs='?', type=int, default='42'),
]
failures = ['-x', 'a b', '1 2']
successes = [
('', NS(foo=42)),
('1', NS(foo=1)),
]
class TestPositionalsNargsNoneNone(ParserTestCase):
"""Test two Positionals that don't specify nargs"""
argument_signatures = [Sig('foo'), Sig('bar')]
failures = ['', '-x', 'a', 'a b c']
successes = [
('a b', NS(foo='a', bar='b')),
]
class TestPositionalsNargsNone1(ParserTestCase):
"""Test a Positional with no nargs followed by one with 1"""
argument_signatures = [Sig('foo'), Sig('bar', nargs=1)]
failures = ['', '--foo', 'a', 'a b c']
successes = [
('a b', NS(foo='a', bar=['b'])),
]
class TestPositionalsNargs2None(ParserTestCase):
"""Test a Positional with 2 nargs followed by one with none"""
argument_signatures = [Sig('foo', nargs=2), Sig('bar')]
failures = ['', '--foo', 'a', 'a b', 'a b c d']
successes = [
('a b c', NS(foo=['a', 'b'], bar='c')),
]
class TestPositionalsNargsNoneZeroOrMore(ParserTestCase):
"""Test a Positional with no nargs followed by one with unlimited"""
argument_signatures = [Sig('foo'), Sig('bar', nargs='*')]
failures = ['', '--foo']
successes = [
('a', NS(foo='a', bar=[])),
('a b', NS(foo='a', bar=['b'])),
('a b c', NS(foo='a', bar=['b', 'c'])),
]
class TestPositionalsNargsNoneOneOrMore(ParserTestCase):
"""Test a Positional with no nargs followed by one with one or more"""
argument_signatures = [Sig('foo'), Sig('bar', nargs='+')]
failures = ['', '--foo', 'a']
successes = [
('a b', NS(foo='a', bar=['b'])),
('a b c', NS(foo='a', bar=['b', 'c'])),
]
class TestPositionalsNargsNoneOptional(ParserTestCase):
"""Test a Positional with no nargs followed by one with an Optional"""
argument_signatures = [Sig('foo'), Sig('bar', nargs='?')]
failures = ['', '--foo', 'a b c']
successes = [
('a', NS(foo='a', bar=None)),
('a b', NS(foo='a', bar='b')),
]
class TestPositionalsNargsZeroOrMoreNone(ParserTestCase):
"""Test a Positional with unlimited nargs followed by one with none"""
argument_signatures = [Sig('foo', nargs='*'), Sig('bar')]
failures = ['', '--foo']
successes = [
('a', NS(foo=[], bar='a')),
('a b', NS(foo=['a'], bar='b')),
('a b c', NS(foo=['a', 'b'], bar='c')),
]
class TestPositionalsNargsOneOrMoreNone(ParserTestCase):
"""Test a Positional with one or more nargs followed by one with none"""
argument_signatures = [Sig('foo', nargs='+'), Sig('bar')]
failures = ['', '--foo', 'a']
successes = [
('a b', NS(foo=['a'], bar='b')),
('a b c', NS(foo=['a', 'b'], bar='c')),
]
class TestPositionalsNargsOptionalNone(ParserTestCase):
"""Test a Positional with an Optional nargs followed by one with none"""
argument_signatures = [Sig('foo', nargs='?', default=42), Sig('bar')]
failures = ['', '--foo', 'a b c']
successes = [
('a', NS(foo=42, bar='a')),
('a b', NS(foo='a', bar='b')),
]
class TestPositionalsNargs2ZeroOrMore(ParserTestCase):
"""Test a Positional with 2 nargs followed by one with unlimited"""
argument_signatures = [Sig('foo', nargs=2), Sig('bar', nargs='*')]
failures = ['', '--foo', 'a']
successes = [
('a b', NS(foo=['a', 'b'], bar=[])),
('a b c', NS(foo=['a', 'b'], bar=['c'])),
]
class TestPositionalsNargs2OneOrMore(ParserTestCase):
"""Test a Positional with 2 nargs followed by one with one or more"""
argument_signatures = [Sig('foo', nargs=2), Sig('bar', nargs='+')]
failures = ['', '--foo', 'a', 'a b']
successes = [
('a b c', NS(foo=['a', 'b'], bar=['c'])),
]
class TestPositionalsNargs2Optional(ParserTestCase):
"""Test a Positional with 2 nargs followed by one optional"""
argument_signatures = [Sig('foo', nargs=2), Sig('bar', nargs='?')]
failures = ['', '--foo', 'a', 'a b c d']
successes = [
('a b', NS(foo=['a', 'b'], bar=None)),
('a b c', NS(foo=['a', 'b'], bar='c')),
]
class TestPositionalsNargsZeroOrMore1(ParserTestCase):
"""Test a Positional with unlimited nargs followed by one with 1"""
argument_signatures = [Sig('foo', nargs='*'), Sig('bar', nargs=1)]
failures = ['', '--foo', ]
successes = [
('a', NS(foo=[], bar=['a'])),
('a b', NS(foo=['a'], bar=['b'])),
('a b c', NS(foo=['a', 'b'], bar=['c'])),
]
class TestPositionalsNargsOneOrMore1(ParserTestCase):
"""Test a Positional with one or more nargs followed by one with 1"""
argument_signatures = [Sig('foo', nargs='+'), Sig('bar', nargs=1)]
failures = ['', '--foo', 'a']
successes = [
('a b', NS(foo=['a'], bar=['b'])),
('a b c', NS(foo=['a', 'b'], bar=['c'])),
]
class TestPositionalsNargsOptional1(ParserTestCase):
"""Test a Positional with an Optional nargs followed by one with 1"""
argument_signatures = [Sig('foo', nargs='?'), Sig('bar', nargs=1)]
failures = ['', '--foo', 'a b c']
successes = [
('a', NS(foo=None, bar=['a'])),
('a b', NS(foo='a', bar=['b'])),
]
class TestPositionalsNargsNoneZeroOrMore1(ParserTestCase):
"""Test three Positionals: no nargs, unlimited nargs and 1 nargs"""
argument_signatures = [
Sig('foo'),
Sig('bar', nargs='*'),
Sig('baz', nargs=1),
]
failures = ['', '--foo', 'a']
successes = [
('a b', NS(foo='a', bar=[], baz=['b'])),
('a b c', NS(foo='a', bar=['b'], baz=['c'])),
]
class TestPositionalsNargsNoneOneOrMore1(ParserTestCase):
"""Test three Positionals: no nargs, one or more nargs and 1 nargs"""
argument_signatures = [
Sig('foo'),
Sig('bar', nargs='+'),
Sig('baz', nargs=1),
]
failures = ['', '--foo', 'a', 'b']
successes = [
('a b c', NS(foo='a', bar=['b'], baz=['c'])),
('a b c d', NS(foo='a', bar=['b', 'c'], baz=['d'])),
]
class TestPositionalsNargsNoneOptional1(ParserTestCase):
"""Test three Positionals: no nargs, optional narg and 1 nargs"""
argument_signatures = [
Sig('foo'),
Sig('bar', nargs='?', default=0.625),
Sig('baz', nargs=1),
]
failures = ['', '--foo', 'a']
successes = [
('a b', NS(foo='a', bar=0.625, baz=['b'])),
('a b c', NS(foo='a', bar='b', baz=['c'])),
]
class TestPositionalsNargsOptionalOptional(ParserTestCase):
"""Test two optional nargs"""
argument_signatures = [
Sig('foo', nargs='?'),
Sig('bar', nargs='?', default=42),
]
failures = ['--foo', 'a b c']
successes = [
('', NS(foo=None, bar=42)),
('a', NS(foo='a', bar=42)),
('a b', NS(foo='a', bar='b')),
]
class TestPositionalsNargsOptionalZeroOrMore(ParserTestCase):
"""Test an Optional narg followed by unlimited nargs"""
argument_signatures = [Sig('foo', nargs='?'), Sig('bar', nargs='*')]
failures = ['--foo']
successes = [
('', NS(foo=None, bar=[])),
('a', NS(foo='a', bar=[])),
('a b', NS(foo='a', bar=['b'])),
('a b c', NS(foo='a', bar=['b', 'c'])),
]
class TestPositionalsNargsOptionalOneOrMore(ParserTestCase):
"""Test an Optional narg followed by one or more nargs"""
argument_signatures = [Sig('foo', nargs='?'), Sig('bar', nargs='+')]
failures = ['', '--foo']
successes = [
('a', NS(foo=None, bar=['a'])),
('a b', NS(foo='a', bar=['b'])),
('a b c', NS(foo='a', bar=['b', 'c'])),
]
class TestPositionalsChoicesString(ParserTestCase):
"""Test a set of single-character choices"""
argument_signatures = [Sig('spam', choices=set('abcdefg'))]
failures = ['', '--foo', 'h', '42', 'ef']
successes = [
('a', NS(spam='a')),
('g', NS(spam='g')),
]
class TestPositionalsChoicesInt(ParserTestCase):
"""Test a set of integer choices"""
argument_signatures = [Sig('spam', type=int, choices=range(20))]
failures = ['', '--foo', 'h', '42', 'ef']
successes = [
('4', NS(spam=4)),
('15', NS(spam=15)),
]
class TestPositionalsActionAppend(ParserTestCase):
"""Test the 'append' action"""
argument_signatures = [
Sig('spam', action='append'),
Sig('spam', action='append', nargs=2),
]
failures = ['', '--foo', 'a', 'a b', 'a b c d']
successes = [
('a b c', NS(spam=['a', ['b', 'c']])),
]
# ========================================
# Combined optionals and positionals tests
# ========================================
class TestOptionalsNumericAndPositionals(ParserTestCase):
"""Tests negative number args when numeric options are present"""
argument_signatures = [
Sig('x', nargs='?'),
Sig('-4', dest='y', action='store_true'),
]
failures = ['-2', '-315']
successes = [
('', NS(x=None, y=False)),
('a', NS(x='a', y=False)),
('-4', NS(x=None, y=True)),
('-4 a', NS(x='a', y=True)),
]
class TestOptionalsAlmostNumericAndPositionals(ParserTestCase):
"""Tests negative number args when almost numeric options are present"""
argument_signatures = [
Sig('x', nargs='?'),
Sig('-k4', dest='y', action='store_true'),
]
failures = ['-k3']
successes = [
('', NS(x=None, y=False)),
('-2', NS(x='-2', y=False)),
('a', NS(x='a', y=False)),
('-k4', NS(x=None, y=True)),
('-k4 a', NS(x='a', y=True)),
]
class TestEmptyAndSpaceContainingArguments(ParserTestCase):
argument_signatures = [
Sig('x', nargs='?'),
Sig('-y', '--yyy', dest='y'),
]
failures = ['-y']
successes = [
([''], NS(x='', y=None)),
(['a badger'], NS(x='a badger', y=None)),
(['-a badger'], NS(x='-a badger', y=None)),
(['-y', ''], NS(x=None, y='')),
(['-y', 'a badger'], NS(x=None, y='a badger')),
(['-y', '-a badger'], NS(x=None, y='-a badger')),
(['--yyy=a badger'], NS(x=None, y='a badger')),
(['--yyy=-a badger'], NS(x=None, y='-a badger')),
]
class TestPrefixCharacterOnlyArguments(ParserTestCase):
parser_signature = Sig(prefix_chars='-+')
argument_signatures = [
Sig('-', dest='x', nargs='?', const='badger'),
Sig('+', dest='y', type=int, default=42),
Sig('-+-', dest='z', action='store_true'),
]
failures = ['-y', '+ -']
successes = [
('', NS(x=None, y=42, z=False)),
('-', NS(x='badger', y=42, z=False)),
('- X', NS(x='X', y=42, z=False)),
('+ -3', NS(x=None, y=-3, z=False)),
('-+-', NS(x=None, y=42, z=True)),
('- ===', NS(x='===', y=42, z=False)),
]
class TestNargsZeroOrMore(ParserTestCase):
"""Tests specifying args for an Optional that accepts zero or more"""
argument_signatures = [Sig('-x', nargs='*'), Sig('y', nargs='*')]
failures = []
successes = [
('', NS(x=None, y=[])),
('-x', NS(x=[], y=[])),
('-x a', NS(x=['a'], y=[])),
('-x a -- b', NS(x=['a'], y=['b'])),
('a', NS(x=None, y=['a'])),
('a -x', NS(x=[], y=['a'])),
('a -x b', NS(x=['b'], y=['a'])),
]
class TestNargsRemainder(ParserTestCase):
"""Tests specifying a positional with nargs=REMAINDER"""
argument_signatures = [Sig('x'), Sig('y', nargs='...'), Sig('-z')]
failures = ['', '-z', '-z Z']
successes = [
('X', NS(x='X', y=[], z=None)),
('-z Z X', NS(x='X', y=[], z='Z')),
('X A B -z Z', NS(x='X', y=['A', 'B', '-z', 'Z'], z=None)),
('X Y --foo', NS(x='X', y=['Y', '--foo'], z=None)),
]
class TestOptionLike(ParserTestCase):
"""Tests options that may or may not be arguments"""
argument_signatures = [
Sig('-x', type=float),
Sig('-3', type=float, dest='y'),
Sig('z', nargs='*'),
]
failures = ['-x', '-y2.5', '-xa', '-x -a',
'-x -3', '-x -3.5', '-3 -3.5',
'-x -2.5', '-x -2.5 a', '-3 -.5',
'a x -1', '-x -1 a', '-3 -1 a']
successes = [
('', NS(x=None, y=None, z=[])),
('-x 2.5', NS(x=2.5, y=None, z=[])),
('-x 2.5 a', NS(x=2.5, y=None, z=['a'])),
('-3.5', NS(x=None, y=0.5, z=[])),
('-3-.5', NS(x=None, y=-0.5, z=[])),
('-3 .5', NS(x=None, y=0.5, z=[])),
('a -3.5', NS(x=None, y=0.5, z=['a'])),
('a', NS(x=None, y=None, z=['a'])),
('a -x 1', NS(x=1.0, y=None, z=['a'])),
('-x 1 a', NS(x=1.0, y=None, z=['a'])),
('-3 1 a', NS(x=None, y=1.0, z=['a'])),
]
class TestDefaultSuppress(ParserTestCase):
"""Test actions with suppressed defaults"""
argument_signatures = [
Sig('foo', nargs='?', default=argparse.SUPPRESS),
Sig('bar', nargs='*', default=argparse.SUPPRESS),
Sig('--baz', action='store_true', default=argparse.SUPPRESS),
]
failures = ['-x']
successes = [
('', NS()),
('a', NS(foo='a')),
('a b', NS(foo='a', bar=['b'])),
('--baz', NS(baz=True)),
('a --baz', NS(foo='a', baz=True)),
('--baz a b', NS(foo='a', bar=['b'], baz=True)),
]
class TestParserDefaultSuppress(ParserTestCase):
"""Test actions with a parser-level default of SUPPRESS"""
parser_signature = Sig(argument_default=argparse.SUPPRESS)
argument_signatures = [
Sig('foo', nargs='?'),
Sig('bar', nargs='*'),
Sig('--baz', action='store_true'),
]
failures = ['-x']
successes = [
('', NS()),
('a', NS(foo='a')),
('a b', NS(foo='a', bar=['b'])),
('--baz', NS(baz=True)),
('a --baz', NS(foo='a', baz=True)),
('--baz a b', NS(foo='a', bar=['b'], baz=True)),
]
class TestParserDefault42(ParserTestCase):
"""Test actions with a parser-level default of 42"""
parser_signature = Sig(argument_default=42)
argument_signatures = [
Sig('--version', action='version', version='1.0'),
Sig('foo', nargs='?'),
Sig('bar', nargs='*'),
Sig('--baz', action='store_true'),
]
failures = ['-x']
successes = [
('', NS(foo=42, bar=42, baz=42, version=42)),
('a', NS(foo='a', bar=42, baz=42, version=42)),
('a b', NS(foo='a', bar=['b'], baz=42, version=42)),
('--baz', NS(foo=42, bar=42, baz=True, version=42)),
('a --baz', NS(foo='a', bar=42, baz=True, version=42)),
('--baz a b', NS(foo='a', bar=['b'], baz=True, version=42)),
]
class TestArgumentsFromFile(TempDirMixin, ParserTestCase):
"""Test reading arguments from a file"""
def setUp(self):
super(TestArgumentsFromFile, self).setUp()
file_texts = [
('hello', 'hello world!\n'),
('recursive', '-a\n'
'A\n'
'@hello'),
('invalid', '@no-such-path\n'),
]
for path, text in file_texts:
with open(path, 'w') as file:
file.write(text)
parser_signature = Sig(fromfile_prefix_chars='@')
argument_signatures = [
Sig('-a'),
Sig('x'),
Sig('y', nargs='+'),
]
failures = ['', '-b', 'X', '@invalid', '@missing']
successes = [
('X Y', NS(a=None, x='X', y=['Y'])),
('X -a A Y Z', NS(a='A', x='X', y=['Y', 'Z'])),
('@hello X', NS(a=None, x='hello world!', y=['X'])),
('X @hello', NS(a=None, x='X', y=['hello world!'])),
('-a B @recursive Y Z', NS(a='A', x='hello world!', y=['Y', 'Z'])),
('X @recursive Z -a B', NS(a='B', x='X', y=['hello world!', 'Z'])),
(["-a", "", "X", "Y"], NS(a='', x='X', y=['Y'])),
]
class TestArgumentsFromFileConverter(TempDirMixin, ParserTestCase):
"""Test reading arguments from a file"""
def setUp(self):
super(TestArgumentsFromFileConverter, self).setUp()
file_texts = [
('hello', 'hello world!\n'),
]
for path, text in file_texts:
with open(path, 'w') as file:
file.write(text)
class FromFileConverterArgumentParser(ErrorRaisingArgumentParser):
def convert_arg_line_to_args(self, arg_line):
for arg in arg_line.split():
if not arg.strip():
continue
yield arg
parser_class = FromFileConverterArgumentParser
parser_signature = Sig(fromfile_prefix_chars='@')
argument_signatures = [
Sig('y', nargs='+'),
]
failures = []
successes = [
('@hello X', NS(y=['hello', 'world!', 'X'])),
]
# =====================
# Type conversion tests
# =====================
class TestFileTypeRepr(TestCase):
def test_r(self):
type = argparse.FileType('r')
self.assertEqual("FileType('r')", repr(type))
def test_wb_1(self):
type = argparse.FileType('wb', 1)
self.assertEqual("FileType('wb', 1)", repr(type))
def test_r_latin(self):
type = argparse.FileType('r', encoding='latin_1')
self.assertEqual("FileType('r', encoding='latin_1')", repr(type))
def test_w_big5_ignore(self):
type = argparse.FileType('w', encoding='big5', errors='ignore')
self.assertEqual("FileType('w', encoding='big5', errors='ignore')",
repr(type))
def test_r_1_replace(self):
type = argparse.FileType('r', 1, errors='replace')
self.assertEqual("FileType('r', 1, errors='replace')", repr(type))
class StdStreamComparer:
def __init__(self, attr):
self.attr = attr
def __eq__(self, other):
return other == getattr(sys, self.attr)
eq_stdin = StdStreamComparer('stdin')
eq_stdout = StdStreamComparer('stdout')
eq_stderr = StdStreamComparer('stderr')
class RFile(object):
seen = {}
def __init__(self, name):
self.name = name
def __eq__(self, other):
if other in self.seen:
text = self.seen[other]
else:
text = self.seen[other] = other.read()
other.close()
if not isinstance(text, str):
text = text.decode('ascii')
return self.name == other.name == text
class TestFileTypeR(TempDirMixin, ParserTestCase):
"""Test the FileType option/argument type for reading files"""
def setUp(self):
super(TestFileTypeR, self).setUp()
for file_name in ['foo', 'bar']:
with open(os.path.join(self.temp_dir, file_name), 'w') as file:
file.write(file_name)
self.create_readonly_file('readonly')
argument_signatures = [
Sig('-x', type=argparse.FileType()),
Sig('spam', type=argparse.FileType('r')),
]
failures = ['-x', '', 'non-existent-file.txt']
successes = [
('foo', NS(x=None, spam=RFile('foo'))),
('-x foo bar', NS(x=RFile('foo'), spam=RFile('bar'))),
('bar -x foo', NS(x=RFile('foo'), spam=RFile('bar'))),
('-x - -', NS(x=eq_stdin, spam=eq_stdin)),
('readonly', NS(x=None, spam=RFile('readonly'))),
]
class TestFileTypeDefaults(TempDirMixin, ParserTestCase):
"""Test that a file is not created unless the default is needed"""
def setUp(self):
super(TestFileTypeDefaults, self).setUp()
file = open(os.path.join(self.temp_dir, 'good'), 'w')
file.write('good')
file.close()
argument_signatures = [
Sig('-c', type=argparse.FileType('r'), default='no-file.txt'),
]
# should provoke no such file error
failures = ['']
# should not provoke error because default file is created
successes = [('-c good', NS(c=RFile('good')))]
class TestFileTypeRB(TempDirMixin, ParserTestCase):
"""Test the FileType option/argument type for reading files"""
def setUp(self):
super(TestFileTypeRB, self).setUp()
for file_name in ['foo', 'bar']:
with open(os.path.join(self.temp_dir, file_name), 'w') as file:
file.write(file_name)
argument_signatures = [
Sig('-x', type=argparse.FileType('rb')),
Sig('spam', type=argparse.FileType('rb')),
]
failures = ['-x', '']
successes = [
('foo', NS(x=None, spam=RFile('foo'))),
('-x foo bar', NS(x=RFile('foo'), spam=RFile('bar'))),
('bar -x foo', NS(x=RFile('foo'), spam=RFile('bar'))),
('-x - -', NS(x=eq_stdin, spam=eq_stdin)),
]
class WFile(object):
seen = set()
def __init__(self, name):
self.name = name
def __eq__(self, other):
if other not in self.seen:
text = 'Check that file is writable.'
