mirror of
https://github.com/zulip/zulip.git
synced 2025-11-05 06:23:38 +00:00
The ' # NOQA' comment is added so that the unused imports check (F401) can still be run while still having the Mypy type check.
491 lines
23 KiB
Python
491 lines
23 KiB
Python
from __future__ import print_function
|
|
from __future__ import absolute_import
|
|
|
|
import os
|
|
import re
|
|
import traceback
|
|
|
|
from .printer import print_err, colors
|
|
|
|
from typing import cast, Any, Callable, Dict, List, Optional, Tuple
|
|
|
|
def build_custom_checkers(by_lang):
|
|
# type: (Dict[str, List[str]]) -> Tuple[Callable[[], bool], Callable[[], bool]]
|
|
RuleList = List[Dict[str, Any]]
|
|
|
|
def custom_check_file(fn, identifier, rules, skip_rules=None, max_length=None):
|
|
# type: (str, str, RuleList, Optional[Any], Optional[int]) -> bool
|
|
failed = False
|
|
color = next(colors)
|
|
|
|
line_tups = []
|
|
for i, line in enumerate(open(fn)):
|
|
line_newline_stripped = line.strip('\n')
|
|
line_fully_stripped = line_newline_stripped.strip()
|
|
skip = False
|
|
for rule in skip_rules or []:
|
|
if re.match(rule, line):
|
|
skip = True
|
|
if line_fully_stripped.endswith(' # nolint'):
|
|
continue
|
|
if skip:
|
|
continue
|
|
tup = (i, line, line_newline_stripped, line_fully_stripped)
|
|
line_tups.append(tup)
|
|
|
|
rules_to_apply = []
|
|
fn_dirname = os.path.dirname(fn)
|
|
for rule in rules:
|
|
exclude_list = rule.get('exclude', set())
|
|
if fn in exclude_list or fn_dirname in exclude_list:
|
|
continue
|
|
if rule.get("include_only"):
|
|
found = False
|
|
for item in rule.get("include_only", set()):
|
|
if item in fn:
|
|
found = True
|
|
if not found:
|
|
continue
|
|
rules_to_apply.append(rule)
|
|
|
|
for rule in rules_to_apply:
|
|
exclude_lines = {
|
|
line for
|
|
(exclude_fn, line) in rule.get('exclude_line', set())
|
|
if exclude_fn == fn
|
|
}
|
|
|
|
pattern = rule['pattern']
|
|
for (i, line, line_newline_stripped, line_fully_stripped) in line_tups:
|
|
if line_fully_stripped in exclude_lines:
|
|
exclude_lines.remove(line_fully_stripped)
|
|
continue
|
|
try:
|
|
line_to_check = line_fully_stripped
|
|
if rule.get('strip') is not None:
|
|
if rule['strip'] == '\n':
|
|
line_to_check = line_newline_stripped
|
|
else:
|
|
raise Exception("Invalid strip rule")
|
|
if re.search(pattern, line_to_check):
|
|
print_err(identifier, color, '{} at {} line {}:'.format(
|
|
rule['description'], fn, i+1))
|
|
print_err(identifier, color, line)
|
|
failed = True
|
|
except Exception:
|
|
print("Exception with %s at %s line %s" % (rule['pattern'], fn, i+1))
|
|
traceback.print_exc()
|
|
|
|
if exclude_lines:
|
|
print('Please remove exclusions for file %s: %s' % (fn, exclude_lines))
|
|
|
|
lastLine = None
|
|
for (i, line, line_newline_stripped, line_fully_stripped) in line_tups:
|
|
if isinstance(line, bytes):
|
|
line_length = len(line.decode("utf-8"))
|
|
else:
|
|
line_length = len(line)
|
|
if (max_length is not None and line_length > max_length and
|
|
'# type' not in line and 'test' not in fn and 'example' not in fn and
|
|
not re.match("\[[ A-Za-z0-9_:,&()-]*\]: http.*", line) and
|
|
not re.match("`\{\{ external_api_uri_subdomain \}\}[^`]+`", line) and
|
|
"#ignorelongline" not in line and 'migrations' not in fn):
|
|
print("Line too long (%s) at %s line %s: %s" % (len(line), fn, i+1, line_newline_stripped))
|
|
failed = True
|
|
lastLine = line
|
|
|
|
if lastLine and ('\n' not in lastLine):
|
|
print("No newline at the end of file. Fix with `sed -i '$a\\' %s`" % (fn,))
