3RNN/Lib/site-packages/pygments/lexers/data.py
2024-05-26 19:49:15 +02:00

764 lines
26 KiB
Python

"""
pygments.lexers.data
~~~~~~~~~~~~~~~~~~~~
Lexers for data file format.
:copyright: Copyright 2006-2024 by the Pygments team, see AUTHORS.
:license: BSD, see LICENSE for details.
"""
from pygments.lexer import Lexer, ExtendedRegexLexer, LexerContext, \
include, bygroups
from pygments.token import Comment, Error, Keyword, Literal, Name, Number, \
Punctuation, String, Whitespace
__all__ = ['YamlLexer', 'JsonLexer', 'JsonBareObjectLexer', 'JsonLdLexer']
class YamlLexerContext(LexerContext):
"""Indentation context for the YAML lexer."""
def __init__(self, *args, **kwds):
super().__init__(*args, **kwds)
self.indent_stack = []
self.indent = -1
self.next_indent = 0
self.block_scalar_indent = None
class YamlLexer(ExtendedRegexLexer):
"""
Lexer for YAML, a human-friendly data serialization
language.
"""
name = 'YAML'
url = 'http://yaml.org/'
aliases = ['yaml']
filenames = ['*.yaml', '*.yml']
mimetypes = ['text/x-yaml']
version_added = '0.11'
def something(token_class):
"""Do not produce empty tokens."""
def callback(lexer, match, context):
text = match.group()
if not text:
return
yield match.start(), token_class, text
context.pos = match.end()
return callback
def reset_indent(token_class):
"""Reset the indentation levels."""
def callback(lexer, match, context):
text = match.group()
context.indent_stack = []
context.indent = -1
context.next_indent = 0
context.block_scalar_indent = None
yield match.start(), token_class, text
context.pos = match.end()
return callback
def save_indent(token_class, start=False):
"""Save a possible indentation level."""
def callback(lexer, match, context):
text = match.group()
extra = ''
if start:
context.next_indent = len(text)
if context.next_indent < context.indent:
while context.next_indent < context.indent:
context.indent = context.indent_stack.pop()
if context.next_indent > context.indent:
extra = text[context.indent:]
text = text[:context.indent]
else:
context.next_indent += len(text)
if text:
yield match.start(), token_class, text
if extra:
yield match.start()+len(text), token_class.Error, extra
context.pos = match.end()
return callback
def set_indent(token_class, implicit=False):
"""Set the previously saved indentation level."""
def callback(lexer, match, context):
text = match.group()
if context.indent < context.next_indent:
context.indent_stack.append(context.indent)
context.indent = context.next_indent
if not implicit:
context.next_indent += len(text)
yield match.start(), token_class, text
context.pos = match.end()
return callback
def set_block_scalar_indent(token_class):
"""Set an explicit indentation level for a block scalar."""
def callback(lexer, match, context):
text = match.group()
context.block_scalar_indent = None
if not text:
return
increment = match.group(1)
if increment:
current_indent = max(context.indent, 0)
increment = int(increment)
context.block_scalar_indent = current_indent + increment
if text:
yield match.start(), token_class, text
context.pos = match.end()
return callback
def parse_block_scalar_empty_line(indent_token_class, content_token_class):
"""Process an empty line in a block scalar."""
def callback(lexer, match, context):
text = match.group()
if (context.block_scalar_indent is None or
len(text) <= context.block_scalar_indent):
if text:
yield match.start(), indent_token_class, text
else:
indentation = text[:context.block_scalar_indent]
content = text[context.block_scalar_indent:]
yield match.start(), indent_token_class, indentation
yield (match.start()+context.block_scalar_indent,
content_token_class, content)
context.pos = match.end()
return callback
def parse_block_scalar_indent(token_class):
"""Process indentation spaces in a block scalar."""
def callback(lexer, match, context):
text = match.group()
if context.block_scalar_indent is None:
if len(text) <= max(context.indent, 0):
context.stack.pop()
context.stack.pop()
return
context.block_scalar_indent = len(text)
else:
if len(text) < context.block_scalar_indent:
context.stack.pop()
context.stack.pop()
return
if text:
yield match.start(), token_class, text
context.pos = match.end()
return callback
def parse_plain_scalar_indent(token_class):
"""Process indentation spaces in a plain scalar."""