if 'b' in other.mode:
text = text.encode('ascii')
other.write(text)
other.close()
self.seen.add(other)
return self.name == other.name
@unittest.skipIf(hasattr(os, 'geteuid') and os.geteuid() == 0,
"non-root user required")
class TestFileTypeW(TempDirMixin, ParserTestCase):
"""Test the FileType option/argument type for writing files"""
def setUp(self):
super(TestFileTypeW, self).setUp()
self.create_readonly_file('readonly')
argument_signatures = [
Sig('-x', type=argparse.FileType('w')),
Sig('spam', type=argparse.FileType('w')),
]
failures = ['-x', '', 'readonly']
successes = [
('foo', NS(x=None, spam=WFile('foo'))),
('-x foo bar', NS(x=WFile('foo'), spam=WFile('bar'))),
('bar -x foo', NS(x=WFile('foo'), spam=WFile('bar'))),
('-x - -', NS(x=eq_stdout, spam=eq_stdout)),
]
class TestFileTypeWB(TempDirMixin, ParserTestCase):
argument_signatures = [
Sig('-x', type=argparse.FileType('wb')),
Sig('spam', type=argparse.FileType('wb')),
]
failures = ['-x', '']
successes = [
('foo', NS(x=None, spam=WFile('foo'))),
('-x foo bar', NS(x=WFile('foo'), spam=WFile('bar'))),
('bar -x foo', NS(x=WFile('foo'), spam=WFile('bar'))),
('-x - -', NS(x=eq_stdout, spam=eq_stdout)),
]
class TestFileTypeOpenArgs(TestCase):
"""Test that open (the builtin) is correctly called"""
def test_open_args(self):
FT = argparse.FileType
cases = [
(FT('rb'), ('rb', -1, None, None)),
(FT('w', 1), ('w', 1, None, None)),
(FT('w', errors='replace'), ('w', -1, None, 'replace')),
(FT('wb', encoding='big5'), ('wb', -1, 'big5', None)),
(FT('w', 0, 'l1', 'strict'), ('w', 0, 'l1', 'strict')),
]
with mock.patch('builtins.open') as m:
for type, args in cases:
type('foo')
m.assert_called_with('foo', *args)
class TestFileTypeMissingInitialization(TestCase):
"""
Test that add_argument throws an error if FileType class
object was passed instead of instance of FileType
"""
def test(self):
parser = argparse.ArgumentParser()
with self.assertRaises(ValueError) as cm:
parser.add_argument('-x', type=argparse.FileType)
self.assertEqual(
'%r is a FileType class object, instance of it must be passed'
% (argparse.FileType,),
str(cm.exception)
)
class TestTypeCallable(ParserTestCase):
"""Test some callables as option/argument types"""
argument_signatures = [
Sig('--eggs', type=complex),
Sig('spam', type=float),
]
failures = ['a', '42j', '--eggs a', '--eggs 2i']
successes = [
('--eggs=42 42', NS(eggs=42, spam=42.0)),
('--eggs 2j -- -1.5', NS(eggs=2j, spam=-1.5)),
('1024.675', NS(eggs=None, spam=1024.675)),
]
class TestTypeUserDefined(ParserTestCase):
"""Test a user-defined option/argument type"""
class MyType(TestCase):
def __init__(self, value):
self.value = value
def __eq__(self, other):
return (type(self), self.value) == (type(other), other.value)
argument_signatures = [
Sig('-x', type=MyType),
Sig('spam', type=MyType),
]
failures = []
successes = [
('a -x b', NS(x=MyType('b'), spam=MyType('a'))),
('-xf g', NS(x=MyType('f'), spam=MyType('g'))),
]
class TestTypeClassicClass(ParserTestCase):
"""Test a classic class type"""
class C:
def __init__(self, value):
self.value = value
def __eq__(self, other):
return (type(self), self.value) == (type(other), other.value)
argument_signatures = [
Sig('-x', type=C),
Sig('spam', type=C),
]
failures = []
successes = [
('a -x b', NS(x=C('b'), spam=C('a'))),
('-xf g', NS(x=C('f'), spam=C('g'))),
]
class TestTypeRegistration(TestCase):
"""Test a user-defined type by registering it"""
def test(self):
def get_my_type(string):
return 'my_type{%s}' % string
parser = argparse.ArgumentParser()
parser.register('type', 'my_type', get_my_type)
parser.add_argument('-x', type='my_type')
parser.add_argument('y', type='my_type')
self.assertEqual(parser.parse_args('1'.split()),
NS(x=None, y='my_type{1}'))
self.assertEqual(parser.parse_args('-x 1 42'.split()),
NS(x='my_type{1}', y='my_type{42}'))
# ============
# Action tests
# ============
class TestActionUserDefined(ParserTestCase):
"""Test a user-defined option/argument action"""
class OptionalAction(argparse.Action):
def __call__(self, parser, namespace, value, option_string=None):
try:
# check destination and option string
assert self.dest == 'spam', 'dest: %s' % self.dest
assert option_string == '-s', 'flag: %s' % option_string
# when option is before argument, badger=2, and when
# option is after argument, badger=<whatever was set>
expected_ns = NS(spam=0.25)
if value in [0.125, 0.625]:
expected_ns.badger = 2
elif value in [2.0]:
expected_ns.badger = 84
else:
raise AssertionError('value: %s' % value)
assert expected_ns == namespace, ('expected %s, got %s' %
(expected_ns, namespace))
except AssertionError:
e = sys.exc_info()[1]
raise ArgumentParserError('opt_action failed: %s' % e)
setattr(namespace, 'spam', value)
class PositionalAction(argparse.Action):
def __call__(self, parser, namespace, value, option_string=None):
try:
assert option_string is None, ('option_string: %s' %
option_string)
# check destination
assert self.dest == 'badger', 'dest: %s' % self.dest
# when argument is before option, spam=0.25, and when
# option is after argument, spam=<whatever was set>
expected_ns = NS(badger=2)
if value in [42, 84]:
expected_ns.spam = 0.25
elif value in [1]:
expected_ns.spam = 0.625
elif value in [2]:
expected_ns.spam = 0.125
else:
raise AssertionError('value: %s' % value)
assert expected_ns == namespace, ('expected %s, got %s' %
(expected_ns, namespace))
except AssertionError:
e = sys.exc_info()[1]
raise ArgumentParserError('arg_action failed: %s' % e)
setattr(namespace, 'badger', value)
argument_signatures = [
Sig('-s', dest='spam', action=OptionalAction,
type=float, default=0.25),
Sig('badger', action=PositionalAction,
type=int, nargs='?', default=2),
]
failures = []
successes = [
('-s0.125', NS(spam=0.125, badger=2)),
('42', NS(spam=0.25, badger=42)),
('-s 0.625 1', NS(spam=0.625, badger=1)),
('84 -s2', NS(spam=2.0, badger=84)),
]
class TestActionRegistration(TestCase):
"""Test a user-defined action supplied by registering it"""
class MyAction(argparse.Action):
def __call__(self, parser, namespace, values, option_string=None):
setattr(namespace, self.dest, 'foo[%s]' % values)
def test(self):
parser = argparse.ArgumentParser()
parser.register('action', 'my_action', self.MyAction)
parser.add_argument('badger', action='my_action')
self.assertEqual(parser.parse_args(['1']), NS(badger='foo[1]'))
self.assertEqual(parser.parse_args(['42']), NS(badger='foo[42]'))
class TestActionExtend(ParserTestCase):
argument_signatures = [
Sig('--foo', action="extend", nargs="+", type=str),
]
failures = ()
successes = [
('--foo f1 --foo f2 f3 f4', NS(foo=['f1', 'f2', 'f3', 'f4'])),
]
# ================
# Subparsers tests
# ================
class TestAddSubparsers(TestCase):
"""Test the add_subparsers method"""
def assertArgumentParserError(self, *args, **kwargs):
self.assertRaises(ArgumentParserError, *args, **kwargs)
def _get_parser(self, subparser_help=False, prefix_chars=None,
aliases=False):
# create a parser with a subparsers argument
if prefix_chars:
parser = ErrorRaisingArgumentParser(
prog='PROG', description='main description', prefix_chars=prefix_chars)
parser.add_argument(
prefix_chars[0] * 2 + 'foo', action='store_true', help='foo help')
else:
parser = ErrorRaisingArgumentParser(
prog='PROG', description='main description')
parser.add_argument(
'--foo', action='store_true', help='foo help')
parser.add_argument(
'bar', type=float, help='bar help')
# check that only one subparsers argument can be added
subparsers_kwargs = {'required': False}
if aliases:
subparsers_kwargs['metavar'] = 'COMMAND'
subparsers_kwargs['title'] = 'commands'
else:
subparsers_kwargs['help'] = 'command help'
subparsers = parser.add_subparsers(**subparsers_kwargs)
self.assertArgumentParserError(parser.add_subparsers)
# add first sub-parser
parser1_kwargs = dict(description='1 description')
if subparser_help:
parser1_kwargs['help'] = '1 help'
if aliases:
parser1_kwargs['aliases'] = ['1alias1', '1alias2']
parser1 = subparsers.add_parser('1', **parser1_kwargs)
parser1.add_argument('-w', type=int, help='w help')
parser1.add_argument('x', choices='abc', help='x help')
# add second sub-parser
parser2_kwargs = dict(description='2 description')
if subparser_help:
parser2_kwargs['help'] = '2 help'
parser2 = subparsers.add_parser('2', **parser2_kwargs)
parser2.add_argument('-y', choices='123', help='y help')
parser2.add_argument('z', type=complex, nargs='*', help='z help')
# add third sub-parser
parser3_kwargs = dict(description='3 description')
if subparser_help:
parser3_kwargs['help'] = '3 help'
parser3 = subparsers.add_parser('3', **parser3_kwargs)
parser3.add_argument('t', type=int, help='t help')
parser3.add_argument('u', nargs='...', help='u help')
# return the main parser
return parser
def setUp(self):
super().setUp()
self.parser = self._get_parser()
self.command_help_parser = self._get_parser(subparser_help=True)
def test_parse_args_failures(self):
# check some failure cases:
for args_str in ['', 'a', 'a a', '0.5 a', '0.5 1',
'0.5 1 -y', '0.5 2 -w']:
args = args_str.split()
self.assertArgumentParserError(self.parser.parse_args, args)
def test_parse_args(self):
# check some non-failure cases:
self.assertEqual(
self.parser.parse_args('0.5 1 b -w 7'.split()),
NS(foo=False, bar=0.5, w=7, x='b'),
)
self.assertEqual(
self.parser.parse_args('0.25 --foo 2 -y 2 3j -- -1j'.split()),
NS(foo=True, bar=0.25, y='2', z=[3j, -1j]),
)
self.assertEqual(
self.parser.parse_args('--foo 0.125 1 c'.split()),
NS(foo=True, bar=0.125, w=None, x='c'),
)
self.assertEqual(
self.parser.parse_args('-1.5 3 11 -- a --foo 7 -- b'.split()),
NS(foo=False, bar=-1.5, t=11, u=['a', '--foo', '7', '--', 'b']),
)
def test_parse_known_args(self):
self.assertEqual(
self.parser.parse_known_args('0.5 1 b -w 7'.split()),
(NS(foo=False, bar=0.5, w=7, x='b'), []),
)
self.assertEqual(
self.parser.parse_known_args('0.5 -p 1 b -w 7'.split()),
(NS(foo=False, bar=0.5, w=7, x='b'), ['-p']),
)
self.assertEqual(
self.parser.parse_known_args('0.5 1 b -w 7 -p'.split()),
(NS(foo=False, bar=0.5, w=7, x='b'), ['-p']),
)
self.assertEqual(
self.parser.parse_known_args('0.5 1 b -q -rs -w 7'.split()),
(NS(foo=False, bar=0.5, w=7, x='b'), ['-q', '-rs']),
)
self.assertEqual(
self.parser.parse_known_args('0.5 -W 1 b -X Y -w 7 Z'.split()),
(NS(foo=False, bar=0.5, w=7, x='b'), ['-W', '-X', 'Y', 'Z']),
)
def test_dest(self):
parser = ErrorRaisingArgumentParser()
parser.add_argument('--foo', action='store_true')
subparsers = parser.add_subparsers(dest='bar')
parser1 = subparsers.add_parser('1')
parser1.add_argument('baz')
self.assertEqual(NS(foo=False, bar='1', baz='2'),
parser.parse_args('1 2'.split()))
def _test_required_subparsers(self, parser):
# Should parse the sub command
ret = parser.parse_args(['run'])
self.assertEqual(ret.command, 'run')
# Error when the command is missing
self.assertArgumentParserError(parser.parse_args, ())
def test_required_subparsers_via_attribute(self):
parser = ErrorRaisingArgumentParser()
subparsers = parser.add_subparsers(dest='command')
subparsers.required = True
subparsers.add_parser('run')
self._test_required_subparsers(parser)
def test_required_subparsers_via_kwarg(self):
parser = ErrorRaisingArgumentParser()
subparsers = parser.add_subparsers(dest='command', required=True)
subparsers.add_parser('run')
self._test_required_subparsers(parser)
def test_required_subparsers_default(self):
parser = ErrorRaisingArgumentParser()
subparsers = parser.add_subparsers(dest='command')
subparsers.add_parser('run')
# No error here
ret = parser.parse_args(())
self.assertIsNone(ret.command)
def test_optional_subparsers(self):
parser = ErrorRaisingArgumentParser()
subparsers = parser.add_subparsers(dest='command', required=False)
subparsers.add_parser('run')
# No error here
ret = parser.parse_args(())
self.assertIsNone(ret.command)
def test_help(self):
self.assertEqual(self.parser.format_usage(),
'usage: PROG [-h] [--foo] bar {1,2,3} ...\n')
self.assertEqual(self.parser.format_help(), textwrap.dedent('''\
usage: PROG [-h] [--foo] bar {1,2,3} ...
main description
positional arguments:
bar bar help
{1,2,3} command help
optional arguments:
-h, --help show this help message and exit
--foo foo help
'''))
def test_help_extra_prefix_chars(self):
# Make sure - is still used for help if it is a non-first prefix char
parser = self._get_parser(prefix_chars='+:-')
self.assertEqual(parser.format_usage(),
'usage: PROG [-h] [++foo] bar {1,2,3} ...\n')
self.assertEqual(parser.format_help(), textwrap.dedent('''\
usage: PROG [-h] [++foo] bar {1,2,3} ...