|
|
failed = True
|
|
|
|
return failed
|
|
|
|
whitespace_rules = [
|
|
# This linter should be first since bash_rules depends on it.
|
|
{'pattern': '\s+$',
|
|
'strip': '\n',
|
|
'description': 'Fix trailing whitespace'},
|
|
{'pattern': '\t',
|
|
'strip': '\n',
|
|
'exclude': set(['zerver/lib/bugdown/codehilite.py',
|
|
'tools/travis/success-http-headers.txt']),
|
|
'description': 'Fix tab-based whitespace'},
|
|
] # type: RuleList
|
|
markdown_whitespace_rules = list([rule for rule in whitespace_rules if rule['pattern'] != '\s+$']) + [
|
|
# Two spaces trailing a line with other content is okay--it's a markdown line break.
|
|
# This rule finds one space trailing a non-space, three or more trailing spaces, and
|
|
# spaces on an empty line.
|
|
{'pattern': '((?<!\s)\s$)|(\s\s\s+$)|(^\s+$)',
|
|
'strip': '\n',
|
|
'description': 'Fix trailing whitespace'},
|
|
{'pattern': '^#+[A-Za-z0-9]',
|
|
'strip': '\n',
|
|
'description': 'Missing space after # in heading'},
|
|
] # type: RuleList
|
|
js_rules = cast(RuleList, [
|
|
{'pattern': '[^_]function\(',
|
|
'description': 'The keyword "function" should be followed by a space'},
|
|
{'pattern': '.*blueslip.warning\(.*',
|
|
'description': 'The module blueslip has no function warning, try using blueslip.warn'},
|
|
{'pattern': '[)]{$',
|
|
'description': 'Missing space between ) and {'},
|
|
{'pattern': '["\']json/',
|
|
'description': 'Relative URL for JSON route not supported by i18n'},
|
|
# This rule is constructed with + to avoid triggering on itself
|
|
{'pattern': " =" + '[^ =>~"]',
|
|
'description': 'Missing whitespace after "="'},
|
|
{'pattern': '^[ ]*//[A-Za-z0-9]',
|
|
'description': 'Missing space after // in comment'},
|
|
{'pattern': 'if[(]',
|
|
'description': 'Missing space between if and ('},
|
|
{'pattern': 'else{$',
|
|
'description': 'Missing space between else and {'},
|
|
{'pattern': '^else {$',
|
|
'description': 'Write JS else statements on same line as }'},
|
|
{'pattern': '^else if',
|
|
'description': 'Write JS else statements on same line as }'},
|
|
{'pattern': 'console[.][a-z]',
|
|
'exclude': set(['static/js/blueslip.js',
|
|
'frontend_tests/zjsunit',
|
|
'frontend_tests/casper_lib/common.js',
|
|
'frontend_tests/node_tests',
|
|
'static/js/debug.js']),
|
|
'description': 'console.log and similar should not be used in webapp'},
|
|
{'pattern': 'i18n[.]t',
|
|
'include_only': set(['static/js/portico/']),
|
|
'description': 'i18n.t is not available in portico pages yet'},
|
|
{'pattern': '[.]text\(["\'][a-zA-Z]',
|
|
'description': 'Strings passed to $().text should be wrapped in i18n.t() for internationalization'},
|
|
{'pattern': 'compose_error\(["\']',
|
|
'description': 'Argument to compose_error should be a literal string enclosed '
|
|
'by i18n.t()'},
|
|
{'pattern': 'ui.report_success\(',
|
|
'description': 'Deprecated function, use ui_report.success.'},
|
|
{'pattern': 'report.success\(["\']',
|
|
'description': 'Argument to report_success should be a literal string enclosed '
|
|
'by i18n.t()'},
|
|
{'pattern': 'ui.report_error\(',
|
|
'description': 'Deprecated function, use ui_report.error.'},
|
|
{'pattern': 'report.error\(["\']',
|
|
'description': 'Argument to report_error should be a literal string enclosed '
|
|
'by i18n.t()'},
|
|
]) + whitespace_rules
|
|
python_rules = cast(RuleList, [
|
|
{'pattern': '^(?!#)@login_required',
|
|
'description': '@login_required is unsupported; use @zulip_login_required'},
|
|
{'pattern': '".*"%\([a-z_].*\)?$',
|
|
'description': 'Missing space around "%"'},
|
|
{'pattern': "'.*'%\([a-z_].*\)?$",
|
|
'exclude': set(['analytics/lib/counts.py',
|
|