def callback(lexer, match, context):
text = match.group()
if len(text) <= context.indent:
context.stack.pop()
context.stack.pop()
return
if text:
yield match.start(), token_class, text
context.pos = match.end()
return callback
tokens = {
# the root rules
'root': [
# ignored whitespaces
(r'[ ]+(?=#|$)', Whitespace),
# line breaks
(r'\n+', Whitespace),
# a comment
(r'#[^\n]*', Comment.Single),
# the '%YAML' directive
(r'^%YAML(?=[ ]|$)', reset_indent(Name.Tag), 'yaml-directive'),
# the %TAG directive
(r'^%TAG(?=[ ]|$)', reset_indent(Name.Tag), 'tag-directive'),
# document start and document end indicators
(r'^(?:---|\.\.\.)(?=[ ]|$)', reset_indent(Name.Namespace),
'block-line'),
# indentation spaces
(r'[ ]*(?!\s|$)', save_indent(Whitespace, start=True),
('block-line', 'indentation')),
],
# trailing whitespaces after directives or a block scalar indicator
'ignored-line': [
# ignored whitespaces
(r'[ ]+(?=#|$)', Whitespace),
# a comment
(r'#[^\n]*', Comment.Single),
# line break
(r'\n', Whitespace, '#pop:2'),
],
# the %YAML directive
'yaml-directive': [
# the version number
(r'([ ]+)([0-9]+\.[0-9]+)',
bygroups(Whitespace, Number), 'ignored-line'),
],
# the %TAG directive
'tag-directive': [
# a tag handle and the corresponding prefix
(r'([ ]+)(!|![\w-]*!)'
r'([ ]+)(!|!?[\w;/?:@&=+$,.!~*\'()\[\]%-]+)',
bygroups(Whitespace, Keyword.Type, Whitespace, Keyword.Type),
'ignored-line'),
],
# block scalar indicators and indentation spaces
'indentation': [
# trailing whitespaces are ignored
(r'[ ]*$', something(Whitespace), '#pop:2'),
# whitespaces preceding block collection indicators
(r'[ ]+(?=[?:-](?:[ ]|$))', save_indent(Whitespace)),
# block collection indicators
(r'[?:-](?=[ ]|$)', set_indent(Punctuation.Indicator)),
# the beginning a block line
(r'[ ]*', save_indent(Whitespace), '#pop'),
],
# an indented line in the block context
'block-line': [
# the line end
(r'[ ]*(?=#|$)', something(Whitespace), '#pop'),
# whitespaces separating tokens
(r'[ ]+', Whitespace),
# key with colon
(r'''([^#,?\[\]{}"'\n]+)(:)(?=[ ]|$)''',
bygroups(Name.Tag, set_indent(Punctuation, implicit=True))),
# tags, anchors and aliases,
include('descriptors'),
# block collections and scalars
include('block-nodes'),
# flow collections and quoted scalars
include('flow-nodes'),
# a plain scalar
(r'(?=[^\s?:,\[\]{}#&*!|>\'"%@`-]|[?:-]\S)',
something(Name.Variable),
'plain-scalar-in-block-context'),
],
# tags, anchors, aliases
'descriptors': [
# a full-form tag
(r'!<[\w#;/?:@&=+$,.!~*\'()\[\]%-]+>', Keyword.Type),
# a tag in the form '!', '!suffix' or '!handle!suffix'
(r'!(?:[\w-]+!)?'