main description
positional arguments:
bar bar help
{1,2,3} command help
optional arguments:
-h, --help show this help message and exit
++foo foo help
'''))
def test_help_non_breaking_spaces(self):
parser = ErrorRaisingArgumentParser(
prog='PROG', description='main description')
parser.add_argument(
"--non-breaking", action='store_false',
help='help message containing non-breaking spaces shall not '
'wrap\N{NO-BREAK SPACE}at non-breaking spaces')
self.assertEqual(parser.format_help(), textwrap.dedent('''\
usage: PROG [-h] [--non-breaking]
main description
optional arguments:
-h, --help show this help message and exit
--non-breaking help message containing non-breaking spaces shall not
wrap\N{NO-BREAK SPACE}at non-breaking spaces
'''))
def test_help_alternate_prefix_chars(self):
parser = self._get_parser(prefix_chars='+:/')
self.assertEqual(parser.format_usage(),
'usage: PROG [+h] [++foo] bar {1,2,3} ...\n')
self.assertEqual(parser.format_help(), textwrap.dedent('''\
usage: PROG [+h] [++foo] bar {1,2,3} ...
main description
positional arguments:
bar bar help
{1,2,3} command help
optional arguments:
+h, ++help show this help message and exit
++foo foo help
'''))
def test_parser_command_help(self):
self.assertEqual(self.command_help_parser.format_usage(),
'usage: PROG [-h] [--foo] bar {1,2,3} ...\n')
self.assertEqual(self.command_help_parser.format_help(),
textwrap.dedent('''\
usage: PROG [-h] [--foo] bar {1,2,3} ...
main description
positional arguments:
bar bar help
{1,2,3} command help
1 1 help
2 2 help
3 3 help
optional arguments:
-h, --help show this help message and exit
--foo foo help
'''))
def test_subparser_title_help(self):
parser = ErrorRaisingArgumentParser(prog='PROG',
description='main description')
parser.add_argument('--foo', action='store_true', help='foo help')
parser.add_argument('bar', help='bar help')
subparsers = parser.add_subparsers(title='subcommands',
description='command help',
help='additional text')
parser1 = subparsers.add_parser('1')
parser2 = subparsers.add_parser('2')
self.assertEqual(parser.format_usage(),
'usage: PROG [-h] [--foo] bar {1,2} ...\n')
self.assertEqual(parser.format_help(), textwrap.dedent('''\
usage: PROG [-h] [--foo] bar {1,2} ...
main description
positional arguments:
bar bar help
optional arguments:
-h, --help show this help message and exit
--foo foo help
subcommands:
command help
{1,2} additional text
'''))
def _test_subparser_help(self, args_str, expected_help):
with self.assertRaises(ArgumentParserError) as cm:
self.parser.parse_args(args_str.split())
self.assertEqual(expected_help, cm.exception.stdout)
def test_subparser1_help(self):
self._test_subparser_help('5.0 1 -h', textwrap.dedent('''\
usage: PROG bar 1 [-h] [-w W] {a,b,c}
1 description
positional arguments:
{a,b,c} x help
optional arguments:
-h, --help show this help message and exit
-w W w help
'''))
def test_subparser2_help(self):
self._test_subparser_help('5.0 2 -h', textwrap.dedent('''\
usage: PROG bar 2 [-h] [-y {1,2,3}] [z ...]
2 description
positional arguments:
z z help
optional arguments:
-h, --help show this help message and exit
-y {1,2,3} y help
'''))
def test_alias_invocation(self):
parser = self._get_parser(aliases=True)
self.assertEqual(
parser.parse_known_args('0.5 1alias1 b'.split()),
(NS(foo=False, bar=0.5, w=None, x='b'), []),
)
self.assertEqual(
parser.parse_known_args('0.5 1alias2 b'.split()),
(NS(foo=False, bar=0.5, w=None, x='b'), []),
)
def test_error_alias_invocation(self):
parser = self._get_parser(aliases=True)
self.assertArgumentParserError(parser.parse_args,
'0.5 1alias3 b'.split())
def test_alias_help(self):
parser = self._get_parser(aliases=True, subparser_help=True)
self.maxDiff = None
self.assertEqual(parser.format_help(), textwrap.dedent("""\
usage: PROG [-h] [--foo] bar COMMAND ...
main description
positional arguments:
bar bar help
optional arguments:
-h, --help show this help message and exit
--foo foo help
commands:
COMMAND
1 (1alias1, 1alias2)
1 help
2 2 help
3 3 help
"""))
# ============
# Groups tests
# ============
class TestPositionalsGroups(TestCase):
"""Tests that order of group positionals matches construction order"""
def test_nongroup_first(self):
parser = ErrorRaisingArgumentParser()
parser.add_argument('foo')
group = parser.add_argument_group('g')
group.add_argument('bar')
parser.add_argument('baz')
expected = NS(foo='1', bar='2', baz='3')
result = parser.parse_args('1 2 3'.split())
self.assertEqual(expected, result)
def test_group_first(self):
parser = ErrorRaisingArgumentParser()
group = parser.add_argument_group('xxx')
group.add_argument('foo')
parser.add_argument('bar')
parser.add_argument('baz')
expected = NS(foo='1', bar='2', baz='3')
result = parser.parse_args('1 2 3'.split())
self.assertEqual(expected, result)
def test_interleaved_groups(self):
parser = ErrorRaisingArgumentParser()
group = parser.add_argument_group('xxx')
parser.add_argument('foo')
group.add_argument('bar')
parser.add_argument('baz')
group = parser.add_argument_group('yyy')
group.add_argument('frell')
expected = NS(foo='1', bar='2', baz='3', frell='4')
result = parser.parse_args('1 2 3 4'.split())
self.assertEqual(expected, result)
# ===================
# Parent parser tests
# ===================
class TestParentParsers(TestCase):
"""Tests that parsers can be created with parent parsers"""
def assertArgumentParserError(self, *args, **kwargs):
self.assertRaises(ArgumentParserError, *args, **kwargs)
def setUp(self):
super().setUp()
self.wxyz_parent = ErrorRaisingArgumentParser(add_help=False)
self.wxyz_parent.add_argument('--w')
x_group = self.wxyz_parent.add_argument_group('x')
x_group.add_argument('-y')
self.wxyz_parent.add_argument('z')
self.abcd_parent = ErrorRaisingArgumentParser(add_help=False)
self.abcd_parent.add_argument('a')
self.abcd_parent.add_argument('-b')
c_group = self.abcd_parent.add_argument_group('c')
c_group.add_argument('--d')
self.w_parent = ErrorRaisingArgumentParser(add_help=False)
self.w_parent.add_argument('--w')
self.z_parent = ErrorRaisingArgumentParser(add_help=False)
self.z_parent.add_argument('z')
# parents with mutually exclusive groups
self.ab_mutex_parent = ErrorRaisingArgumentParser(add_help=False)
group = self.ab_mutex_parent.add_mutually_exclusive_group()
group.add_argument('-a', action='store_true')
group.add_argument('-b', action='store_true')
self.main_program = os.path.basename(sys.argv[0])
def test_single_parent(self):
parser = ErrorRaisingArgumentParser(parents=[self.wxyz_parent])
self.assertEqual(parser.parse_args('-y 1 2 --w 3'.split()),
NS(w='3', y='1', z='2'))
def test_single_parent_mutex(self):
self._test_mutex_ab(self.ab_mutex_parent.parse_args)
parser = ErrorRaisingArgumentParser(parents=[self.ab_mutex_parent])
self._test_mutex_ab(parser.parse_args)
def test_single_granparent_mutex(self):
parents = [self.ab_mutex_parent]
parser = ErrorRaisingArgumentParser(add_help=False, parents=parents)
parser = ErrorRaisingArgumentParser(parents=[parser])
self._test_mutex_ab(parser.parse_args)
def _test_mutex_ab(self, parse_args):
self.assertEqual(parse_args([]), NS(a=False, b=False))
self.assertEqual(parse_args(['-a']), NS(a=True, b=False))
self.assertEqual(parse_args(['-b']), NS(a=False, b=True))
self.assertArgumentParserError(parse_args, ['-a', '-b'])
self.assertArgumentParserError(parse_args, ['-b', '-a'])
self.assertArgumentParserError(parse_args, ['-c'])
self.assertArgumentParserError(parse_args, ['-a', '-c'])
self.assertArgumentParserError(parse_args, ['-b', '-c'])
def test_multiple_parents(self):
parents = [self.abcd_parent, self.wxyz_parent]
parser = ErrorRaisingArgumentParser(parents=parents)
self.assertEqual(parser.parse_args('--d 1 --w 2 3 4'.split()),
NS(a='3', b=None, d='1', w='2', y=None, z='4'))
def test_multiple_parents_mutex(self):
parents = [self.ab_mutex_parent, self.wxyz_parent]
parser = ErrorRaisingArgumentParser(parents=parents)
self.assertEqual(parser.parse_args('-a --w 2 3'.split()),
NS(a=True, b=False, w='2', y=None, z='3'))
self.assertArgumentParserError(
parser.parse_args, '-a --w 2 3 -b'.split())
self.assertArgumentParserError(
parser.parse_args, '-a -b --w 2 3'.split())
def test_conflicting_parents(self):
self.assertRaises(
argparse.ArgumentError,
argparse.ArgumentParser,
parents=[self.w_parent, self.wxyz_parent])
def test_conflicting_parents_mutex(self):
self.assertRaises(
argparse.ArgumentError,
argparse.ArgumentParser,
parents=[self.abcd_parent, self.ab_mutex_parent])
def test_same_argument_name_parents(self):
parents = [self.wxyz_parent, self.z_parent]
parser = ErrorRaisingArgumentParser(parents=parents)
self.assertEqual(parser.parse_args('1 2'.split()),
NS(w=None, y=None, z='2'))
def test_subparser_parents(self):
parser = ErrorRaisingArgumentParser()
subparsers = parser.add_subparsers()
abcde_parser = subparsers.add_parser('bar', parents=[self.abcd_parent])
abcde_parser.add_argument('e')
self.assertEqual(parser.parse_args('bar -b 1 --d 2 3 4'.split()),
NS(a='3', b='1', d='2', e='4'))
def test_subparser_parents_mutex(self):
parser = ErrorRaisingArgumentParser()
subparsers = parser.add_subparsers()
parents = [self.ab_mutex_parent]
abc_parser = subparsers.add_parser('foo', parents=parents)
c_group = abc_parser.add_argument_group('c_group')
c_group.add_argument('c')
parents = [self.wxyz_parent, self.ab_mutex_parent]
wxyzabe_parser = subparsers.add_parser('bar', parents=parents)
wxyzabe_parser.add_argument('e')
self.assertEqual(parser.parse_args('foo -a 4'.split()),
NS(a=True, b=False, c='4'))
self.assertEqual(parser.parse_args('bar -b --w 2 3 4'.split()),
NS(a=False, b=True, w='2', y=None, z='3', e='4'))
self.assertArgumentParserError(
parser.parse_args, 'foo -a -b 4'.split())
self.assertArgumentParserError(
parser.parse_args, 'bar -b -a 4'.split())
def test_parent_help(self):
parents = [self.abcd_parent, self.wxyz_parent]
parser = ErrorRaisingArgumentParser(parents=parents)
parser_help = parser.format_help()
progname = self.main_program
self.assertEqual(parser_help, textwrap.dedent('''\
usage: {}{}[-h] [-b B] [--d D] [--w W] [-y Y] a z
positional arguments:
a
z
optional arguments:
-h, --help show this help message and exit
-b B
--w W
c:
--d D
x:
-y Y
'''.format(progname, ' ' if progname else '' )))
def test_groups_parents(self):
parent = ErrorRaisingArgumentParser(add_help=False)
g = parent.add_argument_group(title='g', description='gd')
g.add_argument('-w')
g.add_argument('-x')
m = parent.add_mutually_exclusive_group()
m.add_argument('-y')
m.add_argument('-z')
parser = ErrorRaisingArgumentParser(parents=[parent])
self.assertRaises(ArgumentParserError, parser.parse_args,
['-y', 'Y', '-z', 'Z'])
parser_help = parser.format_help()
progname = self.main_program
self.assertEqual(parser_help, textwrap.dedent('''\
usage: {}{}[-h] [-w W] [-x X] [-y Y | -z Z]
optional arguments:
-h, --help show this help message and exit
-y Y
-z Z
g:
gd
-w W
-x X
'''.format(progname, ' ' if progname else '' )))
# ==============================
# Mutually exclusive group tests
# ==============================
class TestMutuallyExclusiveGroupErrors(TestCase):
def test_invalid_add_argument_group(self):
parser = ErrorRaisingArgumentParser()
raises = self.assertRaises
raises(TypeError, parser.add_mutually_exclusive_group, title='foo')
def test_invalid_add_argument(self):
parser = ErrorRaisingArgumentParser()
group = parser.add_mutually_exclusive_group()
add_argument = group.add_argument
raises = self.assertRaises
raises(ValueError, add_argument, '--foo', required=True)
raises(ValueError, add_argument, 'bar')
raises(ValueError, add_argument, 'bar', nargs='+')
raises(ValueError, add_argument, 'bar', nargs=1)
raises(ValueError, add_argument, 'bar', nargs=argparse.PARSER)
def test_help(self):
parser = ErrorRaisingArgumentParser(prog='PROG')
group1 = parser.add_mutually_exclusive_group()
group1.add_argument('--foo', action='store_true')
group1.add_argument('--bar', action='store_false')
group2 = parser.add_mutually_exclusive_group()
group2.add_argument('--soup', action='store_true')
group2.add_argument('--nuts', action='store_false')
expected = '''\
usage: PROG [-h] [--foo | --bar] [--soup | --nuts]
optional arguments:
-h, --help show this help message and exit
--foo
--bar
--soup
--nuts
'''
self.assertEqual(parser.format_help(), textwrap.dedent(expected))
class MEMixin(object):
def test_failures_when_not_required(self):
parse_args = self.get_parser(required=False).parse_args
error = ArgumentParserError
for args_string in self.failures:
self.assertRaises(error, parse_args, args_string.split())
def test_failures_when_required(self):
parse_args = self.get_parser(required=True).parse_args
error = ArgumentParserError
for args_string in self.failures + ['']:
self.assertRaises(error, parse_args, args_string.split())
def test_successes_when_not_required(self):
parse_args = self.get_parser(required=False).parse_args
successes = self.successes + self.successes_when_not_required
for args_string, expected_ns in successes:
actual_ns = parse_args(args_string.split())
self.assertEqual(actual_ns, expected_ns)
def test_successes_when_required(self):
parse_args = self.get_parser(required=True).parse_args
for args_string, expected_ns in self.successes:
actual_ns = parse_args(args_string.split())
self.assertEqual(actual_ns, expected_ns)
def test_usage_when_not_required(self):
format_usage = self.get_parser(required=False).format_usage
expected_usage = self.usage_when_not_required
self.assertEqual(format_usage(), textwrap.dedent(expected_usage))
def test_usage_when_required(self):
format_usage = self.get_parser(required=True).format_usage
expected_usage = self.usage_when_required
self.assertEqual(format_usage(), textwrap.dedent(expected_usage))
def test_help_when_not_required(self):
format_help = self.get_parser(required=False).format_help
help = self.usage_when_not_required + self.help
self.assertEqual(format_help(), textwrap.dedent(help))
def test_help_when_required(self):
format_help = self.get_parser(required=True).format_help
help = self.usage_when_required + self.help
self.assertEqual(format_help(), textwrap.dedent(help))
class TestMutuallyExclusiveSimple(MEMixin, TestCase):
def get_parser(self, required=None):
parser = ErrorRaisingArgumentParser(prog='PROG')
group = parser.add_mutually_exclusive_group(required=required)
group.add_argument('--bar', help='bar help')
group.add_argument('--baz', nargs='?', const='Z', help='baz help')
return parser
failures = ['--bar X --baz Y', '--bar X --baz']
successes = [
('--bar X', NS(bar='X', baz=None)),
('--bar X --bar Z', NS(bar='Z', baz=None)),
('--baz Y', NS(bar=None, baz='Y')),
('--baz', NS(bar=None, baz='Z')),
]
successes_when_not_required = [
('', NS(bar=None, baz=None)),
]
usage_when_not_required = '''\
usage: PROG [-h] [--bar BAR | --baz [BAZ]]
'''
usage_when_required = '''\
usage: PROG [-h] (--bar BAR | --baz [BAZ])
'''
help = '''\
optional arguments:
-h, --help show this help message and exit
--bar BAR bar help
--baz [BAZ] baz help
'''
class TestMutuallyExclusiveLong(MEMixin, TestCase):
def get_parser(self, required=None):
parser = ErrorRaisingArgumentParser(prog='PROG')
parser.add_argument('--abcde', help='abcde help')
parser.add_argument('--fghij', help='fghij help')
group = parser.add_mutually_exclusive_group(required=required)
group.add_argument('--klmno', help='klmno help')
group.add_argument('--pqrst', help='pqrst help')
return parser
failures = ['--klmno X --pqrst Y']
successes = [
('--klmno X', NS(abcde=None, fghij=None, klmno='X', pqrst=None)),
('--abcde Y --klmno X',
NS(abcde='Y', fghij=None, klmno='X', pqrst=None)),
('--pqrst X', NS(abcde=None, fghij=None, klmno=None, pqrst='X')),
('--pqrst X --fghij Y',
NS(abcde=None, fghij='Y', klmno=None, pqrst='X')),
]
successes_when_not_required = [
('', NS(abcde=None, fghij=None, klmno=None, pqrst=None)),
]
usage_when_not_required = '''\
usage: PROG [-h] [--abcde ABCDE] [--fghij FGHIJ]
[--klmno KLMNO | --pqrst PQRST]
'''
usage_when_required = '''\
usage: PROG [-h] [--abcde ABCDE] [--fghij FGHIJ]
(--klmno KLMNO | --pqrst PQRST)
'''
help = '''\
optional arguments:
-h, --help show this help message and exit
--abcde ABCDE abcde help
--fghij FGHIJ fghij help
--klmno KLMNO klmno help
--pqrst PQRST pqrst help
'''
class TestMutuallyExclusiveFirstSuppressed(MEMixin, TestCase):
def get_parser(self, required):
parser = ErrorRaisingArgumentParser(prog='PROG')
group = parser.add_mutually_exclusive_group(required=required)
group.add_argument('-x', help=argparse.SUPPRESS)
group.add_argument('-y', action='store_false', help='y help')
return parser
failures = ['-x X -y']
successes = [
('-x X', NS(x='X', y=True)),
('-x X -x Y', NS(x='Y', y=True)),
('-y', NS(x=None, y=False)),
]
successes_when_not_required = [
('', NS(x=None, y=True)),
]
usage_when_not_required = '''\
usage: PROG [-h] [-y]
'''
usage_when_required = '''\
usage: PROG [-h] -y
'''
help = '''\
optional arguments:
-h, --help show this help message and exit
-y y help
'''
class TestMutuallyExclusiveManySuppressed(MEMixin, TestCase):
def get_parser(self, required):
parser = ErrorRaisingArgumentParser(prog='PROG')
group = parser.add_mutually_exclusive_group(required=required)
add = group.add_argument
add('--spam', action='store_true', help=argparse.SUPPRESS)
add('--badger', action='store_false', help=argparse.SUPPRESS)
add('--bladder', help=argparse.SUPPRESS)
return parser
failures = [
'--spam --badger',
'--badger --bladder B',
'--bladder B --spam',
]
successes = [
('--spam', NS(spam=True, badger=True, bladder=None)),
('--badger', NS(spam=False, badger=False, bladder=None)),
('--bladder B', NS(spam=False, badger=True, bladder='B')),
('--spam --spam', NS(spam=True, badger=True, bladder=None)),
]
successes_when_not_required = [
('', NS(spam=False, badger=True, bladder=None)),
]
usage_when_required = usage_when_not_required = '''\
usage: PROG [-h]
'''
help = '''\
optional arguments:
-h, --help show this help message and exit
'''
class TestMutuallyExclusiveOptionalAndPositional(MEMixin, TestCase):
def get_parser(self, required):
parser = ErrorRaisingArgumentParser(prog='PROG')
group = parser.add_mutually_exclusive_group(required=required)
group.add_argument('--foo', action='store_true', help='FOO')
group.add_argument('--spam', help='SPAM')
group.add_argument('badger', nargs='*', default='X', help='BADGER')
return parser
failures = [
'--foo --spam S',
'--spam S X',
'X --foo',
'X Y Z --spam S',
'--foo X Y',
]
successes = [
('--foo', NS(foo=True, spam=None, badger='X')),
('--spam S', NS(foo=False, spam='S', badger='X')),
('X', NS(foo=False, spam=None, badger=['X'])),
('X Y Z', NS(foo=False, spam=None, badger=['X', 'Y', 'Z'])),
]
successes_when_not_required = [
('', NS(foo=False, spam=None, badger='X')),
]
usage_when_not_required = '''\
usage: PROG [-h] [--foo | --spam SPAM | badger ...]
