mirror of https://github.com/ansible/ansible.git
You cannot select more than 25 topics
Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
781 lines
22 KiB
Python
781 lines
22 KiB
Python
"""Execute Ansible sanity tests."""
|
|
|
|
from __future__ import absolute_import, print_function
|
|
|
|
import glob
|
|
import json
|
|
import os
|
|
import re
|
|
|
|
from xml.etree.ElementTree import (
|
|
fromstring,
|
|
Element,
|
|
)
|
|
|
|
from lib.util import (
|
|
ApplicationError,
|
|
SubprocessError,
|
|
display,
|
|
run_command,
|
|
deepest_path,
|
|
parse_to_dict,
|
|
)
|
|
|
|
from lib.ansible_util import (
|
|
ansible_environment,
|
|
)
|
|
|
|
from lib.target import (
|
|
walk_external_targets,
|
|
walk_internal_targets,
|
|
walk_sanity_targets,
|
|
)
|
|
|
|
from lib.executor import (
|
|
get_changes_filter,
|
|
AllTargetsSkipped,
|
|
Delegate,
|
|
install_command_requirements,
|
|
SUPPORTED_PYTHON_VERSIONS,
|
|
intercept_command,
|
|
)
|
|
|
|
from lib.config import (
|
|
SanityConfig,
|
|
)
|
|
|
|
from lib.test import (
|
|
TestSuccess,
|
|
TestFailure,
|
|
TestSkipped,
|
|
TestMessage,
|
|
calculate_best_confidence,
|
|
)
|
|
|
|
COMMAND = 'sanity'
|
|
|
|
PEP8_SKIP_PATH = 'test/sanity/pep8/skip.txt'
|
|
PEP8_LEGACY_PATH = 'test/sanity/pep8/legacy-files.txt'
|
|
|
|
PYLINT_SKIP_PATH = 'test/sanity/pylint/skip.txt'
|
|
|
|
|
|
def command_sanity(args):
|
|
"""
|
|
:type args: SanityConfig
|
|
"""
|
|
changes = get_changes_filter(args)
|
|
require = (args.require or []) + changes
|
|
targets = SanityTargets(args.include, args.exclude, require)
|
|
|
|
if not targets.include:
|
|
raise AllTargetsSkipped()
|
|
|
|
if args.delegate:
|
|
raise Delegate(require=changes)
|
|
|
|
install_command_requirements(args)
|
|
|
|
tests = sanity_get_tests()
|
|
|
|
if args.test:
|
|
tests = [t for t in tests if t.name in args.test]
|
|
|
|
if args.skip_test:
|
|
tests = [t for t in tests if t.name not in args.skip_test]
|
|
|
|
total = 0
|
|
failed = []
|
|
|
|
for test in tests:
|
|
if args.list_tests:
|
|
display.info(test.name)
|
|
continue
|
|
|
|
if test.intercept:
|
|
versions = SUPPORTED_PYTHON_VERSIONS
|
|
else:
|
|
versions = None,
|
|
|
|
for version in versions:
|
|
if args.python and version and version != args.python:
|
|
continue
|
|
|
|
display.info('Sanity check using %s%s' % (test.name, ' with Python %s' % version if version else ''))
|
|
|
|
options = ''
|
|
|
|
if test.script:
|
|
result = test.func(args, targets, test.script)
|
|
elif test.intercept:
|
|
result = test.func(args, targets, python_version=version)
|
|
options = ' --python %s' % version
|
|
else:
|
|
result = test.func(args, targets)
|
|
|
|
result.write(args)
|
|
|
|
total += 1
|
|
|
|
if isinstance(result, SanityFailure):
|
|
failed.append(result.test + options)
|
|
|
|
if failed:
|
|
message = 'The %d sanity test(s) listed below (out of %d) failed. See error output above for details.\n%s' % (
|
|
len(failed), total, '\n'.join(failed))
|
|
|
|
if args.failure_ok:
|
|
display.error(message)
|
|
else:
|
|
raise ApplicationError(message)
|
|
|
|
|
|
def command_sanity_code_smell(args, _, script):
|
|
"""
|
|
:type args: SanityConfig
|
|