r'[\w#;/?:@&=+$,.!~*\'()\[\]%-]*', Keyword.Type),
# an anchor
(r'&[\w-]+', Name.Label),
# an alias
(r'\*[\w-]+', Name.Variable),
],
# block collections and scalars
'block-nodes': [
# implicit key
(r':(?=[ ]|$)', set_indent(Punctuation.Indicator, implicit=True)),
# literal and folded scalars
(r'[|>]', Punctuation.Indicator,
('block-scalar-content', 'block-scalar-header')),
],
# flow collections and quoted scalars
'flow-nodes': [
# a flow sequence
(r'\[', Punctuation.Indicator, 'flow-sequence'),
# a flow mapping
(r'\{', Punctuation.Indicator, 'flow-mapping'),
# a single-quoted scalar
(r'\'', String, 'single-quoted-scalar'),
# a double-quoted scalar
(r'\"', String, 'double-quoted-scalar'),
],
# the content of a flow collection
'flow-collection': [
# whitespaces
(r'[ ]+', Whitespace),
# line breaks
(r'\n+', Whitespace),
# a comment
(r'#[^\n]*', Comment.Single),
# simple indicators
(r'[?:,]', Punctuation.Indicator),
# tags, anchors and aliases
include('descriptors'),
# nested collections and quoted scalars
include('flow-nodes'),
# a plain scalar
(r'(?=[^\s?:,\[\]{}#&*!|>\'"%@`])',
something(Name.Variable),
'plain-scalar-in-flow-context'),
],
# a flow sequence indicated by '[' and ']'
'flow-sequence': [
# include flow collection rules
include('flow-collection'),
# the closing indicator
(r'\]', Punctuation.Indicator, '#pop'),
],
# a flow mapping indicated by '{' and '}'
'flow-mapping': [
# key with colon
(r'''([^,:?\[\]{}"'\n]+)(:)(?=[ ]|$)''',
bygroups(Name.Tag, Punctuation)),
# include flow collection rules
include('flow-collection'),
# the closing indicator
(r'\}', Punctuation.Indicator, '#pop'),
],
# block scalar lines
'block-scalar-content': [
# line break
(r'\n', Whitespace),
# empty line
(r'^[ ]+$',
parse_block_scalar_empty_line(Whitespace, Name.Constant)),
# indentation spaces (we may leave the state here)
(r'^[ ]*', parse_block_scalar_indent(Whitespace)),
# line content
(r'[\S\t ]+', Name.Constant),
],
# the content of a literal or folded scalar
'block-scalar-header': [
# indentation indicator followed by chomping flag
(r'([1-9])?[+-]?(?=[ ]|$)',
set_block_scalar_indent(Punctuation.Indicator),
'ignored-line'),
# chomping flag followed by indentation indicator
(r'[+-]?([1-9])?(?=[ ]|$)',
set_block_scalar_indent(Punctuation.Indicator),
'ignored-line'),
],
# ignored and regular whitespaces in quoted scalars
'quoted-scalar-whitespaces': [
# leading and trailing whitespaces are ignored
(r'^[ ]+', Whitespace),
(r'[ ]+$', Whitespace),
# line breaks are ignored
(r'\n+', Whitespace),
# other whitespaces are a part of the value
(r'[ ]+', Name.Variable),
],
# single-quoted scalars
'single-quoted-scalar': [
# include whitespace and line break rules
include('quoted-scalar-whitespaces'),
# escaping of the quote character
(r'\'\'', String.Escape),
# regular non-whitespace characters
(r'[^\s\']+', String),
# the closing quote
(r'\'', String, '#pop'),
],
# double-quoted scalars
'double-quoted-scalar': [
# include whitespace and line break rules
include('quoted-scalar-whitespaces'),
# escaping of special characters
(r'\\[0abt\tn\nvfre "\\N_LP]', String),
# escape codes
(r'\\(?:x[0-9A-Fa-f]{2}|u[0-9A-Fa-f]{4}|U[0-9A-Fa-f]{8})',
String.Escape),
# regular non-whitespace characters
(r'[^\s"\\]+', String),
# the closing quote
(r'"', String, '#pop'),
],
# the beginning of a new line while scanning a plain scalar
'plain-scalar-in-block-context-new-line': [
# empty lines
(r'^[ ]+$', Whitespace),
# line breaks
(r'\n+', Whitespace),
# document start and document end indicators
(r'^(?=---|\.\.\.)', something(Name.Namespace), '#pop:3'),
# indentation spaces (we may leave the block line state here)
(r'^[ ]*', parse_plain_scalar_indent(Whitespace), '#pop'),
],
# a plain scalar in the block context
'plain-scalar-in-block-context': [
# the scalar ends with the ':' indicator
(r'[ ]*(?=:[ ]|:$)', something(Whitespace), '#pop'),
# the scalar ends with whitespaces followed by a comment
(r'[ ]+(?=#)', Whitespace, '#pop'),
# trailing whitespaces are ignored
(r'[ ]+$', Whitespace),
# line breaks are ignored
(r'\n+', Whitespace, 'plain-scalar-in-block-context-new-line'),
# other whitespaces are a part of the value
(r'[ ]+', Literal.Scalar.Plain),
# regular non-whitespace characters
(r'(?::(?!\s)|[^\s:])+', Literal.Scalar.Plain),
],
# a plain scalar is the flow context
'plain-scalar-in-flow-context': [
# the scalar ends with an indicator character
(r'[ ]*(?=[,:?\[\]{}])', something(Whitespace), '#pop'),
# the scalar ends with a comment
(r'[ ]+(?=#)', Whitespace, '#pop'),
# leading and trailing whitespaces are ignored
(r'^[ ]+', Whitespace),
(r'[ ]+$', Whitespace),
# line breaks are ignored
(r'\n+', Whitespace),
# other whitespaces are a part of the value
(r'[ ]+', Name.Variable),
# regular non-whitespace characters
(r'[^\s,:?\[\]{}]+', Name.Variable),
],
}
def get_tokens_unprocessed(self, text=None, context=None):
if context is None:
context = YamlLexerContext(text, 0)
return super().get_tokens_unprocessed(text, context)
class JsonLexer(Lexer):
"""
For JSON data structures.