'''
usage_when_required = '''\
usage: PROG [-h] (--foo | --spam SPAM | badger ...)
'''
help = '''\
positional arguments:
badger BADGER
optional arguments:
-h, --help show this help message and exit
--foo FOO
--spam SPAM SPAM
'''
class TestMutuallyExclusiveOptionalsMixed(MEMixin, TestCase):
def get_parser(self, required):
parser = ErrorRaisingArgumentParser(prog='PROG')
parser.add_argument('-x', action='store_true', help='x help')
group = parser.add_mutually_exclusive_group(required=required)
group.add_argument('-a', action='store_true', help='a help')
group.add_argument('-b', action='store_true', help='b help')
parser.add_argument('-y', action='store_true', help='y help')
group.add_argument('-c', action='store_true', help='c help')
return parser
failures = ['-a -b', '-b -c', '-a -c', '-a -b -c']
successes = [
('-a', NS(a=True, b=False, c=False, x=False, y=False)),
('-b', NS(a=False, b=True, c=False, x=False, y=False)),
('-c', NS(a=False, b=False, c=True, x=False, y=False)),
('-a -x', NS(a=True, b=False, c=False, x=True, y=False)),
('-y -b', NS(a=False, b=True, c=False, x=False, y=True)),
('-x -y -c', NS(a=False, b=False, c=True, x=True, y=True)),
]
successes_when_not_required = [
('', NS(a=False, b=False, c=False, x=False, y=False)),
('-x', NS(a=False, b=False, c=False, x=True, y=False)),
('-y', NS(a=False, b=False, c=False, x=False, y=True)),
]
usage_when_required = usage_when_not_required = '''\
usage: PROG [-h] [-x] [-a] [-b] [-y] [-c]
'''
help = '''\
optional arguments:
-h, --help show this help message and exit
-x x help
-a a help
-b b help
-y y help
-c c help
'''
class TestMutuallyExclusiveInGroup(MEMixin, TestCase):
def get_parser(self, required=None):
parser = ErrorRaisingArgumentParser(prog='PROG')
titled_group = parser.add_argument_group(
title='Titled group', description='Group description')
mutex_group = \
titled_group.add_mutually_exclusive_group(required=required)
mutex_group.add_argument('--bar', help='bar help')
mutex_group.add_argument('--baz', help='baz help')
return parser
failures = ['--bar X --baz Y', '--baz X --bar Y']
successes = [
('--bar X', NS(bar='X', baz=None)),
('--baz Y', NS(bar=None, baz='Y')),
]
successes_when_not_required = [
('', NS(bar=None, baz=None)),
]
usage_when_not_required = '''\
usage: PROG [-h] [--bar BAR | --baz BAZ]
'''
usage_when_required = '''\
usage: PROG [-h] (--bar BAR | --baz BAZ)
'''
help = '''\
optional arguments:
-h, --help show this help message and exit
Titled group:
Group description
--bar BAR bar help
--baz BAZ baz help
'''
class TestMutuallyExclusiveOptionalsAndPositionalsMixed(MEMixin, TestCase):
def get_parser(self, required):
parser = ErrorRaisingArgumentParser(prog='PROG')
parser.add_argument('x', help='x help')
parser.add_argument('-y', action='store_true', help='y help')
group = parser.add_mutually_exclusive_group(required=required)
group.add_argument('a', nargs='?', help='a help')
group.add_argument('-b', action='store_true', help='b help')
group.add_argument('-c', action='store_true', help='c help')
return parser
failures = ['X A -b', '-b -c', '-c X A']
successes = [
('X A', NS(a='A', b=False, c=False, x='X', y=False)),
('X -b', NS(a=None, b=True, c=False, x='X', y=False)),
('X -c', NS(a=None, b=False, c=True, x='X', y=False)),
('X A -y', NS(a='A', b=False, c=False, x='X', y=True)),
('X -y -b', NS(a=None, b=True, c=False, x='X', y=True)),
]
successes_when_not_required = [
('X', NS(a=None, b=False, c=False, x='X', y=False)),
('X -y', NS(a=None, b=False, c=False, x='X', y=True)),
]
usage_when_required = usage_when_not_required = '''\
usage: PROG [-h] [-y] [-b] [-c] x [a]
'''
help = '''\
positional arguments:
x x help
a a help
optional arguments:
-h, --help show this help message and exit
-y y help
-b b help
-c c help
'''
class TestMutuallyExclusiveNested(MEMixin, TestCase):
def get_parser(self, required):
parser = ErrorRaisingArgumentParser(prog='PROG')
group = parser.add_mutually_exclusive_group(required=required)
group.add_argument('-a')
group.add_argument('-b')
group2 = group.add_mutually_exclusive_group(required=required)
group2.add_argument('-c')
group2.add_argument('-d')
group3 = group2.add_mutually_exclusive_group(required=required)
group3.add_argument('-e')
group3.add_argument('-f')
return parser
usage_when_not_required = '''\
usage: PROG [-h] [-a A | -b B | [-c C | -d D | [-e E | -f F]]]
'''
usage_when_required = '''\
usage: PROG [-h] (-a A | -b B | (-c C | -d D | (-e E | -f F)))
'''
help = '''\
optional arguments:
-h, --help show this help message and exit
-a A
-b B
-c C
-d D
-e E
-f F
'''
# We are only interested in testing the behavior of format_usage().
test_failures_when_not_required = None
test_failures_when_required = None
test_successes_when_not_required = None
test_successes_when_required = None
# =================================================
# Mutually exclusive group in parent parser tests
# =================================================
class MEPBase(object):
def get_parser(self, required=None):
parent = super(MEPBase, self).get_parser(required=required)
parser = ErrorRaisingArgumentParser(
prog=parent.prog, add_help=False, parents=[parent])
return parser
class TestMutuallyExclusiveGroupErrorsParent(
MEPBase, TestMutuallyExclusiveGroupErrors):
pass
class TestMutuallyExclusiveSimpleParent(
MEPBase, TestMutuallyExclusiveSimple):
pass
class TestMutuallyExclusiveLongParent(
MEPBase, TestMutuallyExclusiveLong):
pass
class TestMutuallyExclusiveFirstSuppressedParent(
MEPBase, TestMutuallyExclusiveFirstSuppressed):
pass
class TestMutuallyExclusiveManySuppressedParent(
MEPBase, TestMutuallyExclusiveManySuppressed):
pass
class TestMutuallyExclusiveOptionalAndPositionalParent(
MEPBase, TestMutuallyExclusiveOptionalAndPositional):
pass
class TestMutuallyExclusiveOptionalsMixedParent(
MEPBase, TestMutuallyExclusiveOptionalsMixed):
pass
class TestMutuallyExclusiveOptionalsAndPositionalsMixedParent(
MEPBase, TestMutuallyExclusiveOptionalsAndPositionalsMixed):
pass
# =================
# Set default tests
# =================
class TestSetDefaults(TestCase):
def test_set_defaults_no_args(self):
parser = ErrorRaisingArgumentParser()
parser.set_defaults(x='foo')
parser.set_defaults(y='bar', z=1)
self.assertEqual(NS(x='foo', y='bar', z=1),
parser.parse_args([]))
self.assertEqual(NS(x='foo', y='bar', z=1),
parser.parse_args([], NS()))
self.assertEqual(NS(x='baz', y='bar', z=1),
parser.parse_args([], NS(x='baz')))
self.assertEqual(NS(x='baz', y='bar', z=2),
parser.parse_args([], NS(x='baz', z=2)))
def test_set_defaults_with_args(self):
parser = ErrorRaisingArgumentParser()
parser.set_defaults(x='foo', y='bar')
parser.add_argument('-x', default='xfoox')
self.assertEqual(NS(x='xfoox', y='bar'),
parser.parse_args([]))
self.assertEqual(NS(x='xfoox', y='bar'),
parser.parse_args([], NS()))
self.assertEqual(NS(x='baz', y='bar'),
parser.parse_args([], NS(x='baz')))
self.assertEqual(NS(x='1', y='bar'),
parser.parse_args('-x 1'.split()))
self.assertEqual(NS(x='1', y='bar'),
parser.parse_args('-x 1'.split(), NS()))
self.assertEqual(NS(x='1', y='bar'),
parser.parse_args('-x 1'.split(), NS(x='baz')))
def test_set_defaults_subparsers(self):
parser = ErrorRaisingArgumentParser()
parser.set_defaults(x='foo')
subparsers = parser.add_subparsers()
parser_a = subparsers.add_parser('a')
parser_a.set_defaults(y='bar')
self.assertEqual(NS(x='foo', y='bar'),
parser.parse_args('a'.split()))
def test_set_defaults_parents(self):
parent = ErrorRaisingArgumentParser(add_help=False)
parent.set_defaults(x='foo')
parser = ErrorRaisingArgumentParser(parents=[parent])
self.assertEqual(NS(x='foo'), parser.parse_args([]))
def test_set_defaults_on_parent_and_subparser(self):
parser = argparse.ArgumentParser()
xparser = parser.add_subparsers().add_parser('X')
parser.set_defaults(foo=1)
xparser.set_defaults(foo=2)
self.assertEqual(NS(foo=2), parser.parse_args(['X']))
def test_set_defaults_same_as_add_argument(self):
parser = ErrorRaisingArgumentParser()
parser.set_defaults(w='W', x='X', y='Y', z='Z')
parser.add_argument('-w')
parser.add_argument('-x', default='XX')
parser.add_argument('y', nargs='?')
parser.add_argument('z', nargs='?', default='ZZ')
# defaults set previously
self.assertEqual(NS(w='W', x='XX', y='Y', z='ZZ'),
parser.parse_args([]))
# reset defaults
parser.set_defaults(w='WW', x='X', y='YY', z='Z')
self.assertEqual(NS(w='WW', x='X', y='YY', z='Z'),
parser.parse_args([]))
def test_set_defaults_same_as_add_argument_group(self):
parser = ErrorRaisingArgumentParser()
parser.set_defaults(w='W', x='X', y='Y', z='Z')
group = parser.add_argument_group('foo')
group.add_argument('-w')
group.add_argument('-x', default='XX')
group.add_argument('y', nargs='?')
group.add_argument('z', nargs='?', default='ZZ')
# defaults set previously
self.assertEqual(NS(w='W', x='XX', y='Y', z='ZZ'),
parser.parse_args([]))
# reset defaults
parser.set_defaults(w='WW', x='X', y='YY', z='Z')
self.assertEqual(NS(w='WW', x='X', y='YY', z='Z'),
parser.parse_args([]))
# =================
# Get default tests
# =================
class TestGetDefault(TestCase):
def test_get_default(self):
parser = ErrorRaisingArgumentParser()
self.assertIsNone(parser.get_default("foo"))
self.assertIsNone(parser.get_default("bar"))
parser.add_argument("--foo")
self.assertIsNone(parser.get_default("foo"))
self.assertIsNone(parser.get_default("bar"))
parser.add_argument("--bar", type=int, default=42)
self.assertIsNone(parser.get_default("foo"))
self.assertEqual(42, parser.get_default("bar"))
parser.set_defaults(foo="badger")
self.assertEqual("badger", parser.get_default("foo"))
self.assertEqual(42, parser.get_default("bar"))
# ==========================
# Namespace 'contains' tests
# ==========================
class TestNamespaceContainsSimple(TestCase):
def test_empty(self):
ns = argparse.Namespace()
self.assertNotIn('', ns)
self.assertNotIn('x', ns)
def test_non_empty(self):
ns = argparse.Namespace(x=1, y=2)
self.assertNotIn('', ns)
self.assertIn('x', ns)
self.assertIn('y', ns)
self.assertNotIn('xx', ns)
self.assertNotIn('z', ns)
# =====================
# Help formatting tests
# =====================
class TestHelpFormattingMetaclass(type):
def __init__(cls, name, bases, bodydict):
if name == 'HelpTestCase':
return
class AddTests(object):
def __init__(self, test_class, func_suffix, std_name):
self.func_suffix = func_suffix
self.std_name = std_name
for test_func in [self.test_format,
self.test_print,
self.test_print_file]:
test_name = '%s_%s' % (test_func.__name__, func_suffix)
def test_wrapper(self, test_func=test_func):
test_func(self)
try:
test_wrapper.__name__ = test_name
except TypeError:
pass
setattr(test_class, test_name, test_wrapper)
def _get_parser(self, tester):
parser = argparse.ArgumentParser(
*tester.parser_signature.args,
**tester.parser_signature.kwargs)
for argument_sig in getattr(tester, 'argument_signatures', []):
parser.add_argument(*argument_sig.args,
**argument_sig.kwargs)
group_sigs = getattr(tester, 'argument_group_signatures', [])
for group_sig, argument_sigs in group_sigs:
group = parser.add_argument_group(*group_sig.args,
**group_sig.kwargs)
for argument_sig in argument_sigs:
group.add_argument(*argument_sig.args,
**argument_sig.kwargs)
subparsers_sigs = getattr(tester, 'subparsers_signatures', [])
if subparsers_sigs:
subparsers = parser.add_subparsers()
for subparser_sig in subparsers_sigs:
subparsers.add_parser(*subparser_sig.args,
**subparser_sig.kwargs)
return parser
def _test(self, tester, parser_text):
expected_text = getattr(tester, self.func_suffix)
expected_text = textwrap.dedent(expected_text)
tester.assertEqual(expected_text, parser_text)
def test_format(self, tester):
parser = self._get_parser(tester)
format = getattr(parser, 'format_%s' % self.func_suffix)
self._test(tester, format())
def test_print(self, tester):
parser = self._get_parser(tester)
print_ = getattr(parser, 'print_%s' % self.func_suffix)
old_stream = getattr(sys, self.std_name)
Merged revisions 79464,79471,79623,79626,79630,79632,79643,79648-79649,79679,79685,79711,79761,79774,79777,79792-79794,79877,79898-79900 via svnmerge from svn+ssh://pythondev@svn.python.org/python/trunk ........ r79464 | michael.foord | 2010-03-27 07:55:19 -0500 (Sat, 27 Mar 2010) | 1 line A fix for running unittest tests on platforms without the audioop module (e.g. jython and IronPython) ........ r79471 | michael.foord | 2010-03-27 14:10:11 -0500 (Sat, 27 Mar 2010) | 4 lines Addition of delta keyword argument to unittest.TestCase.assertAlmostEquals and assertNotAlmostEquals This allows the comparison of objects by specifying a maximum difference; this includes the comparing of non-numeric objects that don't support rounding. ........ r79623 | michael.foord | 2010-04-02 16:42:47 -0500 (Fri, 02 Apr 2010) | 1 line Addition of -b command line option to unittest for buffering stdout and stderr during test runs. ........ r79626 | michael.foord | 2010-04-02 17:08:29 -0500 (Fri, 02 Apr 2010) | 1 line TestResult stores original sys.stdout and tests no longer use sys.__stdout__ (etc) in tests for unittest -b command line option ........ r79630 | michael.foord | 2010-04-02 17:30:56 -0500 (Fri, 02 Apr 2010) | 1 line unittest tests no longer replace the sys.stdout put in place by regrtest ........ r79632 | michael.foord | 2010-04-02 17:55:59 -0500 (Fri, 02 Apr 2010) | 1 line Issue #8038: Addition of unittest.TestCase.assertNotRegexpMatches ........ r79643 | michael.foord | 2010-04-02 20:15:21 -0500 (Fri, 02 Apr 2010) | 1 line Support dotted module names for test discovery paths in unittest. Issue 8038. ........ r79648 | michael.foord | 2010-04-02 21:21:39 -0500 (Fri, 02 Apr 2010) | 1 line Cross platform unittest.TestResult newline handling when buffering stdout / stderr. ........ r79649 | michael.foord | 2010-04-02 21:33:55 -0500 (Fri, 02 Apr 2010) | 1 line Another attempt at a fix for unittest.test.test_result for windows line endings ........ r79679 | michael.foord | 2010-04-03 09:52:18 -0500 (Sat, 03 Apr 2010) | 1 line Adding -b command line option to the unittest usage message. ........ r79685 | michael.foord | 2010-04-03 10:20:00 -0500 (Sat, 03 Apr 2010) | 1 line Minor tweak to unittest command line usage message ........ r79711 | michael.foord | 2010-04-03 12:03:11 -0500 (Sat, 03 Apr 2010) | 1 line Documenting new features in unittest ........ r79761 | michael.foord | 2010-04-04 17:41:54 -0500 (Sun, 04 Apr 2010) | 1 line unittest documentation formatting changes ........ r79774 | michael.foord | 2010-04-04 18:28:44 -0500 (Sun, 04 Apr 2010) | 1 line Adding documentation for new unittest.main() parameters ........ r79777 | michael.foord | 2010-04-04 19:39:50 -0500 (Sun, 04 Apr 2010) | 1 line Document signal handling functions in unittest.rst ........ r79792 | michael.foord | 2010-04-05 05:26:26 -0500 (Mon, 05 Apr 2010) | 1 line Documentation fixes for unittest ........ r79793 | michael.foord | 2010-04-05 05:28:27 -0500 (Mon, 05 Apr 2010) | 1 line Furterh documentation fix for unittest.rst ........ r79794 | michael.foord | 2010-04-05 05:30:14 -0500 (Mon, 05 Apr 2010) | 1 line Further documentation fix for unittest.rst ........ r79877 | michael.foord | 2010-04-06 18:18:16 -0500 (Tue, 06 Apr 2010) | 1 line Fix module directory finding logic for dotted paths in unittest test discovery. ........ r79898 | michael.foord | 2010-04-07 18:04:22 -0500 (Wed, 07 Apr 2010) | 1 line unittest.result.TestResult does not create its buffers until they're used. It uses StringIO not cStringIO. Issue 8333. ........ r79899 | michael.foord | 2010-04-07 19:04:24 -0500 (Wed, 07 Apr 2010) | 1 line Switch regrtest to use StringIO instead of cStringIO for test_multiprocessing on Windows. Issue 8333. ........ r79900 | michael.foord | 2010-04-07 23:33:20 -0500 (Wed, 07 Apr 2010) | 1 line Correction of unittest documentation typos and omissions ........