'analytics/tests/test_counts.py',
|
|
]),
|
|
'exclude_line': set([
|
|
('zerver/views/users.py',
|
|
"return json_error(_(\"Email '%(email)s' not allowed for realm '%(realm)s'\") %"),
|
|
('zproject/settings.py',
|
|
"'format': '%(asctime)s %(levelname)-8s %(message)s'"),
|
|
('static/templates/settings/bot-settings.handlebars',
|
|
"'https://hostname.example.com/bots/followup'"),
|
|
]),
|
|
'description': 'Missing space around "%"'},
|
|
# This rule is constructed with + to avoid triggering on itself
|
|
{'pattern': " =" + '[^ =>~"]',
|
|
'description': 'Missing whitespace after "="'},
|
|
{'pattern': '":\w[^"]*$',
|
|
'description': 'Missing whitespace after ":"'},
|
|
{'pattern': "':\w[^']*$",
|
|
'description': 'Missing whitespace after ":"'},
|
|
{'pattern': "^\s+[#]\w",
|
|
'strip': '\n',
|
|
'description': 'Missing whitespace after "#"'},
|
|
{'pattern': "assertEquals[(]",
|
|
'description': 'Use assertEqual, not assertEquals (which is deprecated).'},
|
|
{'pattern': "== None",
|
|
'description': 'Use `is None` to check whether something is None'},
|
|
{'pattern': "type:[(]",
|
|
'description': 'Missing whitespace after ":" in type annotation'},
|
|
{'pattern': "# type [(]",
|
|
'description': 'Missing : after type in type annotation'},
|
|
{'pattern': "#type",
|
|
'description': 'Missing whitespace after "#" in type annotation'},
|
|
{'pattern': 'if[(]',
|
|
'description': 'Missing space between if and ('},
|
|
{'pattern': ", [)]",
|
|
'description': 'Unnecessary whitespace between "," and ")"'},
|
|
{'pattern': "% [(]",
|
|
'description': 'Unnecessary whitespace between "%" and "("'},
|
|
# This next check could have false positives, but it seems pretty
|
|
# rare; if we find any, they can be added to the exclude list for
|
|
# this rule.
|
|
{'pattern': ' % [a-zA-Z0-9_.]*\)?$',
|
|
'exclude_line': set([
|
|
('tools/tests/test_template_parser.py', '{% foo'),
|
|
]),
|
|
'description': 'Used % comprehension without a tuple'},
|
|
{'pattern': '.*%s.* % \([a-zA-Z0-9_.]*\)$',
|
|
'description': 'Used % comprehension without a tuple'},
|
|
{'pattern': 'django.utils.translation',
|
|
'include_only': set(['test/']),
|
|
'description': 'Test strings should not be tagged for translationx'},
|
|
{'pattern': 'json_success\({}\)',
|
|
'description': 'Use json_success() to return nothing'},
|
|
# To avoid json_error(_variable) and json_error(_(variable))
|
|
{'pattern': '\Wjson_error\(_\(?\w+\)',
|
|
'exclude': set(['zerver/tests']),
|
|
'description': 'Argument to json_error should be a literal string enclosed by _()'},
|
|
{'pattern': '\Wjson_error\([\'"].+[),]$',
|
|
'exclude': set(['zerver/tests']),
|
|
'exclude_line': set([
|
|
# We don't want this string tagged for translation.
|
|
('zerver/views/compatibility.py', 'return json_error("Client is too old")'),
|
|
]),
|
|
'description': 'Argument to json_error should a literal string enclosed by _()'},
|
|
# To avoid JsonableError(_variable) and JsonableError(_(variable))
|
|
{'pattern': '\WJsonableError\(_\(?\w.+\)',
|
|
'exclude': set(['zerver/tests']),
|
|
'description': 'Argument to JsonableError should be a literal string enclosed by _()'},
|
|
{'pattern': '\WJsonableError\(["\'].+\)',
|
|
'exclude': set(['zerver/tests']),
|
|
'description': 'Argument to JsonableError should be a literal string enclosed by _()'},
|
|
{'pattern': '([a-zA-Z0-9_]+)=REQ\([\'"]\\1[\'"]',
|
|
'description': 'REQ\'s first argument already defaults to parameter name'},
|
|
{'pattern': 'self\.client\.(get|post|patch|put|delete)',
|
|
'exclude': set(['zilencer/tests.py']),
|
|
'description': \
|
|
'''Do not call self.client directly for put/patch/post/get.
