mirror of
https://github.com/thegeeklab/ansible-later.git
synced 2024-11-26 23:00:36 +00:00
576 lines
19 KiB
Python
576 lines
19 KiB
Python
|
# -*- coding: utf-8 -*-
|
||
|
# Copyright (C) 2016 Adrien Vergé
|
||
|
#
|
||
|
# This program is free software: you can redistribute it and/or modify
|
||
|
# it under the terms of the GNU General Public License as published by
|
||
|
# the Free Software Foundation, either version 3 of the License, or
|
||
|
# (at your option) any later version.
|
||
|
#
|
||
|
# This program is distributed in the hope that it will be useful,
|
||
|
# but WITHOUT ANY WARRANTY; without even the implied warranty of
|
||
|
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
||
|
# GNU General Public License for more details.
|
||
|
#
|
||
|
# You should have received a copy of the GNU General Public License
|
||
|
# along with this program. If not, see <http://www.gnu.org/licenses/>.
|
||
|
|
||
|
"""
|
||
|
Use this rule to control the indentation.
|
||
|
|
||
|
.. rubric:: Options
|
||
|
|
||
|
* ``spaces`` defines the indentation width, in spaces. Set either to an integer
|
||
|
(e.g. ``2`` or ``4``, representing the number of spaces in an indentation
|
||
|
level) or to ``consistent`` to allow any number, as long as it remains the
|
||
|
same within the file.
|
||
|
* ``indent-sequences`` defines whether block sequences should be indented or
|
||
|
not (when in a mapping, this indentation is not mandatory -- some people
|
||
|
perceive the ``-`` as part of the indentation). Possible values: ``true``,
|
||
|
``false``, ``whatever`` and ``consistent``. ``consistent`` requires either
|
||
|
all block sequences to be indented, or none to be. ``whatever`` means either
|
||
|
indenting or not indenting individual block sequences is OK.
|
||
|
* ``check-multi-line-strings`` defines whether to lint indentation in
|
||
|
multi-line strings. Set to ``true`` to enable, ``false`` to disable.
|
||
|
|
||
|
.. rubric:: Examples
|
||
|
|
||
|
#. With ``indentation: {spaces: 1}``
|
||
|
|
||
|
the following code snippet would **PASS**:
|
||
|
::
|
||
|
|
||
|
history:
|
||
|
- name: Unix
|
||
|
date: 1969
|
||
|
- name: Linux
|
||
|
date: 1991
|
||
|
nest:
|
||
|
recurse:
|
||
|
- haystack:
|
||
|
needle
|
||
|
|
||
|
#. With ``indentation: {spaces: 4}``
|
||
|
|
||
|
the following code snippet would **PASS**:
|
||
|
::
|
||
|
|
||
|
history:
|
||
|
- name: Unix
|
||
|
date: 1969
|
||
|
- name: Linux
|
||
|
date: 1991
|
||
|
nest:
|
||
|
recurse:
|
||
|
- haystack:
|
||
|
needle
|
||
|
|
||
|
the following code snippet would **FAIL**:
|
||
|
::
|
||
|
|
||
|
history:
|
||
|
- name: Unix
|
||
|
date: 1969
|
||
|
- name: Linux
|
||
|
date: 1991
|
||
|
nest:
|
||
|
recurse:
|
||
|
- haystack:
|
||
|
needle
|
||
|
|
||
|
#. With ``indentation: {spaces: consistent}``
|
||
|
|
||
|
the following code snippet would **PASS**:
|
||
|
::
|
||
|
|
||
|
history:
|
||
|
- name: Unix
|
||
|
