rana-cli/wakatime/packages/pygments2/pygments/lexers/templates.py

1743 lines
55 KiB
Python
Raw Normal View History

# -*- coding: utf-8 -*-
"""
pygments.lexers.templates
~~~~~~~~~~~~~~~~~~~~~~~~~
Lexers for various template engines' markup.
:copyright: Copyright 2006-2013 by the Pygments team, see AUTHORS.
:license: BSD, see LICENSE for details.
"""
import re
from pygments.lexers.web import \
PhpLexer, HtmlLexer, XmlLexer, JavascriptLexer, CssLexer, LassoLexer
from pygments.lexers.agile import PythonLexer, PerlLexer
from pygments.lexers.compiled import JavaLexer
from pygments.lexers.jvm import TeaLangLexer
from pygments.lexer import Lexer, DelegatingLexer, RegexLexer, bygroups, \
include, using, this
from pygments.token import Error, Punctuation, \
Text, Comment, Operator, Keyword, Name, String, Number, Other, Token
from pygments.util import html_doctype_matches, looks_like_xml
__all__ = ['HtmlPhpLexer', 'XmlPhpLexer', 'CssPhpLexer',
'JavascriptPhpLexer', 'ErbLexer', 'RhtmlLexer',
'XmlErbLexer', 'CssErbLexer', 'JavascriptErbLexer',
'SmartyLexer', 'HtmlSmartyLexer', 'XmlSmartyLexer',
'CssSmartyLexer', 'JavascriptSmartyLexer', 'DjangoLexer',
'HtmlDjangoLexer', 'CssDjangoLexer', 'XmlDjangoLexer',
'JavascriptDjangoLexer', 'GenshiLexer', 'HtmlGenshiLexer',
'GenshiTextLexer', 'CssGenshiLexer', 'JavascriptGenshiLexer',
'MyghtyLexer', 'MyghtyHtmlLexer', 'MyghtyXmlLexer',
'MyghtyCssLexer', 'MyghtyJavascriptLexer', 'MasonLexer', 'MakoLexer',
'MakoHtmlLexer', 'MakoXmlLexer', 'MakoJavascriptLexer',
'MakoCssLexer', 'JspLexer', 'CheetahLexer', 'CheetahHtmlLexer',
'CheetahXmlLexer', 'CheetahJavascriptLexer', 'EvoqueLexer',
'EvoqueHtmlLexer', 'EvoqueXmlLexer', 'ColdfusionLexer',
'ColdfusionHtmlLexer', 'VelocityLexer', 'VelocityHtmlLexer',
'VelocityXmlLexer', 'SspLexer', 'TeaTemplateLexer', 'LassoHtmlLexer',
'LassoXmlLexer', 'LassoCssLexer', 'LassoJavascriptLexer']
class ErbLexer(Lexer):
"""
Generic `ERB <http://ruby-doc.org/core/classes/ERB.html>`_ (Ruby Templating)
lexer.
Just highlights ruby code between the preprocessor directives, other data
is left untouched by the lexer.
All options are also forwarded to the `RubyLexer`.
"""
name = 'ERB'
aliases = ['erb']
mimetypes = ['application/x-ruby-templating']
_block_re = re.compile(r'(<%%|%%>|<%=|<%#|<%-|<%|-%>|%>|^%[^%].*?$)', re.M)
def __init__(self, **options):
from pygments.lexers.agile import RubyLexer
self.ruby_lexer = RubyLexer(**options)
Lexer.__init__(self, **options)
def get_tokens_unprocessed(self, text):
"""
Since ERB doesn't allow "<%" and other tags inside of ruby
blocks we have to use a split approach here that fails for
that too.
"""
tokens = self._block_re.split(text)
tokens.reverse()
state = idx = 0
try:
while True:
# text
if state == 0:
val = tokens.pop()
yield idx, Other, val
idx += len(val)
state = 1
# block starts
elif state == 1:
tag = tokens.pop()
# literals
if tag in ('<%%', '%%>'):
yield idx, Other, tag
idx += 3
state = 0
# comment
elif tag == '<%#':
yield idx, Comment.Preproc, tag
val = tokens.pop()
yield idx + 3, Comment, val
idx += 3 + len(val)
state = 2
# blocks or output
elif tag in ('<%', '<%=', '<%-'):
yield idx, Comment.Preproc, tag
idx += len(tag)
data = tokens.pop()
r_idx = 0
for r_idx, r_token, r_value in \
self.ruby_lexer.get_tokens_unprocessed(data):
yield r_idx + idx, r_token, r_value
idx += len(data)
state = 2
elif tag in ('%>', '-%>'):
yield idx, Error, tag
idx += len(tag)
state = 0
# % raw ruby statements
else:
yield idx, Comment.Preproc, tag[0]
r_idx = 0
for r_idx, r_token, r_value in \
self.ruby_lexer.get_tokens_unprocessed(tag[1:]):
yield idx + 1 + r_idx, r_token, r_value
idx += len(tag)
state = 0
# block ends
elif state == 2:
tag = tokens.pop()
if tag not in ('%>', '-%>'):
yield idx, Other, tag
else:
yield idx, Comment.Preproc, tag
idx += len(tag)
state = 0
except IndexError:
return
def analyse_text(text):
if '<%' in text and '%>' in text:
return 0.4
class SmartyLexer(RegexLexer):
"""
Generic `Smarty <http://smarty.php.net/>`_ template lexer.
