bazarr/libs/dynaconf/vendor/ruamel/yaml/tokens.py

287 lines
7.3 KiB
Python

# # header
# coding: utf-8
from __future__ import unicode_literals
if False: # MYPY
from typing import Text, Any, Dict, Optional, List # NOQA
from .error import StreamMark # NOQA
SHOWLINES = True
class Token(object):
__slots__ = 'start_mark', 'end_mark', '_comment'
def __init__(self, start_mark, end_mark):
# type: (StreamMark, StreamMark) -> None
self.start_mark = start_mark
self.end_mark = end_mark
def __repr__(self):
# type: () -> Any
# attributes = [key for key in self.__slots__ if not key.endswith('_mark') and
# hasattr('self', key)]
attributes = [key for key in self.__slots__ if not key.endswith('_mark')]
attributes.sort()
arguments = ', '.join(['%s=%r' % (key, getattr(self, key)) for key in attributes])
if SHOWLINES:
try:
arguments += ', line: ' + str(self.start_mark.line)
except: # NOQA
pass
try:
arguments += ', comment: ' + str(self._comment)
except: # NOQA
pass
return '{}({})'.format(self.__class__.__name__, arguments)
def add_post_comment(self, comment):
# type: (Any) -> None
if not hasattr(self, '_comment'):
self._comment = [None, None]
self._comment[0] = comment
def add_pre_comments(self, comments):
# type: (Any) -> None
if not hasattr(self, '_comment'):
self._comment = [None, None]
assert self._comment[1] is None
self._comment[1] = comments
def get_comment(self):
# type: () -> Any
return getattr(self, '_comment', None)
@property
def comment(self):
# type: () -> Any
return getattr(self, '_comment', None)
def move_comment(self, target, empty=False):
# type: (Any, bool) -> Any
"""move a comment from this token to target (normally next token)
used to combine e.g. comments before a BlockEntryToken to the
ScalarToken that follows it
empty is a special for empty values -> comment after key
"""
c = self.comment
if c is None:
return
# don't push beyond last element
if isinstance(target, (StreamEndToken, DocumentStartToken)):
return
delattr(self, '_comment')
tc = target.comment
if not tc: # target comment, just insert
# special for empty value in key: value issue 25
if empty:
c = [c[0], c[1], None, None, c[0]]
target._comment = c
# nprint('mco2:', self, target, target.comment, empty)
return self
if c[0] and tc[0] or c[1] and tc[1]:
raise NotImplementedError('overlap in comment %r %r' % (c, tc))
if c[0]:
tc[0] = c[0]
if c[1]:
tc[1] = c[1]
return self
def split_comment(self):
# type: () -> Any
""" split the post part of a comment, and return it
as comment to be added. Delete second part if [None, None]
abc: # this goes to sequence
# this goes to first element
- first element
"""
comment = self.comment
if comment is None or comment[0] is None:
return None # nothing to do
ret_val = [comment[0], None]
if comment[1] is None:
delattr(self, '_comment')
return ret_val
# class BOMToken(Token):
# id = '<byte order mark>'
class DirectiveToken(Token):
__slots__ = 'name', 'value'
id = '<directive>'
def __init__(self, name, value, start_mark, end_mark):
# type: (Any, Any, Any, Any) -> None
Token.__init__(self, start_mark, end_mark)
self.name = name
self.value = value
class DocumentStartToken(Token):
__slots__ = ()
id = '<document start>'
class DocumentEndToken(Token):
__slots__ = ()
id = '<document end>'
class StreamStartToken(Token):
__slots__ = ('encoding',)
id = '<stream start>'
def __init__(self, start_mark=None, end_mark=None, encoding=None):
# type: (Any, Any, Any) -> None
Token.__init__(self, start_mark, end_mark)
self.encoding = encoding
class StreamEndToken(Token):
__slots__ = ()
id = '<stream end>'
class BlockSequenceStartToken(Token):
__slots__ = ()
id = '<block sequence start>'
class BlockMappingStartToken(Token):
__slots__ = ()
id = '<block mapping start>'
class BlockEndToken(Token):
__slots__ = ()
id = '<block end>'
class FlowSequenceStartToken(Token):
__slots__ = ()
id = '['
class FlowMappingStartToken(Token):
__slots__ = ()
id = '{'
class FlowSequenceEndToken(Token):
__slots__ = ()
id = ']'
class FlowMappingEndToken(Token):
__slots__ = ()
id = '}'
class KeyToken(Token):
__slots__ = ()
id = '?'
# def x__repr__(self):
# return 'KeyToken({})'.format(
# self.start_mark.buffer[self.start_mark.index:].split(None, 1)[0])
class ValueToken(Token):
__slots__ = ()
id = ':'
class BlockEntryToken(Token):
__slots__ = ()
id = '-'
class FlowEntryToken(Token):
__slots__ = ()
id = ','
class AliasToken(Token):
__slots__ = ('value',)
id = '<alias>'
def __init__(self, value, start_mark, end_mark):
# type: (Any, Any, Any) -> None
Token.__init__(self, start_mark, end_mark)
self.value = value
class AnchorToken(Token):
__slots__ = ('value',)
id = '<anchor>'
def __init__(self, value, start_mark, end_mark):
# type: (Any, Any, Any) -> None
Token.__init__(self, start_mark, end_mark)
self.value = value
class TagToken(Token):
__slots__ = ('value',)
id = '<tag>'
def __init__(self, value, start_mark, end_mark):
# type: (Any, Any, Any) -> None
Token.__init__(self, start_mark, end_mark)
self.value = value
class ScalarToken(Token):
__slots__ = 'value', 'plain', 'style'
id = '<scalar>'
def __init__(self, value, plain, start_mark, end_mark, style=None):
# type: (Any, Any, Any, Any, Any) -> None
Token.__init__(self, start_mark, end_mark)
self.value = value
self.plain = plain
self.style = style
class CommentToken(Token):
__slots__ = 'value', 'pre_done'
id = '<comment>'
def __init__(self, value, start_mark, end_mark):
# type: (Any, Any, Any) -> None
Token.__init__(self, start_mark, end_mark)
self.value = value
def reset(self):
# type: () -> None
if hasattr(self, 'pre_done'):
delattr(self, 'pre_done')
def __repr__(self):
# type: () -> Any
v = '{!r}'.format(self.value)
if SHOWLINES:
try:
v += ', line: ' + str(self.start_mark.line)
v += ', col: ' + str(self.start_mark.column)
except: # NOQA
pass
return 'CommentToken({})'.format(v)
def __eq__(self, other):
# type: (Any) -> bool
if self.start_mark != other.start_mark:
return False
if self.end_mark != other.end_mark:
return False
if self.value != other.value:
return False
return True
def __ne__(self, other):
# type: (Any) -> bool
return not self.__eq__(other)