date: 1969
|
||
|
- name: Linux
|
||
|
date: 1991
|
||
|
nest:
|
||
|
recurse:
|
||
|
- haystack:
|
||
|
needle
|
||
|
|
||
|
the following code snippet would **FAIL**:
|
||
|
::
|
||
|
|
||
|
some:
|
||
|
Russian:
|
||
|
dolls
|
||
|
|
||
|
#. With ``indentation: {spaces: 2, indent-sequences: false}``
|
||
|
|
||
|
the following code snippet would **PASS**:
|
||
|
::
|
||
|
|
||
|
list:
|
||
|
- flying
|
||
|
- spaghetti
|
||
|
- monster
|
||
|
|
||
|
the following code snippet would **FAIL**:
|
||
|
::
|
||
|
|
||
|
list:
|
||
|
- flying
|
||
|
- spaghetti
|
||
|
- monster
|
||
|
|
||
|
#. With ``indentation: {spaces: 2, indent-sequences: whatever}``
|
||
|
|
||
|
the following code snippet would **PASS**:
|
||
|
::
|
||
|
|
||
|
list:
|
||
|
- flying:
|
||
|
- spaghetti
|
||
|
- monster
|
||
|
- not flying:
|
||
|
- spaghetti
|
||
|
- sauce
|
||
|
|
||
|
#. With ``indentation: {spaces: 2, indent-sequences: consistent}``
|
||
|
|
||
|
the following code snippet would **PASS**:
|
||
|
::
|
||
|
|
||
|
- flying:
|
||
|
- spaghetti
|
||
|
- monster
|
||
|
- not flying:
|
||
|
- spaghetti
|
||
|
- sauce
|
||
|
|
||
|
the following code snippet would **FAIL**:
|
||
|
::
|
||
|
|
||
|
- flying:
|
||
|
- spaghetti
|
||
|
- monster
|
||
|
- not flying:
|
||
|
- spaghetti
|
||
|
- sauce
|
||
|
|
||
|
#. With ``indentation: {spaces: 4, check-multi-line-strings: true}``
|
||
|
|
||
|
the following code snippet would **PASS**:
|
||
|
::
|
||
|
|
||
|
Blaise Pascal:
|
||
|
Je vous écris une longue lettre parce que
|
||
|
je n'ai pas le temps d'en écrire une courte.
|
||
|
|
||
|
the following code snippet would **PASS**:
|
||
|
::
|
||
|
|
||
|
Blaise Pascal: Je vous écris une longue lettre parce que
|
||
|
je n'ai pas le temps d'en écrire une courte.
|
||
|
|
||
|
the following code snippet would **FAIL**:
|
||
|
::
|
||
|
|
||
|
Blaise Pascal: Je vous écris une longue lettre parce que
|
||
|
je n'ai pas le temps d'en écrire une courte.
|
||
|
|
||
|
the following code snippet would **FAIL**:
|
||
|
::
|
||
|
|
||
|
C code:
|
||
|
void main() {
|
||
|
printf("foo");
|
||
|
}
|
||
|
|
||
|
the following code snippet would **PASS**:
|
||
|
::
|
||
|
|
||
|
C code:
|
||
|
void main() {
|
||
|
printf("bar");
|
||
|
}
|
||
|
"""
|
||
|
|
||
|
import yaml
|
||
|
|
||
|
from yamllint.linter import LintProblem
|
||
|
from yamllint.rules.common import is_explicit_key, get_real_end_line
|
||
|
|
||
|
|
||
|
ID = 'indentation'
|
||
|
TYPE = 'token'
|
||
|
CONF = {'spaces': (int, 'consistent'),
|
||
|
'indent-sequences': (bool, 'whatever', 'consistent'),
|
||
|
'check-multi-line-strings': bool}
|
||
|
DEFAULT = {'spaces': 'consistent',
|
||
|
'indent-sequences': True,
|
||
|
'check-multi-line-strings': False}
|
||
|
|
||
|
ROOT, B_MAP, F_MAP, B_SEQ, F_SEQ, B_ENT, KEY, VAL = range(8)
|
||
|
labels = ('ROOT', 'B_MAP', 'F_MAP', 'B_SEQ', 'F_SEQ', 'B_ENT', 'KEY', 'VAL')
|
||
|
|
||
|
|
||
|
class Parent(object):
|
||