|
|
See WRAPPER_COMMENT in test_helpers.py for details.
|
|
'''},
|
|
# Directly fetching Message objects in e.g. views code is often a security bug.
|
|
{'pattern': '[^r][M]essage.objects.get',
|
|
'exclude': set(["zerver/tests", "zerver/worker/queue_processors.py"]),
|
|
'description': 'Please use access_message() to fetch Message objects',
|
|
},
|
|
{'pattern': '[S]tream.objects.get',
|
|
'include_only': set(["zerver/views/"]),
|
|
'description': 'Please use access_stream_by_*() to fetch Stream objects',
|
|
},
|
|
{'pattern': 'get_stream[(]',
|
|
'include_only': set(["zerver/views/", "zerver/lib/actions.py"]),
|
|
# messages.py needs to support accessing invite-only streams
|
|
# that you are no longer subscribed to, so need get_stream.
|
|
'exclude': set(['zerver/views/messages.py']),
|
|
'exclude_line': set([
|
|
# This is a check for whether a stream rename is invalid because it already exists
|
|
('zerver/lib/actions.py', 'get_stream(new_name, stream.realm)'),
|
|
# This one in check_message is kinda terrible, since it's
|
|
# how most instances are written, but better to exclude something than nothing
|
|
('zerver/lib/actions.py', 'stream = get_stream(stream_name, realm)'),
|
|
('zerver/lib/actions.py', 'get_stream(signups_stream, admin_realm)'),
|
|
]),
|
|
'description': 'Please use access_stream_by_*() to fetch Stream objects',
|
|
},
|
|
{'pattern': '[S]tream.objects.filter',
|
|
'include_only': set(["zerver/views/"]),
|
|
'description': 'Please use access_stream_by_*() to fetch Stream objects',
|
|
},
|
|
{'pattern': '^from (zerver|analytics|confirmation)',
|
|
'include_only': set(["/migrations/"]),
|
|
'exclude': set(['zerver/migrations/0032_verify_all_medium_avatar_images.py',
|
|
'zerver/migrations/0041_create_attachments_for_old_messages.py',
|
|
'zerver/migrations/0060_move_avatars_to_be_uid_based.py']),
|
|
'description': "Don't import models or other code in migrations; see docs/schema-migrations.md",
|
|
},
|
|
{'pattern': 'datetime[.](now|utcnow)',
|
|
'include_only': set(["zerver/", "analytics/"]),
|
|
'description': "Don't use datetime in backend code.\n"
|
|
"See https://zulip.readthedocs.io/en/latest/code-style.html#naive-datetime-objects",