:type _: SanityTargets
|
|
:type script: str
|
|
:rtype: SanityResult
|
|
"""
|
|
test = os.path.splitext(os.path.basename(script))[0]
|
|
|
|
cmd = [script]
|
|
env = ansible_environment(args, color=False)
|
|
|
|
try:
|
|
stdout, stderr = run_command(args, cmd, env=env, capture=True)
|
|
status = 0
|
|
except SubprocessError as ex:
|
|
stdout = ex.stdout
|
|
stderr = ex.stderr
|
|
status = ex.status
|
|
|
|
if stderr or status:
|
|
summary = str(SubprocessError(cmd=cmd, status=status, stderr=stderr, stdout=stdout))
|
|
return SanityFailure(test, summary=summary)
|
|
|
|
return SanitySuccess(test)
|
|
|
|
|
|
def command_sanity_validate_modules(args, targets):
|
|
"""
|
|
:type args: SanityConfig
|
|
:type targets: SanityTargets
|
|
:rtype: SanityResult
|
|
"""
|
|
test = 'validate-modules'
|
|
env = ansible_environment(args, color=False)
|
|
|
|
paths = [deepest_path(i.path, 'lib/ansible/modules/') for i in targets.include_external]
|
|
paths = sorted(set(p for p in paths if p))
|
|
|
|
if not paths:
|
|
return SanitySkipped(test)
|
|
|
|
cmd = [
|
|
'test/sanity/validate-modules/validate-modules',
|
|
'--format', 'json',
|
|
] + paths
|
|
|
|
with open('test/sanity/validate-modules/skip.txt', 'r') as skip_fd:
|
|
skip_paths = skip_fd.read().splitlines()
|
|
|
|
skip_paths += [e.path for e in targets.exclude_external]
|
|
|
|
if skip_paths:
|
|
cmd += ['--exclude', '^(%s)' % '|'.join(skip_paths)]
|
|
|
|
if args.base_branch:
|
|
cmd.extend([
|
|
'--base-branch', args.base_branch,
|
|
])
|
|
else:
|
|
display.warning('Cannot perform module comparison against the base branch. Base branch not detected when running locally.')
|
|
|
|
try:
|
|
stdout, stderr = run_command(args, cmd, env=env, capture=True)
|
|
status = 0
|
|
except SubprocessError as ex:
|
|
stdout = ex.stdout
|
|
stderr = ex.stderr
|
|
status = ex.status
|
|
|
|
if stderr or status not in (0, 3):
|
|
raise SubprocessError(cmd=cmd, status=status, stderr=stderr, stdout=stdout)
|
|
|
|
if args.explain:
|
|
return SanitySkipped(test)
|
|
|
|
messages = json.loads(stdout)
|
|
|
|
results = []
|
|
|
|
for filename in messages:
|
|
output = messages[filename]
|
|
|
|
for item in output['errors']:
|
|
results.append(SanityMessage(
|
|
path=filename,
|
|
line=int(item['line']) if 'line' in item else 0,
|
|
column=int(item['column']) if 'column' in item else 0,
|
|
level='error',
|
|
code='E%s' % item['code'],
|
|
message=item['msg'],
|
|
))
|
|
|
|
if results:
|
|
return SanityFailure(test, messages=results)
|
|
|
|
return SanitySuccess(test)
|
|
|
|
|
|
def command_sanity_shellcheck(args, targets):
|
|
"""
|
|
:type args: SanityConfig
|
|
:type targets: SanityTargets
|
|
:rtype: SanityResult
|
|
"""
|
|
test = 'shellcheck'
|
|
|
|
with open('test/sanity/shellcheck/skip.txt', 'r') as skip_fd:
|
|
skip_paths = set(skip_fd.read().splitlines())
|
|
|
|
with open('test/sanity/shellcheck/exclude.txt', 'r') as exclude_fd:
|
|
exclude = set(exclude_fd.read().splitlines())
|
|
|
|
paths = sorted(i.path for i in targets.include if os.path.splitext(i.path)[1] == '.sh' and i.path not in skip_paths)
|
|
|
|
if not paths:
|
|
return SanitySkipped(test)
|
|
|
|
cmd = [
|
