Hacked By AnonymousFox
# Copyright (C) 2012 Canonical Ltd.
#
# Author: Scott Moser <scott.moser@canonical.com>
#
# This file is part of cloud-init. See LICENSE file for license information.
from collections import defaultdict
from itertools import chain
from typing import Any, Dict, List, Tuple
import yaml
YAMLError = yaml.YAMLError
# SchemaPathMarks track the path to an element within a loaded YAML file.
# The start_mark and end_mark contain the row and column indicators
# which represent the coordinates where the schema element begins and ends.
class SchemaPathMarks:
def __init__(self, path: str, start_mark: yaml.Mark, end_mark: yaml.Mark):
self.path = path
self.start_mark = start_mark
self.end_mark = end_mark
def __contains__(self, other):
"""Return whether other start/end marks are within self marks."""
if (
other.start_mark.line < self.start_mark.line
or other.end_mark.line > self.end_mark.line
):
return False
if (
other.start_mark.line == self.start_mark.line
and other.start_mark.column < self.start_mark.column
):
return False
if (
other.end_mark.line == self.end_mark.line
and other.end_mark.column > self.end_mark.column
):
return False
return True
def __eq__(self, other):
return (
self.start_mark.line == other.start_mark.line
and self.start_mark.column == other.start_mark.column
and self.end_mark.line == other.end_mark.line
and self.end_mark.column == other.end_mark.column
)
class _CustomSafeLoader(yaml.SafeLoader):
def construct_python_unicode(self, node):
return super().construct_scalar(node)
def _find_closest_parent(child_mark, marks):
for mark in marks[::-1]:
if child_mark in mark and not child_mark == mark:
return mark
return None
def _reparent_schema_mark_children(line_marks: List[SchemaPathMarks]):
"""
Update any SchemaPathMarks.path for items not under the proper parent.
"""
for mark in line_marks:
parent = _find_closest_parent(mark, line_marks)
if parent:
path_prefix, _path_idx = mark.path.rsplit(".", 1)
if mark.path == parent.path or not mark.path.startswith(
parent.path
):
# Reparent, replacing only the first match of path_prefix
mark.path = mark.path.replace(path_prefix, parent.path, 1)
def _add_mark_and_reparent_marks(
new_mark: SchemaPathMarks, marks: List[SchemaPathMarks]
) -> List[SchemaPathMarks]:
"""Insert new_mark into marks, ordering ancestors first.
Reparent existing SchemaPathMarks.path when new_mark is a parent of
an existing mark item.
Because schema processing is depth first, leaf/child mappings and
sequences may be processed for SchemaPathMarks before their parents.
This leads to SchemaPathMarks.path of 'grandchildren' being incorrectly
parented by the root dictionary instead of an intermediary parents below
root.
Walk through the list of existing marks and reparent marks that are
contained within the new_mark.
"""
new_marks = []
reparent_paths = False
for mark in marks:
if mark not in new_mark:
new_marks.append(mark)
continue
if new_mark not in new_marks:
reparent_paths = True
# Insert new_mark first as it is a parent of mark
new_marks.append(new_mark)
new_marks.append(mark)
if reparent_paths:
_reparent_schema_mark_children(new_marks)
else:
new_marks.append(new_mark)
return new_marks
class _CustomSafeLoaderWithMarks(yaml.SafeLoader):
"""A loader which provides line and column start and end marks for YAML.
If the YAML loaded represents a dictionary, get_single_data will inject
a top-level "schemamarks" key in that dictionary which can be used at
call-sites to process YAML paths schemamark metadata when annotating
YAML files for errors.
The schemamarks key is dictionary where each key is a dot-delimited path
into the YAML object. Each dot represents an element that is nested under
a parent and list items are represented with the format
`<parent>.<list-index>`.
The values in schemamarks will be the line number in the original content
where YAML element begins to aid in annotation when encountering schema
errors.