|
def __init__(self, type, indent, line_indent=None):
|
||
|
self.type = type
|
||
|
self.indent = indent
|
||
|
self.line_indent = line_indent
|
||
|
self.explicit_key = False
|
||
|
self.implicit_block_seq = False
|
||
|
|
||
|
def __repr__(self):
|
||
|
return '%s:%d' % (labels[self.type], self.indent)
|
||
|
|
||
|
|
||
|
def check_scalar_indentation(conf, token, context):
|
||
|
if token.start_mark.line == token.end_mark.line:
|
||
|
return
|
||
|
|
||
|
def compute_expected_indent(found_indent):
|
||
|
def detect_indent(base_indent):
|
||
|
if not isinstance(context['spaces'], int):
|
||
|
context['spaces'] = found_indent - base_indent
|
||
|
return base_indent + context['spaces']
|
||
|
|
||
|
if token.plain:
|
||
|
return token.start_mark.column
|
||
|
elif token.style in ('"', "'"):
|
||
|
return token.start_mark.column + 1
|
||
|
elif token.style in ('>', '|'):
|
||
|
if context['stack'][-1].type == B_ENT:
|
||
|
# - >
|
||
|
# multi
|
||
|
# line
|
||
|
return detect_indent(token.start_mark.column)
|
||
|
elif context['stack'][-1].type == KEY:
|
||
|
assert context['stack'][-1].explicit_key
|
||
|
# - ? >
|
||
|
# multi-line
|
||
|
# key
|
||
|
# : >
|
||
|
# multi-line
|
||
|
# value
|
||
|
return detect_indent(token.start_mark.column)
|
||
|
elif context['stack'][-1].type == VAL:
|
||
|
if token.start_mark.line + 1 > context['cur_line']:
|
||
|
# - key:
|
||
|
# >
|
||
|
# multi
|
||
|
# line
|
||
|
return detect_indent(context['stack'][-1].indent)
|
||
|
elif context['stack'][-2].explicit_key:
|
||
|
# - ? key
|
||
|
# : >
|
||
|
# multi-line
|
||
|
# value
|
||
|
return detect_indent(token.start_mark.column)
|
||
|
else:
|
||
|
# - key: >
|
||
|
# multi
|
||
|
# line
|
||
|
return detect_indent(context['stack'][-2].indent)
|
||
|
else:
|
||
|
return detect_indent(context['stack'][-1].indent)
|
||
|
|
||
|
expected_indent = None
|
||
|
|
||
|
line_no = token.start_mark.line + 1
|
||
|
|
||
|
line_start = token.start_mark.pointer
|
||
|
while True:
|
||
|
line_start = token.start_mark.buffer.find(
|
||
|
'\n', line_start, token.end_mark.pointer - 1) + 1
|
||
|
if line_start == 0:
|
||
|
break
|
||
|
line_no += 1
|
||
|
|
||
|
indent = 0
|
||
|
while token.start_mark.buffer[line_start + indent] == ' ':
|
||
|
indent += 1
|
||
|
if token.start_mark.buffer[line_start + indent] == '\n':
|
||
|
continue
|
||
|
|
||
|
if expected_indent is None:
|
||
|
expected_indent = compute_expected_indent(indent)
|
||
|
|
||
|
if indent != expected_indent:
|
||
|
yield LintProblem(line_no, indent + 1,
|
||
|
'wrong indentation: expected %d but found %d' %
|
||
|
(expected_indent, indent))
|
||
|
|
||
|
|
||
|
def _check(conf, token, prev, next, nextnext, context):
|
||
|
if 'stack' not in context:
|
||
|
context['stack'] = [Parent(ROOT, 0)]
|
||
|
context['cur_line'] = -1
|
||
|
context['spaces'] = conf['spaces']
|
||
|
context['indent-sequences'] = conf['indent-sequences']
|
||
|
|
||
|