|
'shellcheck',
|
|
'-e', ','.join(sorted(exclude)),
|
|
'--format', 'checkstyle',
|
|
] + paths
|
|
|
|
try:
|
|
stdout, stderr = run_command(args, cmd, capture=True)
|
|
status = 0
|
|
except SubprocessError as ex:
|
|
stdout = ex.stdout
|
|
stderr = ex.stderr
|
|
status = ex.status
|
|
|
|
if stderr or status > 1:
|
|
raise SubprocessError(cmd=cmd, status=status, stderr=stderr, stdout=stdout)
|
|
|
|
if args.explain:
|
|
return SanitySkipped(test)
|
|
|
|
# json output is missing file paths in older versions of shellcheck, so we'll use xml instead
|
|
root = fromstring(stdout) # type: Element
|
|
|
|
results = []
|
|
|
|
for item in root: # type: Element
|
|
for entry in item: # type: Element
|
|
results.append(SanityMessage(
|
|
message=entry.attrib['message'],
|
|
path=item.attrib['name'],
|
|
line=int(entry.attrib['line']),
|
|
column=int(entry.attrib['column']),
|
|
level=entry.attrib['severity'],
|
|
code=entry.attrib['source'].replace('ShellCheck.', ''),
|
|
))
|
|
|
|
if results:
|
|
return SanityFailure(test, messages=results)
|
|
|
|
return SanitySuccess(test)
|
|
|
|
|
|
def command_sanity_pep8(args, targets):
|
|
"""
|
|
:type args: SanityConfig
|
|
:type targets: SanityTargets
|
|
:rtype: SanityResult
|
|
"""
|
|
test = 'pep8'
|
|
|
|
with open(PEP8_SKIP_PATH, 'r') as skip_fd:
|
|
skip_paths = skip_fd.read().splitlines()
|
|
|
|
with open(PEP8_LEGACY_PATH, 'r') as legacy_fd:
|
|
legacy_paths = legacy_fd.read().splitlines()
|
|
|
|
with open('test/sanity/pep8/legacy-ignore.txt', 'r') as ignore_fd:
|
|
legacy_ignore = set(ignore_fd.read().splitlines())
|
|
|
|
with open('test/sanity/pep8/current-ignore.txt', 'r') as ignore_fd:
|
|
current_ignore = sorted(ignore_fd.read().splitlines())
|
|
|
|
skip_paths_set = set(skip_paths)
|
|
legacy_paths_set = set(legacy_paths)
|
|
|
|
paths = sorted(i.path for i in targets.include if os.path.splitext(i.path)[1] == '.py' and i.path not in skip_paths_set)
|
|
|
|
if not paths:
|
|
return SanitySkipped(test)
|
|
|
|
cmd = [
|
|
'pep8',
|
|
'--max-line-length', '160',
|
|
'--config', '/dev/null',
|
|
'--ignore', ','.join(sorted(current_ignore)),
|
|
] + paths
|
|
|
|
try:
|
|
stdout, stderr = run_command(args, cmd, capture=True)
|
|
status = 0
|
|
except SubprocessError as ex:
|
|
stdout = ex.stdout
|
|
stderr = ex.stderr
|
|
status = ex.status
|
|
|
|
if stderr:
|
|
raise SubprocessError(cmd=cmd, status=status, stderr=stderr, stdout=stdout)
|
|
|
|
if args.explain:
|
|
return SanitySkipped(test)
|
|
|
|
pattern = '^(?P<path>[^:]*):(?P<line>[0-9]+):(?P<column>[0-9]+): (?P<code>[WE][0-9]{3}) (?P<message>.*)$'
|
|
|
|
results = [re.search(pattern, line).groupdict() for line in stdout.splitlines()]
|
|
|
|
results = [SanityMessage(
|
|
message=r['message'],
|
|
path=r['path'],
|
|
line=int(r['line']),
|
|
column=int(r['column']),
|
|
level='warning' if r['code'].startswith('W') else 'error',
|
|
code=r['code'],
|
|
) for r in results]
|
|
|
|
failed_result_paths = set([result.path for result in results])
|
|
used_paths = set(paths)
|
|
|
|
errors = []
|
|
summary = {}
|
|
|
|
line = 0
|
|
|
|
for path in legacy_paths:
|
|
line += 1
|
|
|
|
if not os.path.exists(path):