Javascript-style comments are supported (like ``/* */`` and ``//``),
though comments are not part of the JSON specification.
This allows users to highlight JSON as it is used in the wild.
No validation is performed on the input JSON document.
"""
name = 'JSON'
url = 'https://www.json.org'
aliases = ['json', 'json-object']
filenames = ['*.json', '*.jsonl', '*.ndjson', 'Pipfile.lock']
mimetypes = ['application/json', 'application/json-object', 'application/x-ndjson', 'application/jsonl', 'application/json-seq']
version_added = '1.5'
# No validation of integers, floats, or constants is done.
# As long as the characters are members of the following
# sets, the token will be considered valid. For example,
#
# "--1--" is parsed as an integer
# "1...eee" is parsed as a float
# "trustful" is parsed as a constant
#
integers = set('-0123456789')
floats = set('.eE+')
constants = set('truefalsenull') # true|false|null
hexadecimals = set('0123456789abcdefABCDEF')
punctuations = set('{}[],')
whitespaces = {'\u0020', '\u000a', '\u000d', '\u0009'}
def get_tokens_unprocessed(self, text):
"""Parse JSON data."""
in_string = False
in_escape = False
in_unicode_escape = 0
in_whitespace = False
in_constant = False
in_number = False
in_float = False
in_punctuation = False
in_comment_single = False
in_comment_multiline = False
expecting_second_comment_opener = False # // or /*
expecting_second_comment_closer = False # */
start = 0
# The queue is used to store data that may need to be tokenized
# differently based on what follows. In particular, JSON object
# keys are tokenized differently than string values, but cannot
# be distinguished until punctuation is encountered outside the
# string.
#
# A ":" character after the string indicates that the string is
# an object key; any other character indicates the string is a
# regular string value.
#
# The queue holds tuples that contain the following data:
#
# (start_index, token_type, text)
#
# By default the token type of text in double quotes is
# String.Double. The token type will be replaced if a colon
# is encountered after the string closes.
#
queue = []
for stop, character in enumerate(text):
if in_string:
if in_unicode_escape:
if character in self.hexadecimals:
in_unicode_escape -= 1
if not in_unicode_escape:
in_escape = False
else:
in_unicode_escape = 0
in_escape = False
elif in_escape:
if character == 'u':
in_unicode_escape = 4
else:
in_escape = False
elif character == '\\':
in_escape = True
elif character == '"':
queue.append((start, String.Double, text[start:stop + 1]))
in_string = False
in_escape = False
in_unicode_escape = 0
continue
elif in_whitespace:
if character in self.whitespaces:
continue
if queue:
queue.append((start, Whitespace, text[start:stop]))
else:
yield start, Whitespace, text[start:stop]
in_whitespace = False
# Fall through so the new character can be evaluated.
elif in_constant:
if character in self.constants:
continue
yield start, Keyword.Constant, text[start:stop]
in_constant = False
# Fall through so the new character can be evaluated.
elif in_number:
if character in self.integers:
continue
elif character in self.floats:
in_float = True
continue
if in_float:
yield start, Number.Float, text[start:stop]
else:
yield start, Number.Integer, text[start:stop]