The example YAML shows expected schemamarks for both dicts and lists:
one: val1
two:
subtwo: val2
three: [val3, val4]
schemamarks == {
"one": 1, "two": 2, "two.subtwo": 3, "three": 4, "three.0": 4,
"three.1": 4
}
"""
def __init__(self, stream):
super().__init__(stream)
self.schemamarks_by_line: Dict[
int, List[SchemaPathMarks]
] = defaultdict(list)
def _get_nested_path_prefix(self, node):
if node.start_mark.line in self.schemamarks_by_line:
# Find most specific match
most_specific_mark = self.schemamarks_by_line[
node.start_mark.line
][0]
for path_mark in self.schemamarks_by_line[node.start_mark.line][
1:
]:
if node in path_mark and path_mark in most_specific_mark:
most_specific_mark = path_mark
if node in most_specific_mark:
return most_specific_mark.path + "."
for _line_num, schema_marks in sorted(
self.schemamarks_by_line.items(), reverse=True
):
for mark in schema_marks[::-1]:
if node in mark:
return f"{mark.path}."
return ""
def construct_mapping(self, node):
mapping = super().construct_mapping(node)
nested_path_prefix = self._get_nested_path_prefix(node)
for key_node, value_node in node.value:
node_key_path = f"{nested_path_prefix}{key_node.value}"
line_num = key_node.start_mark.line
new_mark = SchemaPathMarks(
node_key_path, key_node.start_mark, value_node.end_mark
)
schema_marks = self.schemamarks_by_line[line_num]
new_marks = _add_mark_and_reparent_marks(new_mark, schema_marks)
self.schemamarks_by_line[line_num] = new_marks
return mapping
def construct_sequence(self, node, deep=False):
sequence = super().construct_sequence(node, deep=True)
nested_path_prefix = self._get_nested_path_prefix(node)
for index, sequence_item in enumerate(node.value):
line_num = sequence_item.start_mark.line
node_key_path = f"{nested_path_prefix}{index}"
new_mark = SchemaPathMarks(
node_key_path, sequence_item.start_mark, sequence_item.end_mark
)
if line_num not in self.schemamarks_by_line:
self.schemamarks_by_line[line_num] = [new_mark]
else:
if line_num == sequence_item.end_mark.line:
schema_marks = self.schemamarks_by_line[line_num]
new_marks = _add_mark_and_reparent_marks(
new_mark, schema_marks
)
self.schemamarks_by_line[line_num] = new_marks
else: # Incorrect multi-line mapping or sequence object.
for inner_line in range(
line_num, sequence_item.end_mark.line
):
if inner_line in self.schemamarks_by_line:
schema_marks = self.schemamarks_by_line[inner_line]
new_marks = _add_mark_and_reparent_marks(
new_mark, schema_marks
)
if (
inner_line == line_num
and schema_marks[0].path != node_key_path
):
new_marks.insert(
0,
SchemaPathMarks(
node_key_path,
schema_marks[0].start_mark,
schema_marks[-1].end_mark,
),
)
self.schemamarks_by_line[inner_line] = new_marks
return sequence
def get_single_data(self):
data = super().get_single_data()
if isinstance(data, dict): # valid cloud-config schema is a dict
data["schemamarks"] = dict(
[
(v.path, v.start_mark.line + 1) # 1-based human-readable
for v in chain(*self.schemamarks_by_line.values())
]
)
return data
_CustomSafeLoader.add_constructor(
"tag:yaml.org,2002:python/unicode",
_CustomSafeLoader.construct_python_unicode,
)
class NoAliasSafeDumper(yaml.dumper.SafeDumper):
"""A class which avoids constructing anchors/aliases on yaml dump"""
def ignore_aliases(self, data):
return True
def load_with_marks(blob) -> Tuple[Any, Dict[str, int]]:
"""Perform YAML SafeLoad and track start and end marks during parse.
JSON schema errors come with an encoded object path such as:
<key1>.<key2>.<list_item_index>
YAML loader needs to preserve a mapping of schema path to line and column
marks to annotate original content with JSON schema error marks for the
command:
cloud-init devel schema --annotate
"""
result = yaml.load(blob, Loader=_CustomSafeLoaderWithMarks)
if not isinstance(result, dict):
schemamarks = {}
else:
schemamarks = result.pop("schemamarks")
return result, schemamarks
def load(blob):
return yaml.load(blob, Loader=_CustomSafeLoader)
def dumps(obj, explicit_start=True, explicit_end=True, noalias=False):
"""Return data in nicely formatted yaml."""
return yaml.dump(
obj,
line_break="\n",
indent=4,
explicit_start=explicit_start,
explicit_end=explicit_end,
default_flow_style=False,
Dumper=(NoAliasSafeDumper if noalias else yaml.dumper.SafeDumper),
)
Hacked By AnonymousFox1.0, Coded By AnonymousFox