|
|
# Keep files out of the list which no longer exist in the repo.
|
|
errors.append(SanityMessage(
|
|
code='A101',
|
|
message='Remove "%s" since it does not exist' % path,
|
|
path=PEP8_LEGACY_PATH,
|
|
line=line,
|
|
column=1,
|
|
confidence=calculate_best_confidence(((PEP8_LEGACY_PATH, line), (path, 0)), args.metadata) if args.metadata.changes else None,
|
|
))
|
|
|
|
if path in used_paths and path not in failed_result_paths:
|
|
# Keep files out of the list which no longer require the relaxed rule set.
|
|
errors.append(SanityMessage(
|
|
code='A201',
|
|
message='Remove "%s" since it passes the current rule set' % path,
|
|
path=PEP8_LEGACY_PATH,
|
|
line=line,
|
|
column=1,
|
|
confidence=calculate_best_confidence(((PEP8_LEGACY_PATH, line), (path, 0)), args.metadata) if args.metadata.changes else None,
|
|
))
|
|
|
|
line = 0
|
|
|
|
for path in skip_paths:
|
|
line += 1
|
|
|
|
if not os.path.exists(path):
|
|
# Keep files out of the list which no longer exist in the repo.
|
|
errors.append(SanityMessage(
|
|
code='A101',
|
|
message='Remove "%s" since it does not exist' % path,
|
|
path=PEP8_SKIP_PATH,
|
|
line=line,
|
|
column=1,
|
|
confidence=calculate_best_confidence(((PEP8_SKIP_PATH, line), (path, 0)), args.metadata) if args.metadata.changes else None,
|
|
))
|
|
|
|
for result in results:
|
|
if result.path in legacy_paths_set and result.code in legacy_ignore:
|
|
# Files on the legacy list are permitted to have errors on the legacy ignore list.
|
|
# However, we want to report on their existence to track progress towards eliminating these exceptions.
|
|
display.info('PEP 8: %s (legacy)' % result, verbosity=3)
|
|
|
|
key = '%s %s' % (result.code, re.sub('[0-9]+', 'NNN', result.message))
|
|
|
|
if key not in summary:
|
|
summary[key] = 0
|
|
|
|
summary[key] += 1
|
|
else:
|
|
# Files not on the legacy list and errors not on the legacy ignore list are PEP 8 policy errors.
|
|
errors.append(result)
|
|
|
|
if summary:
|
|
lines = []
|
|
count = 0
|
|
|
|
for key in sorted(summary):
|
|
count += summary[key]
|
|
lines.append('PEP 8: %5d %s' % (summary[key], key))
|
|
|
|
display.info('PEP 8: There were %d different legacy issues found (%d total):' % (len(summary), count), verbosity=1)
|
|
display.info('PEP 8: Count Code Message', verbosity=1)
|
|
|
|
for line in lines:
|
|
display.info(line, verbosity=1)
|
|
|
|
if errors:
|
|
return SanityFailure(test, messages=errors)
|
|
|
|
return SanitySuccess(test)
|
|
|
|
|
|
def command_sanity_pylint(args, targets):
|
|
"""
|
|
:type args: SanityConfig
|
|
:type targets: SanityTargets
|
|
:rtype: SanityResult
|
|
"""
|
|
test = 'pylint'
|
|
|
|
with open(PYLINT_SKIP_PATH, 'r') as skip_fd:
|
|
skip_paths = skip_fd.read().splitlines()
|
|
|
|
with open('test/sanity/pylint/disable.txt', 'r') as disable_fd:
|
|
disable = set(disable_fd.read().splitlines())
|
|
|
|
skip_paths_set = set(skip_paths)
|
|
|
|
paths = sorted(i.path for i in targets.include if os.path.splitext(i.path)[1] == '.py' and i.path not in skip_paths_set)
|
|
|
|
if not paths:
|
|
return SanitySkipped(test)
|
|
|
|
cmd = [
|
|
'pylint',
|
|
'--jobs', '0',
|
|
'--reports', 'n',
|