# Step 1: Lint
|
||
|
|
||
|
is_visible = (
|
||
|
not isinstance(token, (yaml.StreamStartToken, yaml.StreamEndToken)) and
|
||
|
not isinstance(token, yaml.BlockEndToken) and
|
||
|
not (isinstance(token, yaml.ScalarToken) and token.value == ''))
|
||
|
first_in_line = (is_visible and
|
||
|
token.start_mark.line + 1 > context['cur_line'])
|
||
|
|
||
|
def detect_indent(base_indent, next):
|
||
|
if not isinstance(context['spaces'], int):
|
||
|
context['spaces'] = next.start_mark.column - base_indent
|
||
|
return base_indent + context['spaces']
|
||
|
|
||
|
if first_in_line:
|
||
|
found_indentation = token.start_mark.column
|
||
|
expected = context['stack'][-1].indent
|
||
|
|
||
|
if isinstance(token, (yaml.FlowMappingEndToken,
|
||
|
yaml.FlowSequenceEndToken)):
|
||
|
expected = context['stack'][-1].line_indent
|
||
|
elif (context['stack'][-1].type == KEY and
|
||
|
context['stack'][-1].explicit_key and
|
||
|
not isinstance(token, yaml.ValueToken)):
|
||
|
expected = detect_indent(expected, token)
|
||
|
|
||
|
if found_indentation != expected:
|
||
|
yield LintProblem(token.start_mark.line + 1, found_indentation + 1,
|
||
|
'wrong indentation: expected %d but found %d' %
|
||
|
(expected, found_indentation))
|
||
|
|
||
|
if (isinstance(token, yaml.ScalarToken) and
|
||
|
conf['check-multi-line-strings']):
|
||
|
for problem in check_scalar_indentation(conf, token, context):
|
||
|
yield problem
|
||
|
|
||
|
# Step 2.a:
|
||
|
|
||
|
if is_visible:
|
||
|
context['cur_line'] = get_real_end_line(token)
|
||
|
if first_in_line:
|
||
|
context['cur_line_indent'] = found_indentation
|
||
|
|
||
|
# Step 2.b: Update state
|
||
|
|
||
|
if isinstance(token, yaml.BlockMappingStartToken):
|
||
|
# - a: 1
|
||
|
# or
|
||
|
# - ? a
|
||
|
# : 1
|
||
|
# or
|
||
|
# - ?
|
||
|
# a
|
||
|
# : 1
|
||
|
assert isinstance(next, yaml.KeyToken)
|
||
|
assert next.start_mark.line == token.start_mark.line
|
||
|
|
||
|
indent = token.start_mark.column
|
||
|
|
||
|
context['stack'].append(Parent(B_MAP, indent))
|
||
|
|
||
|
elif isinstance(token, yaml.FlowMappingStartToken):
|
||
|
if next.start_mark.line == token.start_mark.line:
|
||
|
# - {a: 1, b: 2}
|
||
|
indent = next.start_mark.column
|
||
|
else:
|
||
|
# - {
|
||
|
# a: 1, b: 2
|
||
|
# }
|
||
|
indent = detect_indent(context['cur_line_indent'], next)
|
||
|
|
||
|
context['stack'].append(Parent(F_MAP, indent,
|
||
|
line_indent=context['cur_line_indent']))
|
||
|
|
||
|
elif isinstance(token, yaml.BlockSequenceStartToken):
|
||
|
# - - a
|
||
|
# - b
|
||
|
assert isinstance(next, yaml.BlockEntryToken)
|
||
|
assert next.start_mark.line == token.start_mark.line
|
||
|
|
||
|
indent = token.start_mark.column
|
||
|
|
||
|
context['stack'].append(Parent(B_SEQ, indent))
|
||
|
|
||
|
elif (isinstance(token, yaml.BlockEntryToken) and
|
||
|
# in case of an empty entry
|
||
|
not isinstance(next, (yaml.BlockEntryToken, yaml.BlockEndToken))):
