BUG: Use return codes properly, add --count (#1558)

* BUG: Use return codes properly, add --count

* FIX: EX_USAGE (kind of)

* FIX: EX_USAGE (really)

* FIX: Dont bother importing

* FIX: More

* STY: Caps

* MAINT: Version bump
This commit is contained in:
Eric Larson
2020-06-16 12:03:47 -04:00
committed by GitHub
parent 4f895732b4
commit 00f8ffd23d
2 changed files with 172 additions and 118 deletions

View File

@ -32,7 +32,7 @@ encodings = ('utf-8', 'iso-8859-1')
USAGE = """
\t%prog [OPTIONS] [file1 file2 ... fileN]
"""
VERSION = '1.18.dev0'
VERSION = '2.0.dev0'
# Users might want to link this file into /usr/local/bin, so we resolve the
# symbolic link path to the real path if necessary.
@ -51,6 +51,12 @@ _builtin_dictionaries = (
)
_builtin_default = 'clear,rare'
# docs say os.EX_USAGE et al. are only available on Unix systems, so to be safe
# we protect and just use the values they are on macOS and Linux
EX_OK = 0
EX_USAGE = 64
EX_DATAERR = 65
# OPTIONS:
#
# ARGUMENTS:
@ -254,7 +260,7 @@ def parse_options(args):
parser.add_argument('-D', '--dictionary',
action='append',
help='Custom dictionary file that contains spelling '
help='custom dictionary file that contains spelling '
'corrections. If this flag is not specified or '
'equals "-" then the default dictionary is used. '
'This option can be specified multiple times.')
@ -263,24 +269,24 @@ def parse_options(args):
parser.add_argument('--builtin',
dest='builtin', default=_builtin_default,
metavar='BUILTIN-LIST',
help='Comma-separated list of builtin dictionaries '
help='comma-separated list of builtin dictionaries '
'to include (when "-D -" or no "-D" is passed). '
'Current options are:' + builtin_opts + '\n'
'The default is %(default)r.')
parser.add_argument('-I', '--ignore-words',
action='append', metavar='FILE',
help='File that contains words which will be ignored '
help='file that contains words which will be ignored '
'by codespell. File must contain 1 word per line.'
' Words are case sensitive based on how they are '
'written in the dictionary file')
parser.add_argument('-L', '--ignore-words-list',
action='append', metavar='WORDS',
help='Comma separated list of words to be ignored '
help='comma separated list of words to be ignored '
'by codespell. Words are case sensitive based on '
'how they are written in the dictionary file')
parser.add_argument('-r', '--regex',
action='store', type=str,
help='Regular expression which is used to find words. '
help='regular expression which is used to find words. '
'By default any alphanumeric character, the '
'underscore, the hyphen, and the apostrophe is '
'used to build words. This option cannot be '
@ -289,9 +295,14 @@ def parse_options(args):
action='store_true', default=False,
help='print summary of fixes')
parser.add_argument('--count',
action='store_true', default=False,
help='print the number of errors as the last line of '
'stderr')
parser.add_argument('-S', '--skip',
action='append',
help='Comma-separated list of files to skip. It '
help='comma-separated list of files to skip. It '
'accepts globs as well. E.g.: if you want '
'codespell to skip .eps and .txt files, '
'you\'d give "*.eps,*.txt" to this option.')
@ -302,7 +313,7 @@ def parse_options(args):
parser.add_argument('-i', '--interactive',
action='store', type=int, default=0,
help='Set interactive mode when writing changes:\n'
help='set interactive mode when writing changes:\n'
'- 0: no interactivity.\n'
'- 1: ask for confirmation.\n'
'- 2: ask user to choose one fix when more than one is available.\n' # noqa: E501
@ -310,7 +321,7 @@ def parse_options(args):
parser.add_argument('-q', '--quiet-level',
action='store', type=int, default=2,
help='Bitmask that allows suppressing messages:\n'
help='bitmask that allows suppressing messages:\n'
'- 0: print all messages.\n'
'- 1: disable warnings about wrong encoding.\n'
'- 2: disable warnings about binary files.\n'
@ -324,7 +335,7 @@ def parse_options(args):
parser.add_argument('-e', '--hard-encoding-detection',
action='store_true', default=False,
help='Use chardet to detect the encoding of each '
help='use chardet to detect the encoding of each '
'file. This can slow down codespell, but is more '
'reliable in detecting encodings other than '
'utf-8, iso8859-1, and ascii.')
@ -335,7 +346,7 @@ def parse_options(args):
parser.add_argument('-H', '--check-hidden',
action='store_true', default=False,
help='Check hidden files and directories (those '
help='check hidden files and directories (those '
'starting with ".") as well.')
parser.add_argument('-A', '--after-context', type=int, metavar='LINES',
help='print LINES of trailing context')
@ -642,7 +653,7 @@ def main(*args):
print("ERROR: --write-changes cannot be used together with "
"--regex")
parser.print_help()
return 1
return EX_USAGE
word_regex = options.regex or word_regex_def
try:
word_regex = re.compile(word_regex)
@ -650,7 +661,7 @@ def main(*args):
print("ERROR: invalid regular expression \"%s\" (%s)" %
(word_regex, err), file=sys.stderr)
parser.print_help()
return 1
return EX_USAGE
ignore_words_files = options.ignore_words or []
ignore_words = set()
@ -659,7 +670,7 @@ def main(*args):
print("ERROR: cannot find ignore-words file: %s" %
ignore_words_file, file=sys.stderr)
parser.print_help()
return 1
return EX_USAGE
build_ignore_words(ignore_words_file, ignore_words)
ignore_words_list = options.ignore_words_list or []
@ -687,13 +698,13 @@ def main(*args):
print("ERROR: Unknown builtin dictionary: %s" % (u,),
file=sys.stderr)
parser.print_help()
return 1
return EX_USAGE
else:
if not os.path.isfile(dictionary):
print("ERROR: cannot find dictionary file: %s" % dictionary,
file=sys.stderr)
parser.print_help()
return 1
return EX_USAGE
use_dictionaries.append(dictionary)
misspellings = dict()
for dictionary in use_dictionaries:
@ -714,7 +725,7 @@ def main(*args):
print("ERROR: --context/-C cannot be used together with "
"--context-before/-B or --context-after/-A")
parser.print_help()
return 1
return EX_USAGE
context_both = max(0, options.context)
context = (context_both, context_both)
elif (options.before_context is not None) or \
@ -772,4 +783,6 @@ def main(*args):
if summary:
print("\n-------8<-------\nSUMMARY:")
print(summary)
return bad_count
if options.count:
print(bad_count, file=sys.stderr)
return EX_DATAERR if bad_count else EX_OK