|
'--max-line-length', '160',
|
|
'--rcfile', '/dev/null',
|
|
'--output-format', 'json',
|
|
'--disable', ','.join(sorted(disable)),
|
|
] + paths
|
|
|
|
env = ansible_environment(args)
|
|
|
|
try:
|
|
stdout, stderr = run_command(args, cmd, env=env, capture=True)
|
|
status = 0
|
|
except SubprocessError as ex:
|
|
stdout = ex.stdout
|
|
stderr = ex.stderr
|
|
status = ex.status
|
|
|
|
if stderr or status >= 32:
|
|
raise SubprocessError(cmd=cmd, status=status, stderr=stderr, stdout=stdout)
|
|
|
|
if args.explain:
|
|
return SanitySkipped(test)
|
|
|
|
if stdout:
|
|
messages = json.loads(stdout)
|
|
else:
|
|
messages = []
|
|
|
|
errors = [SanityMessage(
|
|
message=m['message'],
|
|
path=m['path'],
|
|
line=int(m['line']),
|
|
column=int(m['column']),
|
|
level=m['type'],
|
|
code=m['symbol'],
|
|
) for m in messages]
|
|
|
|
line = 0
|
|
|
|
for path in skip_paths:
|
|
line += 1
|
|
|
|
if not os.path.exists(path):
|
|
# Keep files out of the list which no longer exist in the repo.
|
|
errors.append(SanityMessage(
|
|
code='A101',
|
|
message='Remove "%s" since it does not exist' % path,
|
|
path=PYLINT_SKIP_PATH,
|
|
line=line,
|
|
column=1,
|
|
confidence=calculate_best_confidence(((PYLINT_SKIP_PATH, line), (path, 0)), args.metadata) if args.metadata.changes else None,
|
|
))
|
|
|
|
if errors:
|
|
return SanityFailure(test, messages=errors)
|
|
|
|
return SanitySuccess(test)
|
|
|
|
|
|
def command_sanity_yamllint(args, targets):
|
|
"""
|
|
:type args: SanityConfig
|
|
:type targets: SanityTargets
|
|
:rtype: SanityResult
|
|
"""
|
|
test = 'yamllint'
|
|
|
|
paths = sorted(i.path for i in targets.include if os.path.splitext(i.path)[1] in ('.yml', '.yaml'))
|
|
|
|
if not paths:
|
|
return SanitySkipped(test)
|
|
|
|
cmd = [
|
|
'yamllint',
|
|
'--format', 'parsable',
|
|
] + paths
|
|
|
|
try:
|
|
stdout, stderr = run_command(args, cmd, capture=True)
|
|
status = 0
|
|
except SubprocessError as ex:
|
|
stdout = ex.stdout
|
|
stderr = ex.stderr
|
|
status = ex.status
|
|
|
|
if stderr:
|
|
raise SubprocessError(cmd=cmd, status=status, stderr=stderr, stdout=stdout)
|
|
|
|
if args.explain:
|
|
return SanitySkipped(test)
|
|
|
|
pattern = r'^(?P<path>[^:]*):(?P<line>[0-9]+):(?P<column>[0-9]+): \[(?P<level>warning|error)\] (?P<message>.*)$'
|
|
|
|
results = [re.search(pattern, line).groupdict() for line in stdout.splitlines()]
|
|
|
|
results = [SanityMessage(
|
|
message=r['message'],
|
|
path=r['path'],
|
|
line=int(r['line']),
|
|
column=int(r['column']),
|
|
level=r['level'],
|
|
) for r in results]
|
|
|
|
if results:
|
|
return SanityFailure(test, messages=results)
|
|
|
|
return SanitySuccess(test)
|
|
|
|
|
|
def command_sanity_rstcheck(args, targets):
|
|
"""
|
|
:type args: SanityConfig
|
|
:type targets: SanityTargets
|
|
:rtype: SanityResult
|
|
"""
|
|
test = 'rstcheck'
|
|
|
|
with open('test/sanity/rstcheck/ignore-substitutions.txt', 'r') as ignore_fd:
|
|
ignore_substitutions = sorted(set(ignore_fd.read().splitlines()))
|
|
|
|
paths = sorted(i.path for i in targets.include if os.path.splitext(i.path)[1] in ('.rst',))
|
|
|
|
if not paths:
|
|
return SanitySkipped(test)
|
|