|
||
|
# It looks like pyyaml doesn't issue BlockSequenceStartTokens when the
|
||
|
# list is not indented. We need to compensate that.
|
||
|
if context['stack'][-1].type != B_SEQ:
|
||
|
context['stack'].append(Parent(B_SEQ, token.start_mark.column))
|
||
|
context['stack'][-1].implicit_block_seq = True
|
||
|
|
||
|
if next.start_mark.line == token.end_mark.line:
|
||
|
# - item 1
|
||
|
# - item 2
|
||
|
indent = next.start_mark.column
|
||
|
elif next.start_mark.column == token.start_mark.column:
|
||
|
# -
|
||
|
# key: value
|
||
|
indent = next.start_mark.column
|
||
|
else:
|
||
|
# -
|
||
|
# item 1
|
||
|
# -
|
||
|
# key:
|
||
|
# value
|
||
|
indent = detect_indent(token.start_mark.column, next)
|
||
|
|
||
|
context['stack'].append(Parent(B_ENT, indent))
|
||
|
|
||
|
elif isinstance(token, yaml.FlowSequenceStartToken):
|
||
|
if next.start_mark.line == token.start_mark.line:
|
||
|
# - [a, b]
|
||
|
indent = next.start_mark.column
|
||
|
else:
|
||
|
# - [
|
||
|
# a, b
|
||
|
# ]
|
||
|
indent = detect_indent(context['cur_line_indent'], next)
|
||
|
|
||
|
context['stack'].append(Parent(F_SEQ, indent,
|
||
|
line_indent=context['cur_line_indent']))
|
||
|
|
||
|
elif isinstance(token, yaml.KeyToken):
|
||
|
indent = context['stack'][-1].indent
|
||
|
|
||
|
context['stack'].append(Parent(KEY, indent))
|
||
|
|
||
|
context['stack'][-1].explicit_key = is_explicit_key(token)
|
||
|
|
||
|
elif isinstance(token, yaml.ValueToken):
|
||
|
assert context['stack'][-1].type == KEY
|
||
|
|
||
|
# Special cases:
|
||
|
# key: &anchor
|
||
|
# value
|
||
|
# and:
|
||
|
# key: !!tag
|
||
|
# value
|
||
|
if isinstance(next, (yaml.AnchorToken, yaml.TagToken)):
|
||
|
if (next.start_mark.line == prev.start_mark.line and
|
||
|
next.start_mark.line < nextnext.start_mark.line):
|
||
|
next = nextnext
|
||
|
|
||
|
# Only if value is not empty
|
||
|
if not isinstance(next, (yaml.BlockEndToken,
|
||
|
yaml.FlowMappingEndToken,
|
||
|
yaml.FlowSequenceEndToken,
|
||
|
yaml.KeyToken)):
|
||
|
if context['stack'][-1].explicit_key:
|
||
|
# ? k
|
||
|
# : value
|
||
|
# or
|
||
|
# ? k
|
||
|
# :
|
||
|
# value
|
||
|
indent = detect_indent(context['stack'][-1].indent, next)
|
||
|
elif next.start_mark.line == prev.start_mark.line:
|
||
|
# k: value
|
||
|
indent = next.start_mark.column
|
||
|
elif isinstance(next, (yaml.BlockSequenceStartToken,
|
||
|
yaml.BlockEntryToken)):
|
||
|
# NOTE: We add BlockEntryToken in the test above because
|
||
|
# sometimes BlockSequenceStartToken are not issued. Try
|
||
|
# yaml.scan()ning this:
|
||
|
# '- lib:\n'
|
||
|
# ' - var\n'
|
||
|
if context['indent-sequences'] is False:
|
||
|
indent = context['stack'][-1].indent
|
||
|
elif context['indent-sequences'] is True:
|
||
|
if (context['spaces'] == 'consistent' and
|
||
|
next.start_mark.column -
|
||
|
context['stack'][-1].indent == 0):