2010-04-11 17:43:16 -03:00
setattr(sys, self.std_name, StdIOBuffer())
try:
print_()
parser_text = getattr(sys, self.std_name).getvalue()
finally:
setattr(sys, self.std_name, old_stream)
self._test(tester, parser_text)
def test_print_file(self, tester):
parser = self._get_parser(tester)
print_ = getattr(parser, 'print_%s' % self.func_suffix)
Merged revisions 79464,79471,79623,79626,79630,79632,79643,79648-79649,79679,79685,79711,79761,79774,79777,79792-79794,79877,79898-79900 via svnmerge from svn+ssh://pythondev@svn.python.org/python/trunk ........ r79464 | michael.foord | 2010-03-27 07:55:19 -0500 (Sat, 27 Mar 2010) | 1 line A fix for running unittest tests on platforms without the audioop module (e.g. jython and IronPython) ........ r79471 | michael.foord | 2010-03-27 14:10:11 -0500 (Sat, 27 Mar 2010) | 4 lines Addition of delta keyword argument to unittest.TestCase.assertAlmostEquals and assertNotAlmostEquals This allows the comparison of objects by specifying a maximum difference; this includes the comparing of non-numeric objects that don't support rounding. ........ r79623 | michael.foord | 2010-04-02 16:42:47 -0500 (Fri, 02 Apr 2010) | 1 line Addition of -b command line option to unittest for buffering stdout and stderr during test runs. ........ r79626 | michael.foord | 2010-04-02 17:08:29 -0500 (Fri, 02 Apr 2010) | 1 line TestResult stores original sys.stdout and tests no longer use sys.__stdout__ (etc) in tests for unittest -b command line option ........ r79630 | michael.foord | 2010-04-02 17:30:56 -0500 (Fri, 02 Apr 2010) | 1 line unittest tests no longer replace the sys.stdout put in place by regrtest ........ r79632 | michael.foord | 2010-04-02 17:55:59 -0500 (Fri, 02 Apr 2010) | 1 line Issue #8038: Addition of unittest.TestCase.assertNotRegexpMatches ........ r79643 | michael.foord | 2010-04-02 20:15:21 -0500 (Fri, 02 Apr 2010) | 1 line Support dotted module names for test discovery paths in unittest. Issue 8038. ........ r79648 | michael.foord | 2010-04-02 21:21:39 -0500 (Fri, 02 Apr 2010) | 1 line Cross platform unittest.TestResult newline handling when buffering stdout / stderr. ........ r79649 | michael.foord | 2010-04-02 21:33:55 -0500 (Fri, 02 Apr 2010) | 1 line Another attempt at a fix for unittest.test.test_result for windows line endings ........ r79679 | michael.foord | 2010-04-03 09:52:18 -0500 (Sat, 03 Apr 2010) | 1 line Adding -b command line option to the unittest usage message. ........ r79685 | michael.foord | 2010-04-03 10:20:00 -0500 (Sat, 03 Apr 2010) | 1 line Minor tweak to unittest command line usage message ........ r79711 | michael.foord | 2010-04-03 12:03:11 -0500 (Sat, 03 Apr 2010) | 1 line Documenting new features in unittest ........ r79761 | michael.foord | 2010-04-04 17:41:54 -0500 (Sun, 04 Apr 2010) | 1 line unittest documentation formatting changes ........ r79774 | michael.foord | 2010-04-04 18:28:44 -0500 (Sun, 04 Apr 2010) | 1 line Adding documentation for new unittest.main() parameters ........ r79777 | michael.foord | 2010-04-04 19:39:50 -0500 (Sun, 04 Apr 2010) | 1 line Document signal handling functions in unittest.rst ........ r79792 | michael.foord | 2010-04-05 05:26:26 -0500 (Mon, 05 Apr 2010) | 1 line Documentation fixes for unittest ........ r79793 | michael.foord | 2010-04-05 05:28:27 -0500 (Mon, 05 Apr 2010) | 1 line Furterh documentation fix for unittest.rst ........ r79794 | michael.foord | 2010-04-05 05:30:14 -0500 (Mon, 05 Apr 2010) | 1 line Further documentation fix for unittest.rst ........ r79877 | michael.foord | 2010-04-06 18:18:16 -0500 (Tue, 06 Apr 2010) | 1 line Fix module directory finding logic for dotted paths in unittest test discovery. ........ r79898 | michael.foord | 2010-04-07 18:04:22 -0500 (Wed, 07 Apr 2010) | 1 line unittest.result.TestResult does not create its buffers until they're used. It uses StringIO not cStringIO. Issue 8333. ........ r79899 | michael.foord | 2010-04-07 19:04:24 -0500 (Wed, 07 Apr 2010) | 1 line Switch regrtest to use StringIO instead of cStringIO for test_multiprocessing on Windows. Issue 8333. ........ r79900 | michael.foord | 2010-04-07 23:33:20 -0500 (Wed, 07 Apr 2010) | 1 line Correction of unittest documentation typos and omissions ........
2010-04-11 17:43:16 -03:00
sfile = StdIOBuffer()
print_(sfile)
parser_text = sfile.getvalue()
self._test(tester, parser_text)
# add tests for {format,print}_{usage,help}
for func_suffix, std_name in [('usage', 'stdout'),
('help', 'stdout')]:
AddTests(cls, func_suffix, std_name)
bases = TestCase,
HelpTestCase = TestHelpFormattingMetaclass('HelpTestCase', bases, {})
class TestHelpBiggerOptionals(HelpTestCase):
"""Make sure that argument help aligns when options are longer"""
parser_signature = Sig(prog='PROG', description='DESCRIPTION',
epilog='EPILOG')
argument_signatures = [
Sig('-v', '--version', action='version', version='0.1'),
Sig('-x', action='store_true', help='X HELP'),
Sig('--y', help='Y HELP'),
Sig('foo', help='FOO HELP'),
Sig('bar', help='BAR HELP'),
]
argument_group_signatures = []
usage = '''\
usage: PROG [-h] [-v] [-x] [--y Y] foo bar
'''
help = usage + '''\
DESCRIPTION
positional arguments:
foo FOO HELP
bar BAR HELP
optional arguments:
-h, --help show this help message and exit
-v, --version show program's version number and exit
-x X HELP
--y Y Y HELP
EPILOG
'''
version = '''\
0.1
'''
class TestShortColumns(HelpTestCase):
'''Test extremely small number of columns.
TestCase prevents "COLUMNS" from being too small in the tests themselves,
but we don't want any exceptions thrown in such cases. Only ugly representation.
'''
def setUp(self):
env = support.EnvironmentVarGuard()
env.set("COLUMNS", '15')
self.addCleanup(env.__exit__)
parser_signature = TestHelpBiggerOptionals.parser_signature
argument_signatures = TestHelpBiggerOptionals.argument_signatures
argument_group_signatures = TestHelpBiggerOptionals.argument_group_signatures
usage = '''\
usage: PROG
[-h]
[-v]
[-x]
[--y Y]
foo
bar
'''
help = usage + '''\
DESCRIPTION
positional arguments:
foo
FOO HELP
bar
BAR HELP
optional arguments:
-h, --help
show this
help
message and
exit
-v, --version
show
program's
version
number and
exit
-x
X HELP
--y Y
Y HELP
EPILOG
'''
version = TestHelpBiggerOptionals.version
class TestHelpBiggerOptionalGroups(HelpTestCase):
"""Make sure that argument help aligns when options are longer"""
parser_signature = Sig(prog='PROG', description='DESCRIPTION',
epilog='EPILOG')
argument_signatures = [
Sig('-v', '--version', action='version', version='0.1'),
Sig('-x', action='store_true', help='X HELP'),
Sig('--y', help='Y HELP'),
Sig('foo', help='FOO HELP'),
Sig('bar', help='BAR HELP'),
]
argument_group_signatures = [
(Sig('GROUP TITLE', description='GROUP DESCRIPTION'), [
Sig('baz', help='BAZ HELP'),
Sig('-z', nargs='+', help='Z HELP')]),
]
usage = '''\
usage: PROG [-h] [-v] [-x] [--y Y] [-z Z [Z ...]] foo bar baz
'''
help = usage + '''\
DESCRIPTION
positional arguments:
foo FOO HELP
bar BAR HELP
optional arguments:
-h, --help show this help message and exit
-v, --version show program's version number and exit
-x X HELP
--y Y Y HELP
GROUP TITLE:
GROUP DESCRIPTION
baz BAZ HELP
-z Z [Z ...] Z HELP
EPILOG
'''
version = '''\
0.1
'''
class TestHelpBiggerPositionals(HelpTestCase):
"""Make sure that help aligns when arguments are longer"""
parser_signature = Sig(usage='USAGE', description='DESCRIPTION')
argument_signatures = [
Sig('-x', action='store_true', help='X HELP'),
Sig('--y', help='Y HELP'),
Sig('ekiekiekifekang', help='EKI HELP'),
Sig('bar', help='BAR HELP'),
]
argument_group_signatures = []
usage = '''\
usage: USAGE
'''
help = usage + '''\
DESCRIPTION
positional arguments:
ekiekiekifekang EKI HELP
bar BAR HELP
optional arguments:
-h, --help show this help message and exit
-x X HELP
--y Y Y HELP
'''
version = ''
class TestHelpReformatting(HelpTestCase):
"""Make sure that text after short names starts on the first line"""
parser_signature = Sig(
prog='PROG',
description=' oddly formatted\n'
'description\n'
'\n'
'that is so long that it should go onto multiple '
'lines when wrapped')
argument_signatures = [
Sig('-x', metavar='XX', help='oddly\n'
' formatted -x help'),
Sig('y', metavar='yyy', help='normal y help'),
]
argument_group_signatures = [
(Sig('title', description='\n'
' oddly formatted group\n'
'\n'
'description'),
[Sig('-a', action='store_true',
help=' oddly \n'
'formatted -a help \n'
' again, so long that it should be wrapped over '
'multiple lines')]),
]
usage = '''\
usage: PROG [-h] [-x XX] [-a] yyy
'''
help = usage + '''\
oddly formatted description that is so long that it should go onto \
multiple
lines when wrapped
positional arguments:
yyy normal y help
optional arguments:
-h, --help show this help message and exit
-x XX oddly formatted -x help
title:
oddly formatted group description
-a oddly formatted -a help again, so long that it should \
be wrapped
over multiple lines
'''
version = ''
class TestHelpWrappingShortNames(HelpTestCase):
"""Make sure that text after short names starts on the first line"""
parser_signature = Sig(prog='PROG', description= 'D\nD' * 30)
argument_signatures = [
Sig('-x', metavar='XX', help='XHH HX' * 20),
Sig('y', metavar='yyy', help='YH YH' * 20),
]
argument_group_signatures = [
(Sig('ALPHAS'), [
Sig('-a', action='store_true', help='AHHH HHA' * 10)]),
]
usage = '''\
usage: PROG [-h] [-x XX] [-a] yyy
'''
help = usage + '''\
D DD DD DD DD DD DD DD DD DD DD DD DD DD DD DD DD DD DD DD DD DD DD \
DD DD DD
DD DD DD DD D
positional arguments:
yyy YH YHYH YHYH YHYH YHYH YHYH YHYH YHYH YHYH YHYH YHYH \
YHYH YHYH
YHYH YHYH YHYH YHYH YHYH YHYH YHYH YH
optional arguments:
-h, --help show this help message and exit
-x XX XHH HXXHH HXXHH HXXHH HXXHH HXXHH HXXHH HXXHH HXXHH \
HXXHH HXXHH
HXXHH HXXHH HXXHH HXXHH HXXHH HXXHH HXXHH HXXHH HXXHH HX
ALPHAS:
-a AHHH HHAAHHH HHAAHHH HHAAHHH HHAAHHH HHAAHHH HHAAHHH \
HHAAHHH
HHAAHHH HHAAHHH HHA
'''
version = ''
class TestHelpWrappingLongNames(HelpTestCase):
"""Make sure that text after long names starts on the next line"""
parser_signature = Sig(usage='USAGE', description= 'D D' * 30)
argument_signatures = [
Sig('-v', '--version', action='version', version='V V' * 30),
Sig('-x', metavar='X' * 25, help='XH XH' * 20),
Sig('y', metavar='y' * 25, help='YH YH' * 20),
]
argument_group_signatures = [
(Sig('ALPHAS'), [
Sig('-a', metavar='A' * 25, help='AH AH' * 20),
Sig('z', metavar='z' * 25, help='ZH ZH' * 20)]),
]
usage = '''\
usage: USAGE
'''
help = usage + '''\
D DD DD DD DD DD DD DD DD DD DD DD DD DD DD DD DD DD DD DD DD DD DD \
DD DD DD
DD DD DD DD D
positional arguments:
yyyyyyyyyyyyyyyyyyyyyyyyy
YH YHYH YHYH YHYH YHYH YHYH YHYH YHYH YHYH \
YHYH YHYH
YHYH YHYH YHYH YHYH YHYH YHYH YHYH YHYH YHYH YH
optional arguments:
-h, --help show this help message and exit
-v, --version show program's version number and exit
-x XXXXXXXXXXXXXXXXXXXXXXXXX
XH XHXH XHXH XHXH XHXH XHXH XHXH XHXH XHXH \
XHXH XHXH
XHXH XHXH XHXH XHXH XHXH XHXH XHXH XHXH XHXH XH
ALPHAS:
-a AAAAAAAAAAAAAAAAAAAAAAAAA
AH AHAH AHAH AHAH AHAH AHAH AHAH AHAH AHAH \
AHAH AHAH
AHAH AHAH AHAH AHAH AHAH AHAH AHAH AHAH AHAH AH
zzzzzzzzzzzzzzzzzzzzzzzzz
ZH ZHZH ZHZH ZHZH ZHZH ZHZH ZHZH ZHZH ZHZH \
ZHZH ZHZH
ZHZH ZHZH ZHZH ZHZH ZHZH ZHZH ZHZH ZHZH ZHZH ZH
'''
version = '''\
V VV VV VV VV VV VV VV VV VV VV VV VV VV VV VV VV VV VV VV VV VV VV \
VV VV VV
VV VV VV VV V
'''
class TestHelpUsage(HelpTestCase):
"""Test basic usage messages"""
parser_signature = Sig(prog='PROG')
argument_signatures = [
Sig('-w', nargs='+', help='w'),
Sig('-x', nargs='*', help='x'),
Sig('a', help='a'),
Sig('b', help='b', nargs=2),
Sig('c', help='c', nargs='?'),
Sig('--foo', help='Whether to foo', action=argparse.BooleanOptionalAction),
Sig('--bar', help='Whether to bar', default=True,
action=argparse.BooleanOptionalAction),
Sig('-f', '--foobar', '--barfoo', action=argparse.BooleanOptionalAction),
]
argument_group_signatures = [
(Sig('group'), [
Sig('-y', nargs='?', help='y'),
Sig('-z', nargs=3, help='z'),
Sig('d', help='d', nargs='*'),
Sig('e', help='e', nargs='+'),
])
]
usage = '''\
usage: PROG [-h] [-w W [W ...]] [-x [X ...]] [--foo | --no-foo]
[--bar | --no-bar]
[-f | --foobar | --no-foobar | --barfoo | --no-barfoo] [-y [Y]]
[-z Z Z Z]
a b b [c] [d ...] e [e ...]