|
|
cmd = [
|
|
'rstcheck',
|
|
'--report', 'warning',
|
|
'--ignore-substitutions', ','.join(ignore_substitutions),
|
|
] + paths
|
|
|
|
try:
|
|
stdout, stderr = run_command(args, cmd, capture=True)
|
|
status = 0
|
|
except SubprocessError as ex:
|
|
stdout = ex.stdout
|
|
stderr = ex.stderr
|
|
status = ex.status
|
|
|
|
if stdout:
|
|
raise SubprocessError(cmd=cmd, status=status, stderr=stderr, stdout=stdout)
|
|
|
|
if args.explain:
|
|
return SanitySkipped(test)
|
|
|
|
pattern = r'^(?P<path>[^:]*):(?P<line>[0-9]+): \((?P<level>INFO|WARNING|ERROR|SEVERE)/[0-4]\) (?P<message>.*)$'
|
|
|
|
results = [parse_to_dict(pattern, line) for line in stderr.splitlines()]
|
|
|
|
results = [SanityMessage(
|
|
message=r['message'],
|
|
path=r['path'],
|
|
line=int(r['line']),
|
|
column=0,
|
|
level=r['level'],
|
|
) for r in results]
|
|
|
|
if results:
|
|
return SanityFailure(test, messages=results)
|
|
|
|
return SanitySuccess(test)
|
|
|
|
|
|
def command_sanity_ansible_doc(args, targets, python_version):
|
|
"""
|
|
:type args: SanityConfig
|
|
:type targets: SanityTargets
|
|
:type python_version: str
|
|
:rtype: SanityResult
|
|
"""
|
|
test = 'ansible-doc'
|
|
|
|
with open('test/sanity/ansible-doc/skip.txt', 'r') as skip_fd:
|
|
skip_modules = set(skip_fd.read().splitlines())
|
|
|
|
modules = sorted(set(m for i in targets.include_external for m in i.modules) -
|
|
set(m for i in targets.exclude_external for m in i.modules) -
|
|
skip_modules)
|
|
|
|
if not modules:
|
|
return SanitySkipped(test, python_version=python_version)
|
|
|
|
env = ansible_environment(args, color=False)
|
|
cmd = ['ansible-doc'] + modules
|
|
|
|
try:
|
|
stdout, stderr = intercept_command(args, cmd, target_name='ansible-doc', env=env, capture=True, python_version=python_version)
|
|
status = 0
|
|
except SubprocessError as ex:
|
|
stdout = ex.stdout
|
|
stderr = ex.stderr
|
|
status = ex.status
|
|
|
|
if status:
|
|
summary = str(SubprocessError(cmd=cmd, status=status, stderr=stderr))
|
|
return SanityFailure(test, summary=summary, python_version=python_version)
|
|
|
|
if stdout:
|
|
display.info(stdout.strip(), verbosity=3)
|
|
|
|
if stderr:
|
|
summary = 'Output on stderr from ansible-doc is considered an error.\n\n%s' % SubprocessError(cmd, stderr=stderr)
|
|
return SanityFailure(test, summary=summary, python_version=python_version)
|
|
|
|
return SanitySuccess(test, python_version=python_version)
|
|
|
|
|
|
def collect_code_smell_tests():
|
|
"""
|
|
:rtype: tuple(SanityFunc)
|
|
"""
|
|
with open('test/sanity/code-smell/skip.txt', 'r') as skip_fd:
|
|
skip_tests = skip_fd.read().splitlines()
|
|
|
|
paths = glob.glob('test/sanity/code-smell/*')
|
|
paths = sorted(p for p in paths if os.access(p, os.X_OK) and os.path.isfile(p) and os.path.basename(p) not in skip_tests)
|
|
|
|
tests = tuple(SanityFunc(os.path.splitext(os.path.basename(p))[0], command_sanity_code_smell, script=p, intercept=False) for p in paths)
|
|
|
|
return tests
|
|
|
|
|
|
def sanity_get_tests():
|
|
"""
|
|
:rtype: tuple(SanityFunc)
|
|
"""
|
|
return SANITY_TESTS
|
|
|
|
|
|
class SanitySuccess(TestSuccess):
|
|
"""Sanity test success."""