View File

@ -3,6 +3,7 @@
from __future__ import print_function
import contextlib
import inspect
import os
import os.path as op
from shutil import copyfile
@ -11,18 +12,54 @@ import sys
import pytest
import codespell_lib as cs
import codespell_lib as cs_
from codespell_lib._codespell import EX_USAGE, EX_OK, EX_DATAERR
def test_constants():
"""Test our EX constants."""
assert EX_OK == 0
assert EX_USAGE == 64
assert EX_DATAERR == 65
class MainWrapper(object):
"""Compatibility wrapper for when we used to return the count."""
def main(self, *args, count=True, std=False, **kwargs):
if count:
args = ('--count',) + args
code = cs_.main(*args, **kwargs)
capsys = inspect.currentframe().f_back.f_locals['capsys']
stdout, stderr = capsys.readouterr()
assert code in (EX_OK, EX_USAGE, EX_DATAERR)
if code == EX_DATAERR: # have some misspellings
code = int(stderr.split('\n')[-2])
elif code == EX_OK and count:
code = int(stderr.split('\n')[-2])
assert code == 0
if std:
return (code, stdout, stderr)
else:
return code
cs = MainWrapper()
def run_codespell(args=(), cwd=None):
"""Helper to run codespell"""
return subprocess.Popen(
"""Run codespell."""
args = ('--count',) + args
proc = subprocess.Popen(
['codespell'] + list(args), cwd=cwd,
stdout=subprocess.PIPE, stderr=subprocess.PIPE).wait()
stdout=subprocess.PIPE, stderr=subprocess.PIPE)
stderr = proc.communicate()[1].decode('utf-8')
count = int(stderr.split('\n')[-2])
return count
def test_command(tmpdir):
"""Test running the codespell executable"""
"""Test running the codespell executable."""
# With no arguments does "."
d = str(tmpdir)
assert run_codespell(cwd=d) == 0
@ -32,13 +69,14 @@ def test_command(tmpdir):
def test_basic(tmpdir, capsys):
"""Test some basic functionality"""
"""Test some basic functionality."""
assert cs.main('_does_not_exist_') == 0
fname = op.join(str(tmpdir), 'tmp')
with open(fname, 'w') as f:
pass
assert cs.main('-D', 'foo', f.name) == 1, 'missing dictionary'
assert 'cannot find dictionary' in capsys.readouterr()[1]
code, _, stderr = cs.main('-D', 'foo', f.name, std=True)
assert code == EX_USAGE, 'missing dictionary'
assert 'cannot find dictionary' in stderr
assert cs.main(fname) == 0, 'empty file'
with open(fname, 'a') as f:
f.write('this is a test file\n')
@ -53,22 +91,21 @@ def test_basic(tmpdir, capsys):
f.write('tim\ngonna\n')
assert cs.main(fname) == 2, 'with a name'
assert cs.main('--builtin', 'clear,rare,names,informal', fname) == 4
capsys.readouterr()
assert cs.main(fname, '--builtin', 'foo') == 1 # bad type sys.exit(1)
stdout = capsys.readouterr()[1]
assert 'Unknown builtin dictionary' in stdout
code, _, stderr = cs.main(fname, '--builtin', 'foo', std=True)
assert code == EX_USAGE # bad type
assert 'Unknown builtin dictionary' in stderr
d = str(tmpdir)
assert cs.main(fname, '-D', op.join(d, 'foo')) == 1 # bad dict
stdout = capsys.readouterr()[1]
assert 'cannot find dictionary' in stdout
code, _, stderr = cs.main(fname, '-D', op.join(d, 'foo'), std=True)
assert code == EX_USAGE # bad dict
assert 'cannot find dictionary' in stderr
os.remove(fname)
with open(op.join(d, 'bad.txt'), 'w') as f:
f.write('abandonned\nAbandonned\nABANDONNED\nAbAnDoNnEd')
assert cs.main(d) == 4
capsys.readouterr()
assert cs.main('-w', d) == 0
assert 'FIXED:' in capsys.readouterr()[1]
code, _, stderr = cs.main('-w', d, std=True)
assert code == 0
assert 'FIXED:' in stderr
with open(op.join(d, 'bad.txt')) as f:
new_content = f.read()
assert cs.main(d) == 0
@ -77,9 +114,10 @@ def test_basic(tmpdir, capsys):
with open(op.join(d, 'bad.txt'), 'w') as f:
f.write('abandonned abandonned\n')
assert cs.main(d) == 2
capsys.readouterr()
assert cs.main('-q', '16', '-w', d) == 0
assert capsys.readouterr() == ('', '')
code, stdout, stderr = cs.main(
'-q', '16', '-w', d, count=False, std=True)
assert code == 0
assert stdout == stderr == ''
assert cs.main(d) == 0
# empty directory
@ -101,8 +139,9 @@ def test_interactivity(tmpdir, capsys):
with FakeStdin('y\n'):
assert cs.main('-i', '3', f.name) == 1
with FakeStdin('n\n'):
assert cs.main('-w', '-i', '3', f.name) == 0
assert '==>' in capsys.readouterr()[0]
code, stdout, _ = cs.main('-w', '-i', '3', f.name, std=True)
assert code == 0
assert '==>' in stdout
with FakeStdin('x\ny\n'):
assert cs.main('-w', '-i', '3', f.name) == 0
assert cs.main(f.name) == 0
@ -141,10 +180,10 @@ def test_interactivity(tmpdir, capsys):
with open(f.name, 'w') as f:
f.write('ackward\n')
assert cs.main(f.name) == 1
capsys.readouterr()
with FakeStdin('x\n1\n'): # blank input -> nothing
assert cs.main('-w', '-i', '3', f.name) == 0
assert 'a valid option' in capsys.readouterr()[0]
code, stdout, _ = cs.main('-w', '-i', '3', f.name, std=True)
assert code == 0
assert 'a valid option' in stdout
assert cs.main(f.name) == 0
with open(f.name, 'r') as f:
assert f.read() == 'backward\n'
@ -153,30 +192,28 @@ def test_interactivity(tmpdir, capsys):
def test_summary(tmpdir, capsys):
"""Test summary functionality"""
"""Test summary functionality."""
with open(op.join(str(tmpdir), 'tmp'), 'w') as f:
pass
try:
cs.main(f.name)
assert capsys.readouterr() == ('', ''), 'no output'
cs.main(f.name, '--summary')
stdout, stderr = capsys.readouterr()
assert stderr == ''
assert 'SUMMARY' in stdout
assert len(stdout.split('\n')) == 5
with open(f.name, 'w') as f:
f.write('abandonned\nabandonned')
cs.main(f.name, '--summary')
stdout, stderr = capsys.readouterr()
assert stderr == ''
assert 'SUMMARY' in stdout
assert len(stdout.split('\n')) == 7
assert 'abandonned' in stdout.split()[-2]
finally:
os.remove(f.name)
code, stdout, stderr = cs.main(f.name, std=True, count=False)
assert code == 0
assert stdout == stderr == '', 'no output'
code, stdout, stderr = cs.main(f.name, '--summary', std=True)
assert code == 0
assert stderr == '0\n'
assert 'SUMMARY' in stdout
assert len(stdout.split('\n')) == 5
with open(f.name, 'w') as f:
f.write('abandonned\nabandonned')
assert code == 0
code, stdout, stderr = cs.main(f.name, '--summary', std=True)
assert stderr == '2\n'
assert 'SUMMARY' in stdout
assert len(stdout.split('\n')) == 7
assert 'abandonned' in stdout.split()[-2]
def test_ignore_dictionary(tmpdir):
def test_ignore_dictionary(tmpdir, capsys):
"""Test ignore dictionary functionality."""
d = str(tmpdir)
with open(op.join(d, 'bad.txt'), 'w') as f:
@ -188,8 +225,8 @@ def test_ignore_dictionary(tmpdir):
assert cs.main('-I', f.name, bad_name) == 1
def test_ignore_word_list(tmpdir):
"""Test ignore word list functionality"""
def test_ignore_word_list(tmpdir, capsys):
"""Test ignore word list functionality."""
d = str(tmpdir)
with open(op.join(d, 'bad.txt'), 'w') as f:
f.write('abandonned\nabondon\nabilty\n')
@ -204,12 +241,12 @@ def test_custom_regex(tmpdir, capsys):
f.write('abandonned_abondon\n')
assert cs.main(d) == 0
assert cs.main('-r', "[a-z]+", d) == 2
capsys.readouterr()
assert cs.main('-r', '[a-z]+', '--write-changes', d) == 1
assert 'ERROR:' in capsys.readouterr()[0]
code, stdout, _ = cs.main('-r', '[a-z]+', '--write-changes', d, std=True)
assert code == EX_USAGE
assert 'ERROR:' in stdout
def test_exclude_file(tmpdir):
def test_exclude_file(tmpdir, capsys):
"""Test exclude file functionality."""
d = str(tmpdir)
with open(op.join(d, 'bad.txt'), 'wb') as f:
@ -229,28 +266,26 @@ def test_encoding(tmpdir, capsys):
pass
# with CaptureStdout() as sio:
assert cs.main(f.name) == 0
try:
with open(f.name, 'wb') as f:
f.write(u'naïve\n'.encode('utf-8'))
assert cs.main(f.name) == 0
assert cs.main('-e', f.name) == 0
with open(f.name, 'ab') as f:
f.write(u'naieve\n'.encode('utf-8'))
assert cs.main(f.name) == 1
# Binary file warning
with open(f.name, 'wb') as f:
f.write(b'\x00\x00naiive\x00\x00')
capsys.readouterr()
assert cs.main(f.name) == 0
assert capsys.readouterr() == ('', '')
assert cs.main('-q', '0', f.name) == 0
assert 'WARNING: Binary file' in capsys.readouterr()[1]
assert capsys.readouterr() == ('', '')
finally:
os.remove(f.name)
with open(f.name, 'wb') as f:
f.write(u'naïve\n'.encode('utf-8'))
assert cs.main(f.name) == 0
assert cs.main('-e', f.name) == 0
with open(f.name, 'ab') as f:
f.write(u'naieve\n'.encode('utf-8'))
assert cs.main(f.name) == 1
# Binary file warning
with open(f.name, 'wb') as f:
f.write(b'\x00\x00naiive\x00\x00')
code, stdout, stderr = cs.main(f.name, std=True, count=False)
assert code == 0
assert stdout == stderr == ''
code, stdout, stderr = cs.main('-q', '0', f.name, std=True, count=False)
assert code == 0
assert stdout == ''
assert 'WARNING: Binary file' in stderr
def test_ignore(tmpdir):
def test_ignore(tmpdir, capsys):
"""Test ignoring of files and directories."""
d = str(tmpdir)
with open(op.join(d, 'good.txt'), 'w') as f:
@ -272,7 +307,7 @@ def test_ignore(tmpdir):
assert cs.main('--skip=*ignoredir/bad*', d) == 1
def test_check_filename(tmpdir):
def test_check_filename(tmpdir, capsys):
"""Test filename check."""
d = str(tmpdir)
# Empty file
@ -291,7 +326,7 @@ def test_check_filename(tmpdir):
@pytest.mark.skipif((not hasattr(os, "mkfifo") or not callable(os.mkfifo)),
reason='requires os.mkfifo')
def test_check_filename_irregular_file(tmpdir):
def test_check_filename_irregular_file(tmpdir, capsys):
"""Test irregular file filename check."""
# Irregular file (!isfile())
d = str(tmpdir)
@ -300,7 +335,7 @@ def test_check_filename_irregular_file(tmpdir):
d = str(tmpdir)
def test_check_hidden(tmpdir):
def test_check_hidden(tmpdir, capsys):
"""Test ignoring of hidden files."""
d = str(tmpdir)