'''
help = usage + '''\
positional arguments:
a a
b b
c c
optional arguments:
-h, --help show this help message and exit
-w W [W ...] w
-x [X ...] x
--foo, --no-foo Whether to foo
--bar, --no-bar Whether to bar (default: True)
-f, --foobar, --no-foobar, --barfoo, --no-barfoo
group:
-y [Y] y
-z Z Z Z z
d d
e e
'''
version = ''
class TestHelpOnlyUserGroups(HelpTestCase):
"""Test basic usage messages"""
parser_signature = Sig(prog='PROG', add_help=False)
argument_signatures = []
argument_group_signatures = [
(Sig('xxxx'), [
Sig('-x', help='x'),
Sig('a', help='a'),
]),
(Sig('yyyy'), [
Sig('b', help='b'),
Sig('-y', help='y'),
]),
]
usage = '''\
usage: PROG [-x X] [-y Y] a b
'''
help = usage + '''\
xxxx:
-x X x
a a
yyyy:
b b
-y Y y
'''
version = ''
class TestHelpUsageLongProg(HelpTestCase):
"""Test usage messages where the prog is long"""
parser_signature = Sig(prog='P' * 60)
argument_signatures = [
Sig('-w', metavar='W'),
Sig('-x', metavar='X'),
Sig('a'),
Sig('b'),
]
argument_group_signatures = []
usage = '''\
usage: PPPPPPPPPPPPPPPPPPPPPPPPPPPPPPPPPPPPPPPPPPPPPPPPPPPPPPPPPPPP
[-h] [-w W] [-x X] a b
'''
help = usage + '''\
positional arguments:
a
b
optional arguments:
-h, --help show this help message and exit
-w W
-x X
'''
version = ''
class TestHelpUsageLongProgOptionsWrap(HelpTestCase):
"""Test usage messages where the prog is long and the optionals wrap"""
parser_signature = Sig(prog='P' * 60)
argument_signatures = [
Sig('-w', metavar='W' * 25),
Sig('-x', metavar='X' * 25),
Sig('-y', metavar='Y' * 25),
Sig('-z', metavar='Z' * 25),
Sig('a'),
Sig('b'),
]
argument_group_signatures = []
usage = '''\
usage: PPPPPPPPPPPPPPPPPPPPPPPPPPPPPPPPPPPPPPPPPPPPPPPPPPPPPPPPPPPP
[-h] [-w WWWWWWWWWWWWWWWWWWWWWWWWW] \
[-x XXXXXXXXXXXXXXXXXXXXXXXXX]
[-y YYYYYYYYYYYYYYYYYYYYYYYYY] [-z ZZZZZZZZZZZZZZZZZZZZZZZZZ]
a b
'''
help = usage + '''\
positional arguments:
a
b
optional arguments:
-h, --help show this help message and exit
-w WWWWWWWWWWWWWWWWWWWWWWWWW
-x XXXXXXXXXXXXXXXXXXXXXXXXX
-y YYYYYYYYYYYYYYYYYYYYYYYYY
-z ZZZZZZZZZZZZZZZZZZZZZZZZZ
'''
version = ''
class TestHelpUsageLongProgPositionalsWrap(HelpTestCase):
"""Test usage messages where the prog is long and the positionals wrap"""
parser_signature = Sig(prog='P' * 60, add_help=False)
argument_signatures = [
Sig('a' * 25),
Sig('b' * 25),
Sig('c' * 25),
]
argument_group_signatures = []
usage = '''\
usage: PPPPPPPPPPPPPPPPPPPPPPPPPPPPPPPPPPPPPPPPPPPPPPPPPPPPPPPPPPPP
aaaaaaaaaaaaaaaaaaaaaaaaa bbbbbbbbbbbbbbbbbbbbbbbbb
ccccccccccccccccccccccccc
'''
help = usage + '''\
positional arguments:
aaaaaaaaaaaaaaaaaaaaaaaaa
bbbbbbbbbbbbbbbbbbbbbbbbb
ccccccccccccccccccccccccc
'''
version = ''
class TestHelpUsageOptionalsWrap(HelpTestCase):
"""Test usage messages where the optionals wrap"""
parser_signature = Sig(prog='PROG')
argument_signatures = [
Sig('-w', metavar='W' * 25),
Sig('-x', metavar='X' * 25),
Sig('-y', metavar='Y' * 25),
Sig('-z', metavar='Z' * 25),
Sig('a'),
Sig('b'),
Sig('c'),
]
argument_group_signatures = []
usage = '''\
usage: PROG [-h] [-w WWWWWWWWWWWWWWWWWWWWWWWWW] \
[-x XXXXXXXXXXXXXXXXXXXXXXXXX]
[-y YYYYYYYYYYYYYYYYYYYYYYYYY] \
[-z ZZZZZZZZZZZZZZZZZZZZZZZZZ]
a b c
'''
help = usage + '''\
positional arguments:
a
b
c
optional arguments:
-h, --help show this help message and exit
-w WWWWWWWWWWWWWWWWWWWWWWWWW
-x XXXXXXXXXXXXXXXXXXXXXXXXX
-y YYYYYYYYYYYYYYYYYYYYYYYYY
-z ZZZZZZZZZZZZZZZZZZZZZZZZZ
'''
version = ''
class TestHelpUsagePositionalsWrap(HelpTestCase):
"""Test usage messages where the positionals wrap"""
parser_signature = Sig(prog='PROG')
argument_signatures = [
Sig('-x'),
Sig('-y'),
Sig('-z'),
Sig('a' * 25),
Sig('b' * 25),
Sig('c' * 25),
]
argument_group_signatures = []
usage = '''\
usage: PROG [-h] [-x X] [-y Y] [-z Z]
aaaaaaaaaaaaaaaaaaaaaaaaa bbbbbbbbbbbbbbbbbbbbbbbbb
ccccccccccccccccccccccccc
'''
help = usage + '''\
positional arguments:
aaaaaaaaaaaaaaaaaaaaaaaaa
bbbbbbbbbbbbbbbbbbbbbbbbb
ccccccccccccccccccccccccc
optional arguments:
-h, --help show this help message and exit
-x X
-y Y
-z Z
'''
version = ''
class TestHelpUsageOptionalsPositionalsWrap(HelpTestCase):
"""Test usage messages where the optionals and positionals wrap"""
parser_signature = Sig(prog='PROG')
argument_signatures = [
Sig('-x', metavar='X' * 25),
Sig('-y', metavar='Y' * 25),
Sig('-z', metavar='Z' * 25),
Sig('a' * 25),
Sig('b' * 25),
Sig('c' * 25),
]
argument_group_signatures = []
usage = '''\
usage: PROG [-h] [-x XXXXXXXXXXXXXXXXXXXXXXXXX] \
[-y YYYYYYYYYYYYYYYYYYYYYYYYY]
[-z ZZZZZZZZZZZZZZZZZZZZZZZZZ]
aaaaaaaaaaaaaaaaaaaaaaaaa bbbbbbbbbbbbbbbbbbbbbbbbb
ccccccccccccccccccccccccc
'''
help = usage + '''\
positional arguments:
aaaaaaaaaaaaaaaaaaaaaaaaa
bbbbbbbbbbbbbbbbbbbbbbbbb
ccccccccccccccccccccccccc
optional arguments:
-h, --help show this help message and exit
-x XXXXXXXXXXXXXXXXXXXXXXXXX
-y YYYYYYYYYYYYYYYYYYYYYYYYY
-z ZZZZZZZZZZZZZZZZZZZZZZZZZ
'''
version = ''
class TestHelpUsageOptionalsOnlyWrap(HelpTestCase):
"""Test usage messages where there are only optionals and they wrap"""
parser_signature = Sig(prog='PROG')
argument_signatures = [
Sig('-x', metavar='X' * 25),
Sig('-y', metavar='Y' * 25),
Sig('-z', metavar='Z' * 25),
]
argument_group_signatures = []
usage = '''\
usage: PROG [-h] [-x XXXXXXXXXXXXXXXXXXXXXXXXX] \
[-y YYYYYYYYYYYYYYYYYYYYYYYYY]
[-z ZZZZZZZZZZZZZZZZZZZZZZZZZ]
'''
help = usage + '''\
optional arguments:
-h, --help show this help message and exit
-x XXXXXXXXXXXXXXXXXXXXXXXXX
-y YYYYYYYYYYYYYYYYYYYYYYYYY
-z ZZZZZZZZZZZZZZZZZZZZZZZZZ
'''
version = ''
class TestHelpUsagePositionalsOnlyWrap(HelpTestCase):
"""Test usage messages where there are only positionals and they wrap"""
parser_signature = Sig(prog='PROG', add_help=False)
argument_signatures = [
Sig('a' * 25),
Sig('b' * 25),
Sig('c' * 25),
]
argument_group_signatures = []
usage = '''\
usage: PROG aaaaaaaaaaaaaaaaaaaaaaaaa bbbbbbbbbbbbbbbbbbbbbbbbb
ccccccccccccccccccccccccc
'''
help = usage + '''\
positional arguments:
aaaaaaaaaaaaaaaaaaaaaaaaa
bbbbbbbbbbbbbbbbbbbbbbbbb
ccccccccccccccccccccccccc
'''
version = ''
class TestHelpVariableExpansion(HelpTestCase):
"""Test that variables are expanded properly in help messages"""
parser_signature = Sig(prog='PROG')
argument_signatures = [
Sig('-x', type=int,
help='x %(prog)s %(default)s %(type)s %%'),
Sig('-y', action='store_const', default=42, const='XXX',
help='y %(prog)s %(default)s %(const)s'),
Sig('--foo', choices='abc',
help='foo %(prog)s %(default)s %(choices)s'),
Sig('--bar', default='baz', choices=[1, 2], metavar='BBB',
help='bar %(prog)s %(default)s %(dest)s'),
Sig('spam', help='spam %(prog)s %(default)s'),
Sig('badger', default=0.5, help='badger %(prog)s %(default)s'),
]
argument_group_signatures = [
(Sig('group'), [
Sig('-a', help='a %(prog)s %(default)s'),
Sig('-b', default=-1, help='b %(prog)s %(default)s'),
])
]
usage = ('''\
usage: PROG [-h] [-x X] [-y] [--foo {a,b,c}] [--bar BBB] [-a A] [-b B]
spam badger
''')
help = usage + '''\
positional arguments:
spam spam PROG None
badger badger PROG 0.5
optional arguments:
-h, --help show this help message and exit
-x X x PROG None int %
-y y PROG 42 XXX
--foo {a,b,c} foo PROG None a, b, c
--bar BBB bar PROG baz bar
group:
-a A a PROG None
-b B b PROG -1
'''
version = ''
class TestHelpVariableExpansionUsageSupplied(HelpTestCase):
"""Test that variables are expanded properly when usage= is present"""
parser_signature = Sig(prog='PROG', usage='%(prog)s FOO')
argument_signatures = []
argument_group_signatures = []
usage = ('''\
usage: PROG FOO
''')
help = usage + '''\
optional arguments:
-h, --help show this help message and exit
'''
version = ''
class TestHelpVariableExpansionNoArguments(HelpTestCase):
"""Test that variables are expanded properly with no arguments"""
parser_signature = Sig(prog='PROG', add_help=False)
argument_signatures = []
argument_group_signatures = []
usage = ('''\
usage: PROG
''')
help = usage
version = ''
class TestHelpSuppressUsage(HelpTestCase):
"""Test that items can be suppressed in usage messages"""
parser_signature = Sig(prog='PROG', usage=argparse.SUPPRESS)
argument_signatures = [
Sig('--foo', help='foo help'),
Sig('spam', help='spam help'),
]
argument_group_signatures = []
help = '''\
positional arguments:
spam spam help
optional arguments:
-h, --help show this help message and exit
--foo FOO foo help
'''
usage = ''
version = ''
class TestHelpSuppressOptional(HelpTestCase):
"""Test that optional arguments can be suppressed in help messages"""
parser_signature = Sig(prog='PROG', add_help=False)
argument_signatures = [
Sig('--foo', help=argparse.SUPPRESS),
Sig('spam', help='spam help'),
]
argument_group_signatures = []
usage = '''\
usage: PROG spam
'''
help = usage + '''\
positional arguments:
spam spam help
'''
version = ''
class TestHelpSuppressOptionalGroup(HelpTestCase):
"""Test that optional groups can be suppressed in help messages"""
parser_signature = Sig(prog='PROG')
argument_signatures = [
Sig('--foo', help='foo help'),
Sig('spam', help='spam help'),
]
argument_group_signatures = [
(Sig('group'), [Sig('--bar', help=argparse.SUPPRESS)]),
]
usage = '''\
usage: PROG [-h] [--foo FOO] spam
'''
help = usage + '''\
positional arguments:
spam spam help
optional arguments:
-h, --help show this help message and exit
--foo FOO foo help
'''
version = ''
class TestHelpSuppressPositional(HelpTestCase):
"""Test that positional arguments can be suppressed in help messages"""
parser_signature = Sig(prog='PROG')
argument_signatures = [
Sig('--foo', help='foo help'),
Sig('spam', help=argparse.SUPPRESS),
]
argument_group_signatures = []
usage = '''\
usage: PROG [-h] [--foo FOO]
'''
help = usage + '''\
optional arguments:
-h, --help show this help message and exit
--foo FOO foo help
'''
version = ''
class TestHelpRequiredOptional(HelpTestCase):
"""Test that required options don't look optional"""
parser_signature = Sig(prog='PROG')
argument_signatures = [
Sig('--foo', required=True, help='foo help'),
]
argument_group_signatures = []
usage = '''\
usage: PROG [-h] --foo FOO
'''
help = usage + '''\
optional arguments:
-h, --help show this help message and exit
--foo FOO foo help
'''
version = ''
class TestHelpAlternatePrefixChars(HelpTestCase):
"""Test that options display with different prefix characters"""
parser_signature = Sig(prog='PROG', prefix_chars='^;', add_help=False)
argument_signatures = [
Sig('^^foo', action='store_true', help='foo help'),
Sig(';b', ';;bar', help='bar help'),
]
argument_group_signatures = []
usage = '''\
usage: PROG [^^foo] [;b BAR]
'''
help = usage + '''\
optional arguments:
^^foo foo help
;b BAR, ;;bar BAR bar help
'''
version = ''
class TestHelpNoHelpOptional(HelpTestCase):
"""Test that the --help argument can be suppressed help messages"""
parser_signature = Sig(prog='PROG', add_help=False)
argument_signatures = [
Sig('--foo', help='foo help'),
Sig('spam', help='spam help'),
]
argument_group_signatures = []
usage = '''\
usage: PROG [--foo FOO] spam
'''
help = usage + '''\
positional arguments:
spam spam help
optional arguments:
--foo FOO foo help
'''
version = ''
class TestHelpNone(HelpTestCase):
"""Test that no errors occur if no help is specified"""
parser_signature = Sig(prog='PROG')
argument_signatures = [
Sig('--foo'),
Sig('spam'),
]
argument_group_signatures = []
usage = '''\
usage: PROG [-h] [--foo FOO] spam
'''
help = usage + '''\
positional arguments:
spam
optional arguments:
-h, --help show this help message and exit
--foo FOO
'''
version = ''
class TestHelpTupleMetavar(HelpTestCase):
"""Test specifying metavar as a tuple"""
parser_signature = Sig(prog='PROG')
argument_signatures = [
Sig('-w', help='w', nargs='+', metavar=('W1', 'W2')),
Sig('-x', help='x', nargs='*', metavar=('X1', 'X2')),
Sig('-y', help='y', nargs=3, metavar=('Y1', 'Y2', 'Y3')),
Sig('-z', help='z', nargs='?', metavar=('Z1', )),
]
argument_group_signatures = []
usage = '''\
usage: PROG [-h] [-w W1 [W2 ...]] [-x [X1 [X2 ...]]] [-y Y1 Y2 Y3] \
[-z [Z1]]
'''
help = usage + '''\
optional arguments:
-h, --help show this help message and exit
-w W1 [W2 ...] w
-x [X1 [X2 ...]] x
-y Y1 Y2 Y3 y
-z [Z1] z
'''
version = ''
class TestHelpRawText(HelpTestCase):
"""Test the RawTextHelpFormatter"""
parser_signature = Sig(
prog='PROG', formatter_class=argparse.RawTextHelpFormatter,
description='Keep the formatting\n'
' exactly as it is written\n'
'\n'
'here\n')
argument_signatures = [
Sig('--foo', help=' foo help should also\n'
'appear as given here'),
Sig('spam', help='spam help'),
]
argument_group_signatures = [
(Sig('title', description=' This text\n'
' should be indented\n'
' exactly like it is here\n'),
[Sig('--bar', help='bar help')]),
]
usage = '''\
usage: PROG [-h] [--foo FOO] [--bar BAR] spam
'''
help = usage + '''\
Keep the formatting
exactly as it is written
here
positional arguments:
spam spam help
optional arguments:
-h, --help show this help message and exit
--foo FOO foo help should also
appear as given here
title:
This text
should be indented
exactly like it is here
--bar BAR bar help
'''
version = ''
class TestHelpRawDescription(HelpTestCase):
"""Test the RawTextHelpFormatter"""
parser_signature = Sig(
prog='PROG', formatter_class=argparse.RawDescriptionHelpFormatter,
description='Keep the formatting\n'
' exactly as it is written\n'
'\n'
'here\n')
argument_signatures = [
Sig('--foo', help=' foo help should not\n'
' retain this odd formatting'),
Sig('spam', help='spam help'),
]
argument_group_signatures = [
(Sig('title', description=' This text\n'
' should be indented\n'
' exactly like it is here\n'),
[Sig('--bar', help='bar help')]),
]
usage = '''\
usage: PROG [-h] [--foo FOO] [--bar BAR] spam
'''
help = usage + '''\
Keep the formatting
exactly as it is written
here
positional arguments:
spam spam help
optional arguments:
-h, --help show this help message and exit
--foo FOO foo help should not retain this odd formatting
title:
This text
should be indented
exactly like it is here
--bar BAR bar help
'''
version = ''
class TestHelpArgumentDefaults(HelpTestCase):
"""Test the ArgumentDefaultsHelpFormatter"""
parser_signature = Sig(
prog='PROG', formatter_class=argparse.ArgumentDefaultsHelpFormatter,
description='description')
argument_signatures = [
Sig('--foo', help='foo help - oh and by the way, %(default)s'),
Sig('--bar', action='store_true', help='bar help'),
Sig('spam', help='spam help'),
Sig('badger', nargs='?', default='wooden', help='badger help'),
]
argument_group_signatures = [
(Sig('title', description='description'),
[Sig('--baz', type=int, default=42, help='baz help')]),
]
usage = '''\
usage: PROG [-h] [--foo FOO] [--bar] [--baz BAZ] spam [badger]
'''
help = usage + '''\
description
positional arguments:
spam spam help
badger badger help (default: wooden)
optional arguments:
-h, --help show this help message and exit
--foo FOO foo help - oh and by the way, None
--bar bar help (default: False)
title:
description
--baz BAZ baz help (default: 42)
'''
version = ''
class TestHelpVersionAction(HelpTestCase):
"""Test the default help for the version action"""
parser_signature = Sig(prog='PROG', description='description')
argument_signatures = [Sig('-V', '--version', action='version', version='3.6')]
argument_group_signatures = []
usage = '''\
usage: PROG [-h] [-V]
'''
help = usage + '''\
description
optional arguments:
-h, --help show this help message and exit
-V, --version show program's version number and exit
'''
version = ''
class TestHelpVersionActionSuppress(HelpTestCase):
"""Test that the --version argument can be suppressed in help messages"""
parser_signature = Sig(prog='PROG')
argument_signatures = [
Sig('-v', '--version', action='version', version='1.0',
help=argparse.SUPPRESS),
Sig('--foo', help='foo help'),
Sig('spam', help='spam help'),
]
argument_group_signatures = []
usage = '''\
usage: PROG [-h] [--foo FOO] spam
'''
help = usage + '''\
positional arguments:
spam spam help
optional arguments:
-h, --help show this help message and exit
--foo FOO foo help
'''
class TestHelpSubparsersOrdering(HelpTestCase):
"""Test ordering of subcommands in help matches the code"""
parser_signature = Sig(prog='PROG',
description='display some subcommands')
argument_signatures = [Sig('-v', '--version', action='version', version='0.1')]
subparsers_signatures = [Sig(name=name)
for name in ('a', 'b', 'c', 'd', 'e')]
usage = '''\
usage: PROG [-h] [-v] {a,b,c,d,e} ...