|
|
},
|
|
{'pattern': 'render_to_response\(',
|
|
'description': "Use render() instead of render_to_response().",
|
|
},
|
|
# This rule might give false positives in virtualenv setup files which should be excluded,
|
|
# and comments which should be rewritten to avoid use of "python2", "python3", etc.
|
|
{'pattern': 'python[23]',
|
|
'exclude': set(['tools/lib/provision.py',
|
|
'tools/setup/setup_venvs.py',
|
|
'scripts/lib/setup_venv.py']),
|
|
'description': 'Explicit python invocations should not include a version'},
|
|
{'pattern': '(^|\s)open\s*\(',
|
|
'description': 'open() should not be used in Zulip\'s bots. Use functions'
|
|
' provided by the bots framework to access the filesystem.',
|
|
'include_only': set(['api/bots/']),
|
|
'exclude': set(['api/bots/john/john.py'])},
|
|
]) + whitespace_rules
|
|
bash_rules = [
|
|
{'pattern': '#!.*sh [-xe]',
|
|
'description': 'Fix shebang line with proper call to /usr/bin/env for Bash path, change -x|-e switches'
|
|
' to set -x|set -e'},
|
|
] + whitespace_rules[0:1] # type: RuleList
|
|
css_rules = cast(RuleList, [
|
|
{'pattern': '^[^:]*:\S[^:]*;$',
|
|
'description': "Missing whitespace after : in CSS"},
|
|
{'pattern': '[a-z]{',
|
|
'description': "Missing whitespace before '{' in CSS."},
|
|
{'pattern': 'https://',
|
|
'description': "Zulip CSS should have no dependencies on external resources"},
|
|
{'pattern': '^[ ][ ][a-zA-Z0-9]',
|
|
'description': "Incorrect 2-space indentation in CSS",
|
|
'exclude': set(['static/third/thirdparty-fonts.css']),
|
|
'strip': '\n'},
|
|
{'pattern': '{\w',
|
|
'description': "Missing whitespace after '{' in CSS (should be newline)."},
|
|
{'pattern': ' thin[; ]',
|
|
'description': "thin CSS attribute is under-specified, please use 1px."},
|
|
{'pattern': ' medium[; ]',
|
|
'description': "medium CSS attribute is under-specified, please use pixels."},
|
|
{'pattern': ' thick[; ]',
|
|
'description': "thick CSS attribute is under-specified, please use pixels."},
|
|
]) + whitespace_rules # type: RuleList
|
|
prose_style_rules = [
|
|
{'pattern': '[^\/\#\-\"]([jJ]avascript)', # exclude usage in hrefs/divs
|
|
'description': "javascript should be spelled JavaScript"},
|
|
{'pattern': '[^\/\-\.\"\'\_\=\>]([gG]ithub)[^\.\-\_\"\<]', # exclude usage in hrefs/divs
|
|
'description': "github should be spelled GitHub"},
|
|
{'pattern': '[oO]rganisation', # exclude usage in hrefs/divs
|
|
'description': "Organization is spelled with a z"},
|
|
{'pattern': '!!! warning',
|
|
'description': "!!! warning is invalid; it's spelled '!!! warn'"},
|
|
] # type: RuleList
|
|
html_rules = whitespace_rules + prose_style_rules + [
|
|
{'pattern': 'placeholder="[^{]',
|
|
'description': "`placeholder` value should be translatable.",
|
|
'exclude_line': [('templates/zerver/register.html', 'placeholder="acme"'),
|
|
('templates/zerver/register.html', 'placeholder="Acme"'),
|
|
('static/templates/settings/realm-domains-modal.handlebars',
|
|
'<td><input type="text" class="new-realm-domain" placeholder="acme.com"></input></td>')],
|
|
'exclude': set(["static/templates/settings/emoji-settings-admin.handlebars",
|
|
"static/templates/settings/realm-filter-settings-admin.handlebars",
|
|
"static/templates/settings/bot-settings.handlebars"])},
|
|
{'pattern': "placeholder='[^{]",
|
|
'description': "`placeholder` value should be translatable."},
|
|
{'pattern': "aria-label='[^{]",
|
|
'description': "`aria-label` value should be translatable."},
|
|
{'pattern': 'aria-label="[^{]',
|
|
'description': "`aria-label` value should be translatable."},
|
|
{'pattern': 'script src="http',
|
|
'description': "Don't directly load dependencies from CDNs. See docs/front-end-build-process.md"},
|
|
{'pattern': "title='[^{]",
|
|
'description': "`title` value should be translatable."},
|
|
{'pattern': 'title="[^{\:]',
|
|
'exclude_line': set([
|
|
('templates/zerver/markdown_help.html',
|
|
'<td><img alt=":heart:" class="emoji" src="/static/generated/emoji/images/emoji/heart.png" title=":heart:" /></td>')
|
|
]),
|
|
'exclude': set(["templates/zerver/emails"]),
|
|
'description': "`title` value should be translatable."},
|
|
{'pattern': '\Walt=["\'][^{"\']',
|
|
'description': "alt argument should be enclosed by _() or it should be an empty string.",
|
|
'exclude': set(['static/templates/settings/display-settings.handlebars',
|
|
'templates/zerver/keyboard_shortcuts.html',
|
|
'templates/zerver/markdown_help.html']),
|
|
},
|
|
{'pattern': '\Walt=["\']{{ ?["\']',
|
|
'description': "alt argument should be enclosed by _().",
|
|
},
|
|
] # type: RuleList
|
|
handlebars_rules = html_rules + [
|
|
{'pattern': "[<]script",
|
|
'description': "Do not use inline <script> tags here; put JavaScript in static/js instead."},
|
|
{'pattern': "{{t '.*' }}[\.\?!]",
|
|
'description': "Period should be part of the translatable string."},
|
|
{'pattern': '{{t ".*" }}[\.\?!]',
|
|
'description': "Period should be part of the translatable string."},
|
|
{'pattern': "{{/tr}}[\.\?!]",
|
|
'description': "Period should be part of the translatable string."},
|
|
]
|
|
jinja2_rules = html_rules + [
|
|
{'pattern': "{% endtrans %}[\.\?!]",
|
|
'description': "Period should be part of the translatable string."},
|
|
{'pattern': "{{ _(.+) }}[\.\?!]",
|
|
'description': "Period should be part of the translatable string."},
|
|
]