# visible file
@ -342,17 +377,16 @@ def test_case_handling(tmpdir, capsys):
pass
# with CaptureStdout() as sio:
assert cs.main(f.name) == 0
try:
with open(f.name, 'wb') as f:
f.write('this has an ACII error'.encode('utf-8'))
assert cs.main(f.name) == 1
assert 'ASCII' in capsys.readouterr()[0]
assert cs.main('-w', f.name) == 0
assert 'FIXED' in capsys.readouterr()[1]
with open(f.name, 'rb') as f:
assert f.read().decode('utf-8') == 'this has an ASCII error'
finally:
os.remove(f.name)
with open(f.name, 'wb') as f:
f.write('this has an ACII error'.encode('utf-8'))
code, stdout, _ = cs.main(f.name, std=True)
assert code == 1
assert 'ASCII' in stdout
code, _, stderr = cs.main('-w', f.name, std=True)
assert code == 0
assert 'FIXED' in stderr
with open(f.name, 'rb') as f:
assert f.read().decode('utf-8') == 'this has an ASCII error'
def test_context(tmpdir, capsys):
@ -362,16 +396,18 @@ def test_context(tmpdir, capsys):
f.write('line 1\nline 2\nline 3 abandonned\nline 4\nline 5')
# symmetric context, fully within file
cs.main('-C', '1', d)
lines = capsys.readouterr()[0].split('\n')
code, stdout, _ = cs.main('-C', '1', d, std=True)
assert code == 1
lines = stdout.split('\n')
assert len(lines) == 5
assert lines[0] == ': line 2'
assert lines[1] == '> line 3 abandonned'
assert lines[2] == ': line 4'
# requested context is bigger than the file
cs.main('-C', '10', d)
lines = capsys.readouterr()[0].split('\n')
code, stdout, _ = cs.main('-C', '10', d, std=True)
assert code == 1
lines = stdout.split('\n')
assert len(lines) == 7
assert lines[0] == ': line 1'
assert lines[1] == ': line 2'
@ -380,23 +416,26 @@ def test_context(tmpdir, capsys):
assert lines[4] == ': line 5'
# only before context
cs.main('-B', '2', d)
lines = capsys.readouterr()[0].split('\n')
code, stdout, _ = cs.main('-B', '2', d, std=True)
assert code == 1
lines = stdout.split('\n')
assert len(lines) == 5
assert lines[0] == ': line 1'
assert lines[1] == ': line 2'
assert lines[2] == '> line 3 abandonned'
# only after context
cs.main('-A', '1', d)
lines = capsys.readouterr()[0].split('\n')
code, stdout, _ = cs.main('-A', '1', d, std=True)
assert code == 1
lines = stdout.split('\n')
assert len(lines) == 4
assert lines[0] == '> line 3 abandonned'
assert lines[1] == ': line 4'
# asymmetric context
cs.main('-B', '2', '-A', '1', d)
lines = capsys.readouterr()[0].split('\n')
code, stdout, _ = cs.main('-B', '2', '-A', '1', d, std=True)
assert code == 1
lines = stdout.split('\n')
assert len(lines) == 6
assert lines[0] == ': line 1'
assert lines[1] == ': line 2'
@ -404,13 +443,15 @@ def test_context(tmpdir, capsys):
assert lines[3] == ': line 4'
# both '-C' and '-A' on the command line
cs.main('-C', '2', '-A', '1', d)
lines = capsys.readouterr()[0].split('\n')
code, stdout, _ = cs.main('-C', '2', '-A', '1', d, std=True)
assert code == EX_USAGE
lines = stdout.split('\n')
assert 'ERROR' in lines[0]
# both '-C' and '-B' on the command line
cs.main('-C', '2', '-B', '1', d)
lines = capsys.readouterr()[0].split('\n')
code, stdout, stderr = cs.main('-C', '2', '-B', '1', d, std=True)
assert code == EX_USAGE
lines = stdout.split('\n')
assert 'ERROR' in lines[0]