'''
help = usage + '''\
display some subcommands
positional arguments:
{a,b,c,d,e}
optional arguments:
-h, --help show this help message and exit
-v, --version show program's version number and exit
'''
version = '''\
0.1
'''
class TestHelpSubparsersWithHelpOrdering(HelpTestCase):
"""Test ordering of subcommands in help matches the code"""
parser_signature = Sig(prog='PROG',
description='display some subcommands')
argument_signatures = [Sig('-v', '--version', action='version', version='0.1')]
subcommand_data = (('a', 'a subcommand help'),
('b', 'b subcommand help'),
('c', 'c subcommand help'),
('d', 'd subcommand help'),
('e', 'e subcommand help'),
)
subparsers_signatures = [Sig(name=name, help=help)
for name, help in subcommand_data]
usage = '''\
usage: PROG [-h] [-v] {a,b,c,d,e} ...
'''
help = usage + '''\
display some subcommands
positional arguments:
{a,b,c,d,e}
a a subcommand help
b b subcommand help
c c subcommand help
d d subcommand help
e e subcommand help
optional arguments:
-h, --help show this help message and exit
-v, --version show program's version number and exit
'''
version = '''\
0.1
'''
class TestHelpMetavarTypeFormatter(HelpTestCase):
def custom_type(string):
return string
parser_signature = Sig(prog='PROG', description='description',
formatter_class=argparse.MetavarTypeHelpFormatter)
argument_signatures = [Sig('a', type=int),
Sig('-b', type=custom_type),
Sig('-c', type=float, metavar='SOME FLOAT')]
argument_group_signatures = []
usage = '''\
usage: PROG [-h] [-b custom_type] [-c SOME FLOAT] int
'''
help = usage + '''\
description
positional arguments:
int
optional arguments:
-h, --help show this help message and exit
-b custom_type
-c SOME FLOAT
'''
version = ''
# =====================================
# Optional/Positional constructor tests
# =====================================
class TestInvalidArgumentConstructors(TestCase):
"""Test a bunch of invalid Argument constructors"""
def assertTypeError(self, *args, **kwargs):
parser = argparse.ArgumentParser()
self.assertRaises(TypeError, parser.add_argument,
*args, **kwargs)
def assertValueError(self, *args, **kwargs):
parser = argparse.ArgumentParser()
self.assertRaises(ValueError, parser.add_argument,
*args, **kwargs)
def test_invalid_keyword_arguments(self):
self.assertTypeError('-x', bar=None)
self.assertTypeError('-y', callback='foo')
self.assertTypeError('-y', callback_args=())
self.assertTypeError('-y', callback_kwargs={})
def test_missing_destination(self):
self.assertTypeError()
for action in ['append', 'store']:
self.assertTypeError(action=action)
def test_invalid_option_strings(self):
self.assertValueError('--')
self.assertValueError('---')
def test_invalid_type(self):
self.assertValueError('--foo', type='int')
self.assertValueError('--foo', type=(int, float))
def test_invalid_action(self):
self.assertValueError('-x', action='foo')
self.assertValueError('foo', action='baz')
self.assertValueError('--foo', action=('store', 'append'))
parser = argparse.ArgumentParser()
with self.assertRaises(ValueError) as cm:
parser.add_argument("--foo", action="store-true")
self.assertIn('unknown action', str(cm.exception))
def test_multiple_dest(self):
parser = argparse.ArgumentParser()
parser.add_argument(dest='foo')
with self.assertRaises(ValueError) as cm:
parser.add_argument('bar', dest='baz')
self.assertIn('dest supplied twice for positional argument',
str(cm.exception))
def test_no_argument_actions(self):
for action in ['store_const', 'store_true', 'store_false',
'append_const', 'count']:
for attrs in [dict(type=int), dict(nargs='+'),
dict(choices='ab')]:
self.assertTypeError('-x', action=action, **attrs)
def test_no_argument_no_const_actions(self):
# options with zero arguments
for action in ['store_true', 'store_false', 'count']:
# const is always disallowed
self.assertTypeError('-x', const='foo', action=action)
# nargs is always disallowed
self.assertTypeError('-x', nargs='*', action=action)
def test_more_than_one_argument_actions(self):
for action in ['store', 'append']:
# nargs=0 is disallowed
self.assertValueError('-x', nargs=0, action=action)
self.assertValueError('spam', nargs=0, action=action)
# const is disallowed with non-optional arguments
for nargs in [1, '*', '+']:
self.assertValueError('-x', const='foo',
nargs=nargs, action=action)
self.assertValueError('spam', const='foo',
nargs=nargs, action=action)
def test_required_const_actions(self):
for action in ['store_const', 'append_const']:
# nargs is always disallowed
self.assertTypeError('-x', nargs='+', action=action)
def test_parsers_action_missing_params(self):
self.assertTypeError('command', action='parsers')
self.assertTypeError('command', action='parsers', prog='PROG')
self.assertTypeError('command', action='parsers',
parser_class=argparse.ArgumentParser)
def test_required_positional(self):
self.assertTypeError('foo', required=True)
def test_user_defined_action(self):
class Success(Exception):
pass
class Action(object):
def __init__(self,
option_strings,
dest,
const,
default,
required=False):
if dest == 'spam':
if const is Success:
if default is Success:
raise Success()
def __call__(self, *args, **kwargs):
pass
parser = argparse.ArgumentParser()
self.assertRaises(Success, parser.add_argument, '--spam',
action=Action, default=Success, const=Success)
self.assertRaises(Success, parser.add_argument, 'spam',
action=Action, default=Success, const=Success)
# ================================
# Actions returned by add_argument
# ================================
class TestActionsReturned(TestCase):
def test_dest(self):
parser = argparse.ArgumentParser()
action = parser.add_argument('--foo')
self.assertEqual(action.dest, 'foo')
action = parser.add_argument('-b', '--bar')
self.assertEqual(action.dest, 'bar')
action = parser.add_argument('-x', '-y')
self.assertEqual(action.dest, 'x')
def test_misc(self):
parser = argparse.ArgumentParser()
action = parser.add_argument('--foo', nargs='?', const=42,
default=84, type=int, choices=[1, 2],
help='FOO', metavar='BAR', dest='baz')
self.assertEqual(action.nargs, '?')
self.assertEqual(action.const, 42)
self.assertEqual(action.default, 84)
self.assertEqual(action.type, int)
self.assertEqual(action.choices, [1, 2])
self.assertEqual(action.help, 'FOO')
self.assertEqual(action.metavar, 'BAR')
self.assertEqual(action.dest, 'baz')
# ================================
# Argument conflict handling tests
# ================================
class TestConflictHandling(TestCase):
def test_bad_type(self):
self.assertRaises(ValueError, argparse.ArgumentParser,
conflict_handler='foo')
def test_conflict_error(self):
parser = argparse.ArgumentParser()
parser.add_argument('-x')
self.assertRaises(argparse.ArgumentError,
parser.add_argument, '-x')
parser.add_argument('--spam')
self.assertRaises(argparse.ArgumentError,
parser.add_argument, '--spam')
def test_resolve_error(self):
get_parser = argparse.ArgumentParser
parser = get_parser(prog='PROG', conflict_handler='resolve')
parser.add_argument('-x', help='OLD X')
parser.add_argument('-x', help='NEW X')
self.assertEqual(parser.format_help(), textwrap.dedent('''\
usage: PROG [-h] [-x X]
optional arguments:
-h, --help show this help message and exit
-x X NEW X
'''))
parser.add_argument('--spam', metavar='OLD_SPAM')
parser.add_argument('--spam', metavar='NEW_SPAM')
self.assertEqual(parser.format_help(), textwrap.dedent('''\
usage: PROG [-h] [-x X] [--spam NEW_SPAM]
optional arguments:
-h, --help show this help message and exit
-x X NEW X
--spam NEW_SPAM
'''))
# =============================
# Help and Version option tests
# =============================
class TestOptionalsHelpVersionActions(TestCase):
"""Test the help and version actions"""
def assertPrintHelpExit(self, parser, args_str):
with self.assertRaises(ArgumentParserError) as cm:
parser.parse_args(args_str.split())
self.assertEqual(parser.format_help(), cm.exception.stdout)
def assertArgumentParserError(self, parser, *args):
self.assertRaises(ArgumentParserError, parser.parse_args, args)
def test_version(self):
parser = ErrorRaisingArgumentParser()
parser.add_argument('-v', '--version', action='version', version='1.0')
self.assertPrintHelpExit(parser, '-h')
self.assertPrintHelpExit(parser, '--help')
self.assertRaises(AttributeError, getattr, parser, 'format_version')
def test_version_format(self):
parser = ErrorRaisingArgumentParser(prog='PPP')
parser.add_argument('-v', '--version', action='version', version='%(prog)s 3.5')
with self.assertRaises(ArgumentParserError) as cm:
parser.parse_args(['-v'])
self.assertEqual('PPP 3.5\n', cm.exception.stdout)
def test_version_no_help(self):
parser = ErrorRaisingArgumentParser(add_help=False)
parser.add_argument('-v', '--version', action='version', version='1.0')
self.assertArgumentParserError(parser, '-h')
self.assertArgumentParserError(parser, '--help')
self.assertRaises(AttributeError, getattr, parser, 'format_version')
def test_version_action(self):
parser = ErrorRaisingArgumentParser(prog='XXX')
parser.add_argument('-V', action='version', version='%(prog)s 3.7')
with self.assertRaises(ArgumentParserError) as cm:
parser.parse_args(['-V'])
self.assertEqual('XXX 3.7\n', cm.exception.stdout)
def test_no_help(self):
parser = ErrorRaisingArgumentParser(add_help=False)
self.assertArgumentParserError(parser, '-h')
self.assertArgumentParserError(parser, '--help')
self.assertArgumentParserError(parser, '-v')
self.assertArgumentParserError(parser, '--version')
def test_alternate_help_version(self):
parser = ErrorRaisingArgumentParser()
parser.add_argument('-x', action='help')
parser.add_argument('-y', action='version')
self.assertPrintHelpExit(parser, '-x')
self.assertArgumentParserError(parser, '-v')
self.assertArgumentParserError(parser, '--version')
self.assertRaises(AttributeError, getattr, parser, 'format_version')
def test_help_version_extra_arguments(self):
parser = ErrorRaisingArgumentParser()
parser.add_argument('--version', action='version', version='1.0')
parser.add_argument('-x', action='store_true')
parser.add_argument('y')
# try all combinations of valid prefixes and suffixes
valid_prefixes = ['', '-x', 'foo', '-x bar', 'baz -x']
valid_suffixes = valid_prefixes + ['--bad-option', 'foo bar baz']
for prefix in valid_prefixes:
for suffix in valid_suffixes:
format = '%s %%s %s' % (prefix, suffix)
self.assertPrintHelpExit(parser, format % '-h')
self.assertPrintHelpExit(parser, format % '--help')
self.assertRaises(AttributeError, getattr, parser, 'format_version')
# ======================
# str() and repr() tests
# ======================
class TestStrings(TestCase):
"""Test str() and repr() on Optionals and Positionals"""
def assertStringEqual(self, obj, result_string):
for func in [str, repr]:
self.assertEqual(func(obj), result_string)
def test_optional(self):
option = argparse.Action(
option_strings=['--foo', '-a', '-b'],
dest='b',
type='int',
nargs='+',
default=42,
choices=[1, 2, 3],
help='HELP',
metavar='METAVAR')
string = (
"Action(option_strings=['--foo', '-a', '-b'], dest='b', "
"nargs='+', const=None, default=42, type='int', "
"choices=[1, 2, 3], help='HELP', metavar='METAVAR')")
self.assertStringEqual(option, string)
def test_argument(self):
argument = argparse.Action(
option_strings=[],
dest='x',
type=float,
nargs='?',
default=2.5,
choices=[0.5, 1.5, 2.5],
help='H HH H',
metavar='MV MV MV')
string = (
"Action(option_strings=[], dest='x', nargs='?', "
"const=None, default=2.5, type=%r, choices=[0.5, 1.5, 2.5], "
"help='H HH H', metavar='MV MV MV')" % float)
self.assertStringEqual(argument, string)
def test_namespace(self):
ns = argparse.Namespace(foo=42, bar='spam')
string = "Namespace(foo=42, bar='spam')"
self.assertStringEqual(ns, string)
def test_namespace_starkwargs_notidentifier(self):
ns = argparse.Namespace(**{'"': 'quote'})
string = """Namespace(**{'"': 'quote'})"""
self.assertStringEqual(ns, string)
def test_namespace_kwargs_and_starkwargs_notidentifier(self):
ns = argparse.Namespace(a=1, **{'"': 'quote'})
string = """Namespace(a=1, **{'"': 'quote'})"""
self.assertStringEqual(ns, string)
def test_namespace_starkwargs_identifier(self):
ns = argparse.Namespace(**{'valid': True})
string = "Namespace(valid=True)"
self.assertStringEqual(ns, string)
def test_parser(self):
parser = argparse.ArgumentParser(prog='PROG')
string = (
"ArgumentParser(prog='PROG', usage=None, description=None, "
"formatter_class=%r, conflict_handler='error', "
"add_help=True)" % argparse.HelpFormatter)
self.assertStringEqual(parser, string)
# ===============
# Namespace tests
# ===============
class TestNamespace(TestCase):
def test_constructor(self):
ns = argparse.Namespace()
self.assertRaises(AttributeError, getattr, ns, 'x')
ns = argparse.Namespace(a=42, b='spam')
self.assertEqual(ns.a, 42)
self.assertEqual(ns.b, 'spam')
def test_equality(self):
ns1 = argparse.Namespace(a=1, b=2)
ns2 = argparse.Namespace(b=2, a=1)
ns3 = argparse.Namespace(a=1)
ns4 = argparse.Namespace(b=2)
self.assertEqual(ns1, ns2)
self.assertNotEqual(ns1, ns3)
self.assertNotEqual(ns1, ns4)
self.assertNotEqual(ns2, ns3)
self.assertNotEqual(ns2, ns4)
self.assertTrue(ns1 != ns3)
self.assertTrue(ns1 != ns4)
self.assertTrue(ns2 != ns3)
self.assertTrue(ns2 != ns4)
def test_equality_returns_notimplemented(self):
# See issue 21481
ns = argparse.Namespace(a=1, b=2)
self.assertIs(ns.__eq__(None), NotImplemented)
self.assertIs(ns.__ne__(None), NotImplemented)
# ===================
# File encoding tests
# ===================
class TestEncoding(TestCase):
def _test_module_encoding(self, path):
path, _ = os.path.splitext(path)
path += ".py"
with open(path, 'r', encoding='utf-8') as f:
f.read()
def test_argparse_module_encoding(self):
self._test_module_encoding(argparse.__file__)
def test_test_argparse_module_encoding(self):
self._test_module_encoding(__file__)
# ===================
# ArgumentError tests
# ===================
class TestArgumentError(TestCase):
def test_argument_error(self):
msg = "my error here"
error = argparse.ArgumentError(None, msg)
self.assertEqual(str(error), msg)
# =======================
# ArgumentTypeError tests
# =======================
class TestArgumentTypeError(TestCase):
def test_argument_type_error(self):
def spam(string):
raise argparse.ArgumentTypeError('spam!')