in_number = False
in_float = False
# Fall through so the new character can be evaluated.
elif in_punctuation:
if character in self.punctuations:
continue
yield start, Punctuation, text[start:stop]
in_punctuation = False
# Fall through so the new character can be evaluated.
elif in_comment_single:
if character != '\n':
continue
if queue:
queue.append((start, Comment.Single, text[start:stop]))
else:
yield start, Comment.Single, text[start:stop]
in_comment_single = False
# Fall through so the new character can be evaluated.
elif in_comment_multiline:
if character == '*':
expecting_second_comment_closer = True
elif expecting_second_comment_closer:
expecting_second_comment_closer = False
if character == '/':
if queue:
queue.append((start, Comment.Multiline, text[start:stop + 1]))
else:
yield start, Comment.Multiline, text[start:stop + 1]
in_comment_multiline = False
continue
elif expecting_second_comment_opener:
expecting_second_comment_opener = False
if character == '/':
in_comment_single = True
continue
elif character == '*':
in_comment_multiline = True
continue
# Exhaust the queue. Accept the existing token types.
yield from queue
queue.clear()
yield start, Error, text[start:stop]
# Fall through so the new character can be evaluated.
start = stop
if character == '"':
in_string = True
elif character in self.whitespaces:
in_whitespace = True
elif character in {'f', 'n', 't'}: # The first letters of true|false|null
# Exhaust the queue. Accept the existing token types.
yield from queue
queue.clear()
in_constant = True
elif character in self.integers:
# Exhaust the queue. Accept the existing token types.
yield from queue
queue.clear()
in_number = True
elif character == ':':
# Yield from the queue. Replace string token types.
for _start, _token, _text in queue:
# There can be only three types of tokens before a ':':
# Whitespace, Comment, or a quoted string.
#
# If it's a quoted string we emit Name.Tag.
# Otherwise, we yield the original token.
#
# In all other cases this would be invalid JSON,
# but this is not a validating JSON lexer, so it's OK.
if _token is String.Double:
yield _start, Name.Tag, _text
else:
yield _start, _token, _text
queue.clear()
in_punctuation = True
elif character in self.punctuations:
# Exhaust the queue. Accept the existing token types.
yield from queue
queue.clear()
in_punctuation = True
elif character == '/':
# This is the beginning of a comment.
expecting_second_comment_opener = True
else:
# Exhaust the queue. Accept the existing token types.
yield from queue
queue.clear()
yield start, Error, character
# Yield any remaining text.
yield from queue
if in_string:
yield start, Error, text[start:]
elif in_float:
yield start, Number.Float, text[start:]
elif in_number:
yield start, Number.Integer, text[start:]
elif in_constant:
yield start, Keyword.Constant, text[start:]
elif in_whitespace:
yield start, Whitespace, text[start:]
elif in_punctuation:
yield start, Punctuation, text[start:]
elif in_comment_single:
yield start, Comment.Single, text[start:]
elif in_comment_multiline:
yield start, Error, text[start:]
elif expecting_second_comment_opener:
yield start, Error, text[start:]
class JsonBareObjectLexer(JsonLexer):
"""
For JSON data structures (with missing object curly braces).
.. deprecated:: 2.8.0
Behaves the same as `JsonLexer` now.
"""
name = 'JSONBareObject'
aliases = []
filenames = []
mimetypes = []
version_added = '2.2'
class JsonLdLexer(JsonLexer):
"""
For JSON-LD linked data.
"""
name = 'JSON-LD'
url = 'https://json-ld.org/'
aliases = ['jsonld', 'json-ld']
filenames = ['*.jsonld']
mimetypes = ['application/ld+json']
version_added = '2.0'
json_ld_keywords = {
f'"@{keyword}"'
for keyword in (
'base',
'container',
'context',
'direction',
'graph',
'id',
'import',
'included',
'index',
'json',
'language',
'list',
'nest',
'none',
'prefix',
'propagate',
'protected',
'reverse',
'set',
'type',
'value',
'version',
'vocab',
)
}
def get_tokens_unprocessed(self, text):
for start, token, value in super().get_tokens_unprocessed(text):
if token is Name.Tag and value in self.json_ld_keywords:
yield start, Name.Decorator, value
else:
yield start, token, value