|
|
def __init__(self, test, python_version=None):
|
|
"""
|
|
:type test: str
|
|
:type python_version: str
|
|
"""
|
|
super(SanitySuccess, self).__init__(COMMAND, test, python_version)
|
|
|
|
|
|
class SanitySkipped(TestSkipped):
|
|
"""Sanity test skipped."""
|
|
def __init__(self, test, python_version=None):
|
|
"""
|
|
:type test: str
|
|
:type python_version: str
|
|
"""
|
|
super(SanitySkipped, self).__init__(COMMAND, test, python_version)
|
|
|
|
|
|
class SanityFailure(TestFailure):
|
|
"""Sanity test failure."""
|
|
def __init__(self, test, python_version=None, messages=None, summary=None):
|
|
"""
|
|
:type test: str
|
|
:type python_version: str
|
|
:type messages: list[SanityMessage]
|
|
:type summary: str
|
|
"""
|
|
super(SanityFailure, self).__init__(COMMAND, test, python_version, messages, summary)
|
|
|
|
|
|
class SanityMessage(TestMessage):
|
|
"""Single sanity test message for one file."""
|
|
pass
|
|
|
|
|
|
class SanityTargets(object):
|
|
"""Sanity test target information."""
|
|
def __init__(self, include, exclude, require):
|
|
"""
|
|
:type include: list[str]
|
|
:type exclude: list[str]
|
|
:type require: list[str]
|
|
"""
|
|
self.all = not include
|
|
self.targets = tuple(sorted(walk_sanity_targets()))
|
|
self.include = walk_internal_targets(self.targets, include, exclude, require)
|
|
self.include_external, self.exclude_external = walk_external_targets(self.targets, include, exclude, require)
|
|
|
|
|
|
class SanityTest(object):
|
|
"""Sanity test base class."""
|
|
def __init__(self, name):
|
|
self.name = name
|
|
|
|
|
|
class SanityFunc(SanityTest):
|
|
"""Sanity test function information."""
|
|
def __init__(self, name, func, intercept=True, script=None):
|
|
"""
|
|
:type name: str
|
|
:type func: (SanityConfig, SanityTargets) -> SanityResult
|
|
:type intercept: bool
|
|
:type script: str | None
|
|
"""
|
|
super(SanityFunc, self).__init__(name)
|
|
|
|
self.func = func
|
|
self.intercept = intercept
|
|
self.script = script
|
|
|
|
|
|
SANITY_TESTS = (
|
|
SanityFunc('shellcheck', command_sanity_shellcheck, intercept=False),
|
|
SanityFunc('pep8', command_sanity_pep8, intercept=False),
|
|
SanityFunc('pylint', command_sanity_pylint, intercept=False),
|
|
SanityFunc('yamllint', command_sanity_yamllint, intercept=False),
|
|
SanityFunc('rstcheck', command_sanity_rstcheck, intercept=False),
|
|
SanityFunc('validate-modules', command_sanity_validate_modules, intercept=False),
|
|
SanityFunc('ansible-doc', command_sanity_ansible_doc),
|
|
)
|
|
|
|
|
|
def sanity_init():
|
|
"""Initialize full sanity test list (includes code-smell scripts determined at runtime)."""
|
|
global SANITY_TESTS # pylint: disable=locally-disabled, global-statement
|
|
SANITY_TESTS = tuple(sorted(SANITY_TESTS + collect_code_smell_tests(), key=lambda k: k.name))
|