|
|
json_rules = [] # type: RuleList # fix newlines at ends of files
|
|
# It is okay that json_rules is empty, because the empty list
|
|
# ensures we'll still check JSON files for whitespace.
|
|
markdown_rules = markdown_whitespace_rules + prose_style_rules + [
|
|
{'pattern': '\[(?P<url>[^\]]+)\]\((?P=url)\)',
|
|
'description': 'Linkified markdown URLs should use cleaner <http://example.com> syntax.'}
|
|
]
|
|
help_markdown_rules = markdown_rules + [
|
|
{'pattern': '[a-z][.][A-Z]',
|
|
'description': "Likely missing space after end of sentence"},
|
|
{'pattern': '[rR]ealm',
|
|
'description': "Realms are referred to as Organizations in user-facing docs."},
|
|
]
|
|
txt_rules = whitespace_rules
|
|
|
|
def check_custom_checks_py():
|
|
# type: () -> bool
|
|
failed = False
|
|
|
|
for fn in by_lang['py']:
|
|
if 'custom_check.py' in fn:
|
|
continue
|
|
if custom_check_file(fn, 'py', python_rules, max_length=140):
|
|
failed = True
|
|
return failed
|
|
|
|
def check_custom_checks_nonpy():
|
|
# type: () -> bool
|
|
failed = False
|
|
|
|
for fn in by_lang['js']:
|
|
if custom_check_file(fn, 'js', js_rules):
|
|
failed = True
|
|
|
|
for fn in by_lang['sh']:
|
|
if custom_check_file(fn, 'sh', bash_rules):
|
|
failed = True
|
|
|
|
for fn in by_lang['css']:
|
|
if custom_check_file(fn, 'css', css_rules):
|
|
failed = True
|
|
|
|
for fn in by_lang['handlebars']:
|
|
if custom_check_file(fn, 'handlebars', handlebars_rules):
|
|
failed = True
|
|
|
|
for fn in by_lang['html']:
|
|
if custom_check_file(fn, 'html', jinja2_rules):
|
|
failed = True
|
|
|
|
for fn in by_lang['json']:
|
|
if custom_check_file(fn, 'json', json_rules):
|
|
failed = True
|
|
|
|
markdown_docs_length_exclude = {
|
|
"api/bots/converter/readme.md",
|
|
"docs/bots-guide.md",
|
|
"docs/dev-env-first-time-contributors.md",
|
|
"docs/webhook-walkthrough.md",
|
|
"docs/life-of-a-request.md",
|
|
"docs/logging.md",
|
|
"docs/migration-renumbering.md",
|
|
"docs/readme-symlink.md",
|
|
"README.md",
|
|
"zerver/webhooks/helloworld/doc.md",
|
|
"zerver/webhooks/trello/doc.md",
|
|
"templates/zerver/integrations/perforce.md",
|
|
}
|
|
for fn in by_lang['md']:
|
|
max_length = None
|
|
if fn not in markdown_docs_length_exclude:
|
|
max_length = 120
|
|
rules = markdown_rules
|
|
if fn.startswith("templates/zerver/help"):
|
|
rules = help_markdown_rules
|
|
if custom_check_file(fn, 'md', rules, max_length=max_length):
|
|
failed = True
|
|
|
|
for fn in by_lang['txt'] + by_lang['text']:
|
|
if custom_check_file(fn, 'txt', txt_rules):
|
|
failed = True
|
|
|
|
for fn in by_lang['yaml']:
|
|
if custom_check_file(fn, 'yaml', txt_rules):
|
|
failed = True
|
|
|
|
return failed
|
|
|
|
return (check_custom_checks_py, check_custom_checks_nonpy)
|