rana-cli/wakatime/packages/pygments/lexers/bibtex.py
2017-02-13 23:25:51 -08:00

160 lines
4.6 KiB
Python

# -*- coding: utf-8 -*-
"""
pygments.lexers.bibtex
~~~~~~~~~~~~~~~~~~~~~~
Lexers for BibTeX bibliography data and styles
:copyright: Copyright 2006-2017 by the Pygments team, see AUTHORS.
:license: BSD, see LICENSE for details.
"""
import re
from pygments.lexer import RegexLexer, ExtendedRegexLexer, include, default, \
words
from pygments.token import Name, Comment, String, Error, Number, Text, \
Keyword, Punctuation
__all__ = ['BibTeXLexer', 'BSTLexer']
class BibTeXLexer(ExtendedRegexLexer):
"""
A lexer for BibTeX bibliography data format.
.. versionadded:: 2.2
"""
name = 'BibTeX'
aliases = ['bib', 'bibtex']
filenames = ['*.bib']
mimetypes = ["text/x-bibtex"]
flags = re.IGNORECASE
ALLOWED_CHARS = r'@!$&*+\-./:;<>?\[\\\]^`|~'
IDENTIFIER = '[{0}][{1}]*'.format('a-z_' + ALLOWED_CHARS, r'\w' + ALLOWED_CHARS)
def open_brace_callback(self, match, ctx):
opening_brace = match.group()
ctx.opening_brace = opening_brace
yield match.start(), Punctuation, opening_brace
ctx.pos = match.end()
def close_brace_callback(self, match, ctx):
closing_brace = match.group()
if (
ctx.opening_brace == '{' and closing_brace != '}' or
ctx.opening_brace == '(' and closing_brace != ')'
):
yield match.start(), Error, closing_brace
else:
yield match.start(), Punctuation, closing_brace
del ctx.opening_brace
ctx.pos = match.end()
tokens = {
'root': [
include('whitespace'),
('@comment', Comment),
('@preamble', Name.Class, ('closing-brace', 'value', 'opening-brace')),
('@string', Name.Class, ('closing-brace', 'field', 'opening-brace')),
('@' + IDENTIFIER, Name.Class,
('closing-brace', 'command-body', 'opening-brace')),
('.+', Comment),
],
'opening-brace': [
include('whitespace'),
(r'[{(]', open_brace_callback, '#pop'),
],
'closing-brace': [
include('whitespace'),
(r'[})]', close_brace_callback, '#pop'),
],
'command-body': [
include('whitespace'),
(r'[^\s\,\}]+', Name.Label, ('#pop', 'fields')),
],
'fields': [
include('whitespace'),
(',', Punctuation, 'field'),
default('#pop'),
],
'field': [
include('whitespace'),
(IDENTIFIER, Name.Attribute, ('value', '=')),
default('#pop'),
],
'=': [
include('whitespace'),
('=', Punctuation, '#pop'),
],
'value': [
include('whitespace'),
(IDENTIFIER, Name.Variable),
('"', String, 'quoted-string'),
(r'\{', String, 'braced-string'),
(r'[\d]+', Number),
('#', Punctuation),
default('#pop'),
],
'quoted-string': [
(r'\{', String, 'braced-string'),
('"', String, '#pop'),
('[^\{\"]+', String),
],
'braced-string': [
(r'\{', String, '#push'),
(r'\}', String, '#pop'),
('[^\{\}]+', String),
],
'whitespace': [
(r'\s+', Text),
],
}
class BSTLexer(RegexLexer):
"""
A lexer for BibTeX bibliography styles.
.. versionadded:: 2.2
"""
name = 'BST'
aliases = ['bst', 'bst-pybtex']
filenames = ['*.bst']
flags = re.IGNORECASE | re.MULTILINE
tokens = {
'root': [
include('whitespace'),
(words(['read', 'sort']), Keyword),
(words(['execute', 'integers', 'iterate', 'reverse', 'strings']),
Keyword, ('group')),
(words(['function', 'macro']), Keyword, ('group', 'group')),
(words(['entry']), Keyword, ('group', 'group', 'group')),
],
'group': [
include('whitespace'),
(r'\{', Punctuation, ('#pop', 'group-end', 'body')),
],
'group-end': [
include('whitespace'),
(r'\}', Punctuation, '#pop'),
],
'body': [
include('whitespace'),
(r"\'[^#\"\{\}\s]+", Name.Function),
(r'[^#\"\{\}\s]+\$', Name.Builtin),
(r'[^#\"\{\}\s]+', Name.Variable),
(r'"[^\"]*"', String),
(r'#-?\d+', Number),
(r'\{', Punctuation, ('group-end', 'body')),
default('#pop'),
],
'whitespace': [
('\s+', Text),
('%.*?$', Comment.SingleLine),
],
}