|
||
|
# In this case, the block sequence item is not indented
|
||
|
# (while it should be), but we don't know yet the
|
||
|
# indentation it should have (because `spaces` is
|
||
|
# `consistent` and its value has not been computed yet
|
||
|
# -- this is probably the beginning of the document).
|
||
|
# So we choose an arbitrary value (2).
|
||
|
indent = 2
|
||
|
else:
|
||
|
indent = detect_indent(context['stack'][-1].indent,
|
||
|
next)
|
||
|
else: # 'whatever' or 'consistent'
|
||
|
if next.start_mark.column == context['stack'][-1].indent:
|
||
|
# key:
|
||
|
# - e1
|
||
|
# - e2
|
||
|
if context['indent-sequences'] == 'consistent':
|
||
|
context['indent-sequences'] = False
|
||
|
indent = context['stack'][-1].indent
|
||
|
else:
|
||
|
if context['indent-sequences'] == 'consistent':
|
||
|
context['indent-sequences'] = True
|
||
|
# key:
|
||
|
# - e1
|
||
|
# - e2
|
||
|
indent = detect_indent(context['stack'][-1].indent,
|
||
|
next)
|
||
|
else:
|
||
|
# k:
|
||
|
# value
|
||
|
indent = detect_indent(context['stack'][-1].indent, next)
|
||
|
|
||
|
context['stack'].append(Parent(VAL, indent))
|
||
|
|
||
|
consumed_current_token = False
|
||
|
while True:
|
||
|
if (context['stack'][-1].type == F_SEQ and
|
||
|
isinstance(token, yaml.FlowSequenceEndToken) and
|
||
|
not consumed_current_token):
|
||
|
context['stack'].pop()
|
||
|
consumed_current_token = True
|
||
|
|
||
|
elif (context['stack'][-1].type == F_MAP and
|
||
|
isinstance(token, yaml.FlowMappingEndToken) and
|
||
|
not consumed_current_token):
|
||
|
context['stack'].pop()
|
||
|
consumed_current_token = True
|
||
|
|
||
|
elif (context['stack'][-1].type in (B_MAP, B_SEQ) and
|
||
|
isinstance(token, yaml.BlockEndToken) and
|
||
|
not context['stack'][-1].implicit_block_seq and
|
||
|
not consumed_current_token):
|
||
|
context['stack'].pop()
|
||
|
consumed_current_token = True
|
||
|
|
||
|
elif (context['stack'][-1].type == B_ENT and
|
||
|
not isinstance(token, yaml.BlockEntryToken) and
|
||
|
context['stack'][-2].implicit_block_seq and
|
||
|
not isinstance(token, (yaml.AnchorToken, yaml.TagToken)) and
|
||
|
not isinstance(next, yaml.BlockEntryToken)):
|
||
|
context['stack'].pop()
|
||
|
context['stack'].pop()
|
||
|
|
||
|
elif (context['stack'][-1].type == B_ENT and
|
||
|
isinstance(next, (yaml.BlockEntryToken, yaml.BlockEndToken))):
|
||
|
context['stack'].pop()
|
||
|
|
||
|
elif (context['stack'][-1].type == VAL and
|
||
|
not isinstance(token, yaml.ValueToken) and
|
||
|
not isinstance(token, (yaml.AnchorToken, yaml.TagToken))):
|
||
|
assert context['stack'][-2].type == KEY
|
||
|
context['stack'].pop()
|
||
|
context['stack'].pop()
|
||
|
|
||
|
elif (context['stack'][-1].type == KEY and
|
||
|
isinstance(next, (yaml.BlockEndToken,
|
||
|
yaml.FlowMappingEndToken,
|
||
|
yaml.FlowSequenceEndToken,
|
||
|
yaml.KeyToken))):
|
||
|
# A key without a value: it's part of a set. Let's drop this key
|
||
|
# and leave room for the next one.
|
||
|
context['stack'].pop()
|
||
|
|
||
|
else:
|
||
|
break
|
||
|
|
||
|
|
||
|
def check(conf, token, prev, next, nextnext, context):
|
||
|
try:
|
||
|
for problem in _check(conf, token, prev, next, nextnext, context):
|
||
|
yield problem
|
||
|
except AssertionError:
|
||
|
yield LintProblem(token.start_mark.line + 1,
|
||
|
token.start_mark.column + 1,
|
||
|
'cannot infer indentation: unexpected token')
|