parser = ErrorRaisingArgumentParser(prog='PROG', add_help=False)
parser.add_argument('x', type=spam)
with self.assertRaises(ArgumentParserError) as cm:
parser.parse_args(['XXX'])
self.assertEqual('usage: PROG x\nPROG: error: argument x: spam!\n',
cm.exception.stderr)
# =========================
# MessageContentError tests
# =========================
class TestMessageContentError(TestCase):
def test_missing_argument_name_in_message(self):
parser = ErrorRaisingArgumentParser(prog='PROG', usage='')
parser.add_argument('req_pos', type=str)
parser.add_argument('-req_opt', type=int, required=True)
parser.add_argument('need_one', type=str, nargs='+')
with self.assertRaises(ArgumentParserError) as cm:
parser.parse_args([])
msg = str(cm.exception)
self.assertRegex(msg, 'req_pos')
self.assertRegex(msg, 'req_opt')
self.assertRegex(msg, 'need_one')
with self.assertRaises(ArgumentParserError) as cm:
parser.parse_args(['myXargument'])
msg = str(cm.exception)
self.assertNotIn(msg, 'req_pos')
self.assertRegex(msg, 'req_opt')
self.assertRegex(msg, 'need_one')
with self.assertRaises(ArgumentParserError) as cm:
parser.parse_args(['myXargument', '-req_opt=1'])
msg = str(cm.exception)
self.assertNotIn(msg, 'req_pos')
self.assertNotIn(msg, 'req_opt')
self.assertRegex(msg, 'need_one')
def test_optional_optional_not_in_message(self):
parser = ErrorRaisingArgumentParser(prog='PROG', usage='')
parser.add_argument('req_pos', type=str)
parser.add_argument('--req_opt', type=int, required=True)
parser.add_argument('--opt_opt', type=bool, nargs='?',
default=True)
with self.assertRaises(ArgumentParserError) as cm:
parser.parse_args([])
msg = str(cm.exception)
self.assertRegex(msg, 'req_pos')
self.assertRegex(msg, 'req_opt')
self.assertNotIn(msg, 'opt_opt')
with self.assertRaises(ArgumentParserError) as cm:
parser.parse_args(['--req_opt=1'])
msg = str(cm.exception)
self.assertRegex(msg, 'req_pos')
self.assertNotIn(msg, 'req_opt')
self.assertNotIn(msg, 'opt_opt')
def test_optional_positional_not_in_message(self):
parser = ErrorRaisingArgumentParser(prog='PROG', usage='')
parser.add_argument('req_pos')
parser.add_argument('optional_positional', nargs='?', default='eggs')
with self.assertRaises(ArgumentParserError) as cm:
parser.parse_args([])
msg = str(cm.exception)
self.assertRegex(msg, 'req_pos')
self.assertNotIn(msg, 'optional_positional')
# ================================================
# Check that the type function is called only once
# ================================================
class TestTypeFunctionCallOnlyOnce(TestCase):
def test_type_function_call_only_once(self):
def spam(string_to_convert):
self.assertEqual(string_to_convert, 'spam!')
return 'foo_converted'
parser = argparse.ArgumentParser()
parser.add_argument('--foo', type=spam, default='bar')
args = parser.parse_args('--foo spam!'.split())
self.assertEqual(NS(foo='foo_converted'), args)
# ==================================================================
# Check semantics regarding the default argument and type conversion
# ==================================================================
class TestTypeFunctionCalledOnDefault(TestCase):
def test_type_function_call_with_non_string_default(self):
def spam(int_to_convert):
self.assertEqual(int_to_convert, 0)
return 'foo_converted'
parser = argparse.ArgumentParser()
parser.add_argument('--foo', type=spam, default=0)
args = parser.parse_args([])
# foo should *not* be converted because its default is not a string.
self.assertEqual(NS(foo=0), args)
def test_type_function_call_with_string_default(self):
def spam(int_to_convert):
return 'foo_converted'
parser = argparse.ArgumentParser()
parser.add_argument('--foo', type=spam, default='0')
args = parser.parse_args([])
# foo is converted because its default is a string.
self.assertEqual(NS(foo='foo_converted'), args)
def test_no_double_type_conversion_of_default(self):
def extend(str_to_convert):
return str_to_convert + '*'
parser = argparse.ArgumentParser()
parser.add_argument('--test', type=extend, default='*')
args = parser.parse_args([])
# The test argument will be two stars, one coming from the default
# value and one coming from the type conversion being called exactly
# once.
self.assertEqual(NS(test='**'), args)
def test_issue_15906(self):
# Issue #15906: When action='append', type=str, default=[] are
# providing, the dest value was the string representation "[]" when it
# should have been an empty list.
parser = argparse.ArgumentParser()
parser.add_argument('--test', dest='test', type=str,
default=[], action='append')
args = parser.parse_args([])
self.assertEqual(args.test, [])
# ======================
# parse_known_args tests
# ======================
class TestParseKnownArgs(TestCase):
def test_arguments_tuple(self):
parser = argparse.ArgumentParser()
parser.parse_args(())
def test_arguments_list(self):
parser = argparse.ArgumentParser()
parser.parse_args([])
def test_arguments_tuple_positional(self):
parser = argparse.ArgumentParser()
parser.add_argument('x')
parser.parse_args(('x',))
def test_arguments_list_positional(self):
parser = argparse.ArgumentParser()
parser.add_argument('x')
parser.parse_args(['x'])
def test_optionals(self):
parser = argparse.ArgumentParser()
parser.add_argument('--foo')
args, extras = parser.parse_known_args('--foo F --bar --baz'.split())
self.assertEqual(NS(foo='F'), args)
self.assertEqual(['--bar', '--baz'], extras)
def test_mixed(self):
parser = argparse.ArgumentParser()
parser.add_argument('-v', nargs='?', const=1, type=int)
parser.add_argument('--spam', action='store_false')
parser.add_argument('badger')
argv = ["B", "C", "--foo", "-v", "3", "4"]
args, extras = parser.parse_known_args(argv)
self.assertEqual(NS(v=3, spam=True, badger="B"), args)
self.assertEqual(["C", "--foo", "4"], extras)
# ===========================
# parse_intermixed_args tests
# ===========================
class TestIntermixedArgs(TestCase):
def test_basic(self):
# test parsing intermixed optionals and positionals
parser = argparse.ArgumentParser(prog='PROG')
parser.add_argument('--foo', dest='foo')
bar = parser.add_argument('--bar', dest='bar', required=True)
parser.add_argument('cmd')
parser.add_argument('rest', nargs='*', type=int)
argv = 'cmd --foo x 1 --bar y 2 3'.split()
args = parser.parse_intermixed_args(argv)
# rest gets [1,2,3] despite the foo and bar strings
self.assertEqual(NS(bar='y', cmd='cmd', foo='x', rest=[1, 2, 3]), args)
args, extras = parser.parse_known_args(argv)
# cannot parse the '1,2,3'
self.assertEqual(NS(bar='y', cmd='cmd', foo='x', rest=[]), args)
self.assertEqual(["1", "2", "3"], extras)
argv = 'cmd --foo x 1 --error 2 --bar y 3'.split()
args, extras = parser.parse_known_intermixed_args(argv)
# unknown optionals go into extras
self.assertEqual(NS(bar='y', cmd='cmd', foo='x', rest=[1]), args)
self.assertEqual(['--error', '2', '3'], extras)
# restores attributes that were temporarily changed
self.assertIsNone(parser.usage)
self.assertEqual(bar.required, True)
def test_remainder(self):
# Intermixed and remainder are incompatible
parser = ErrorRaisingArgumentParser(prog='PROG')
parser.add_argument('-z')
parser.add_argument('x')
parser.add_argument('y', nargs='...')
argv = 'X A B -z Z'.split()
# intermixed fails with '...' (also 'A...')
# self.assertRaises(TypeError, parser.parse_intermixed_args, argv)
with self.assertRaises(TypeError) as cm:
parser.parse_intermixed_args(argv)
self.assertRegex(str(cm.exception), r'\.\.\.')
def test_exclusive(self):
# mutually exclusive group; intermixed works fine
parser = ErrorRaisingArgumentParser(prog='PROG')
group = parser.add_mutually_exclusive_group(required=True)
group.add_argument('--foo', action='store_true', help='FOO')
group.add_argument('--spam', help='SPAM')
parser.add_argument('badger', nargs='*', default='X', help='BADGER')
args = parser.parse_intermixed_args('1 --foo 2'.split())
self.assertEqual(NS(badger=['1', '2'], foo=True, spam=None), args)
self.assertRaises(ArgumentParserError, parser.parse_intermixed_args, '1 2'.split())
self.assertEqual(group.required, True)
def test_exclusive_incompatible(self):
# mutually exclusive group including positional - fail
parser = ErrorRaisingArgumentParser(prog='PROG')
group = parser.add_mutually_exclusive_group(required=True)
group.add_argument('--foo', action='store_true', help='FOO')
group.add_argument('--spam', help='SPAM')
group.add_argument('badger', nargs='*', default='X', help='BADGER')
self.assertRaises(TypeError, parser.parse_intermixed_args, [])
self.assertEqual(group.required, True)
class TestIntermixedMessageContentError(TestCase):
# case where Intermixed gives different error message
# error is raised by 1st parsing step
def test_missing_argument_name_in_message(self):
parser = ErrorRaisingArgumentParser(prog='PROG', usage='')
parser.add_argument('req_pos', type=str)
parser.add_argument('-req_opt', type=int, required=True)
with self.assertRaises(ArgumentParserError) as cm:
parser.parse_args([])
msg = str(cm.exception)
self.assertRegex(msg, 'req_pos')
self.assertRegex(msg, 'req_opt')
with self.assertRaises(ArgumentParserError) as cm:
parser.parse_intermixed_args([])
msg = str(cm.exception)
self.assertNotRegex(msg, 'req_pos')
self.assertRegex(msg, 'req_opt')
# ==========================
# add_argument metavar tests
# ==========================
class TestAddArgumentMetavar(TestCase):
EXPECTED_MESSAGE = "length of metavar tuple does not match nargs"
def do_test_no_exception(self, nargs, metavar):
parser = argparse.ArgumentParser()
parser.add_argument("--foo", nargs=nargs, metavar=metavar)
def do_test_exception(self, nargs, metavar):
parser = argparse.ArgumentParser()
with self.assertRaises(ValueError) as cm:
parser.add_argument("--foo", nargs=nargs, metavar=metavar)
self.assertEqual(cm.exception.args[0], self.EXPECTED_MESSAGE)
# Unit tests for different values of metavar when nargs=None
def test_nargs_None_metavar_string(self):
self.do_test_no_exception(nargs=None, metavar="1")
def test_nargs_None_metavar_length0(self):
self.do_test_exception(nargs=None, metavar=tuple())
def test_nargs_None_metavar_length1(self):
self.do_test_no_exception(nargs=None, metavar=("1",))
def test_nargs_None_metavar_length2(self):
self.do_test_exception(nargs=None, metavar=("1", "2"))
def test_nargs_None_metavar_length3(self):
self.do_test_exception(nargs=None, metavar=("1", "2", "3"))
# Unit tests for different values of metavar when nargs=?
def test_nargs_optional_metavar_string(self):
self.do_test_no_exception(nargs="?", metavar="1")
def test_nargs_optional_metavar_length0(self):
self.do_test_exception(nargs="?", metavar=tuple())
def test_nargs_optional_metavar_length1(self):
self.do_test_no_exception(nargs="?", metavar=("1",))
def test_nargs_optional_metavar_length2(self):
self.do_test_exception(nargs="?", metavar=("1", "2"))
def test_nargs_optional_metavar_length3(self):
self.do_test_exception(nargs="?", metavar=("1", "2", "3"))
# Unit tests for different values of metavar when nargs=*
def test_nargs_zeroormore_metavar_string(self):
self.do_test_no_exception(nargs="*", metavar="1")
def test_nargs_zeroormore_metavar_length0(self):
self.do_test_exception(nargs="*", metavar=tuple())
def test_nargs_zeroormore_metavar_length1(self):
self.do_test_no_exception(nargs="*", metavar=("1",))
def test_nargs_zeroormore_metavar_length2(self):
self.do_test_no_exception(nargs="*", metavar=("1", "2"))
def test_nargs_zeroormore_metavar_length3(self):
self.do_test_exception(nargs="*", metavar=("1", "2", "3"))
# Unit tests for different values of metavar when nargs=+
def test_nargs_oneormore_metavar_string(self):
self.do_test_no_exception(nargs="+", metavar="1")
def test_nargs_oneormore_metavar_length0(self):
self.do_test_exception(nargs="+", metavar=tuple())
def test_nargs_oneormore_metavar_length1(self):
self.do_test_exception(nargs="+", metavar=("1",))
def test_nargs_oneormore_metavar_length2(self):
self.do_test_no_exception(nargs="+", metavar=("1", "2"))
def test_nargs_oneormore_metavar_length3(self):
self.do_test_exception(nargs="+", metavar=("1", "2", "3"))
# Unit tests for different values of metavar when nargs=...
def test_nargs_remainder_metavar_string(self):
self.do_test_no_exception(nargs="...", metavar="1")
def test_nargs_remainder_metavar_length0(self):
self.do_test_no_exception(nargs="...", metavar=tuple())
def test_nargs_remainder_metavar_length1(self):
self.do_test_no_exception(nargs="...", metavar=("1",))
def test_nargs_remainder_metavar_length2(self):
self.do_test_no_exception(nargs="...", metavar=("1", "2"))
def test_nargs_remainder_metavar_length3(self):
self.do_test_no_exception(nargs="...", metavar=("1", "2", "3"))
# Unit tests for different values of metavar when nargs=A...
def test_nargs_parser_metavar_string(self):
self.do_test_no_exception(nargs="A...", metavar="1")
def test_nargs_parser_metavar_length0(self):
self.do_test_exception(nargs="A...", metavar=tuple())
def test_nargs_parser_metavar_length1(self):
self.do_test_no_exception(nargs="A...", metavar=("1",))
def test_nargs_parser_metavar_length2(self):
self.do_test_exception(nargs="A...", metavar=("1", "2"))
def test_nargs_parser_metavar_length3(self):
self.do_test_exception(nargs="A...", metavar=("1", "2", "3"))
# Unit tests for different values of metavar when nargs=1
def test_nargs_1_metavar_string(self):
self.do_test_no_exception(nargs=1, metavar="1")
def test_nargs_1_metavar_length0(self):
self.do_test_exception(nargs=1, metavar=tuple())
def test_nargs_1_metavar_length1(self):
self.do_test_no_exception(nargs=1, metavar=("1",))
def test_nargs_1_metavar_length2(self):
self.do_test_exception(nargs=1, metavar=("1", "2"))
def test_nargs_1_metavar_length3(self):
self.do_test_exception(nargs=1, metavar=("1", "2", "3"))
# Unit tests for different values of metavar when nargs=2
def test_nargs_2_metavar_string(self):
self.do_test_no_exception(nargs=2, metavar="1")
def test_nargs_2_metavar_length0(self):
self.do_test_exception(nargs=2, metavar=tuple())
def test_nargs_2_metavar_length1(self):
self.do_test_exception(nargs=2, metavar=("1",))
def test_nargs_2_metavar_length2(self):
self.do_test_no_exception(nargs=2, metavar=("1", "2"))
def test_nargs_2_metavar_length3(self):
self.do_test_exception(nargs=2, metavar=("1", "2", "3"))
# Unit tests for different values of metavar when nargs=3
def test_nargs_3_metavar_string(self):
self.do_test_no_exception(nargs=3, metavar="1")
def test_nargs_3_metavar_length0(self):
self.do_test_exception(nargs=3, metavar=tuple())
def test_nargs_3_metavar_length1(self):
self.do_test_exception(nargs=3, metavar=("1",))
def test_nargs_3_metavar_length2(self):
self.do_test_exception(nargs=3, metavar=("1", "2"))
def test_nargs_3_metavar_length3(self):
self.do_test_no_exception(nargs=3, metavar=("1", "2", "3"))
class TestInvalidNargs(TestCase):
EXPECTED_INVALID_MESSAGE = "invalid nargs value"
EXPECTED_RANGE_MESSAGE = ("nargs for store actions must be != 0; if you "
"have nothing to store, actions such as store "
"true or store const may be more appropriate")
def do_test_range_exception(self, nargs):
parser = argparse.ArgumentParser()
with self.assertRaises(ValueError) as cm:
parser.add_argument("--foo", nargs=nargs)
self.assertEqual(cm.exception.args[0], self.EXPECTED_RANGE_MESSAGE)
def do_test_invalid_exception(self, nargs):
parser = argparse.ArgumentParser()
with self.assertRaises(ValueError) as cm:
parser.add_argument("--foo", nargs=nargs)
self.assertEqual(cm.exception.args[0], self.EXPECTED_INVALID_MESSAGE)
# Unit tests for different values of nargs
def test_nargs_alphabetic(self):
self.do_test_invalid_exception(nargs='a')
self.do_test_invalid_exception(nargs="abcd")
def test_nargs_zero(self):
self.do_test_range_exception(nargs=0)
# ============================
# from argparse import * tests
# ============================
class TestImportStar(TestCase):
def test(self):
for name in argparse.__all__:
self.assertTrue(hasattr(argparse, name))
def test_all_exports_everything_but_modules(self):
items = [
name
for name, value in vars(argparse).items()
if not (name.startswith("_") or name == 'ngettext')
if not inspect.ismodule(value)
]
self.assertEqual(sorted(items), sorted(argparse.__all__))
class TestWrappingMetavar(TestCase):
def setUp(self):
super().setUp()
self.parser = ErrorRaisingArgumentParser(
'this_is_spammy_prog_with_a_long_name_sorry_about_the_name'
)
# this metavar was triggering library assertion errors due to usage
# message formatting incorrectly splitting on the ] chars within
metavar = '<http[s]://example:1234>'
self.parser.add_argument('--proxy', metavar=metavar)
def test_help_with_metavar(self):
help_text = self.parser.format_help()
self.assertEqual(help_text, textwrap.dedent('''\
usage: this_is_spammy_prog_with_a_long_name_sorry_about_the_name
[-h] [--proxy <http[s]://example:1234>]
optional arguments:
-h, --help show this help message and exit
--proxy <http[s]://example:1234>
'''))
class TestExitOnError(TestCase):
def setUp(self):
self.parser = argparse.ArgumentParser(exit_on_error=False)
self.parser.add_argument('--integers', metavar='N', type=int)
def test_exit_on_error_with_good_args(self):
ns = self.parser.parse_args('--integers 4'.split())
self.assertEqual(ns, argparse.Namespace(integers=4))
def test_exit_on_error_with_bad_args(self):
with self.assertRaises(argparse.ArgumentError):
self.parser.parse_args('--integers a'.split())
def test_main():
support.run_unittest(__name__)
# Remove global references to avoid looking like we have refleaks.
RFile.seen = {}
WFile.seen = set()
if __name__ == '__main__':
test_main()