Just highlights smarty code between the preprocessor directives, other
data is left untouched by the lexer.
"""
name = 'Smarty'
aliases = ['smarty']
filenames = ['*.tpl']
mimetypes = ['application/x-smarty']
flags = re.MULTILINE | re.DOTALL
tokens = {
'root': [
(r'[^{]+', Other),
(r'(\{)(\*.*?\*)(\})',
bygroups(Comment.Preproc, Comment, Comment.Preproc)),
(r'(\{php\})(.*?)(\{/php\})',
bygroups(Comment.Preproc, using(PhpLexer, startinline=True),
Comment.Preproc)),
(r'(\{)(/?[a-zA-Z_][a-zA-Z0-9_]*)(\s*)',
bygroups(Comment.Preproc, Name.Function, Text), 'smarty'),
(r'\{', Comment.Preproc, 'smarty')
],
'smarty': [
(r'\s+', Text),
(r'\}', Comment.Preproc, '#pop'),
(r'#[a-zA-Z_][a-zA-Z0-9_]*#', Name.Variable),
(r'\$[a-zA-Z_][a-zA-Z0-9_]*(\.[a-zA-Z0-9_]+)*', Name.Variable),
(r'[~!%^&*()+=|\[\]:;,.<>/?{}@-]', Operator),
(r'(true|false|null)\b', Keyword.Constant),
(r"[0-9](\.[0-9]*)?(eE[+-][0-9])?[flFLdD]?|"
r"0[xX][0-9a-fA-F]+[Ll]?", Number),
(r'"(\\\\|\\"|[^"])*"', String.Double),
(r"'(\\\\|\\'|[^'])*'", String.Single),
(r'[a-zA-Z_][a-zA-Z0-9_]*', Name.Attribute)
]
}
def analyse_text(text):
rv = 0.0
if re.search('\{if\s+.*?\}.*?\{/if\}', text):
rv += 0.15
if re.search('\{include\s+file=.*?\}', text):
rv += 0.15
if re.search('\{foreach\s+.*?\}.*?\{/foreach\}', text):
rv += 0.15
if re.search('\{\$.*?\}', text):
rv += 0.01
return rv
class VelocityLexer(RegexLexer):
"""
Generic `Velocity <http://velocity.apache.org/>`_ template lexer.
Just highlights velocity directives and variable references, other
data is left untouched by the lexer.
"""
name = 'Velocity'
aliases = ['velocity']
filenames = ['*.vm','*.fhtml']
flags = re.MULTILINE | re.DOTALL
identifier = r'[a-zA-Z_][a-zA-Z0-9_]*'
tokens = {
'root': [
(r'[^{#$]+', Other),
(r'(#)(\*.*?\*)(#)',
bygroups(Comment.Preproc, Comment, Comment.Preproc)),
(r'(##)(.*?$)',
bygroups(Comment.Preproc, Comment)),
(r'(#\{?)(' + identifier + r')(\}?)(\s?\()',
bygroups(Comment.Preproc, Name.Function, Comment.Preproc, Punctuation),
'directiveparams'),
(r'(#\{?)(' + identifier + r')(\}|\b)',
bygroups(Comment.Preproc, Name.Function, Comment.Preproc)),
(r'\$\{?', Punctuation, 'variable')
],
'variable': [
(identifier, Name.Variable),
(r'\(', Punctuation, 'funcparams'),
(r'(\.)(' + identifier + r')',
bygroups(Punctuation, Name.Variable), '#push'),
(r'\}', Punctuation, '#pop'),
(r'', Other, '#pop')
],
'directiveparams': [
(r'(&&|\|\||==?|!=?|[-<>+*%&\|\^/])|\b(eq|ne|gt|lt|ge|le|not|in)\b',
Operator),
(r'\[', Operator, 'rangeoperator'),
(r'\b' + identifier + r'\b', Name.Function),
include('funcparams')
],
'rangeoperator': [
(r'\.\.', Operator),
include('funcparams'),
(r'\]', Operator, '#pop')
],
'funcparams': [
(r'\$\{?', Punctuation, 'variable'),
(r'\s+', Text),
(r',', Punctuation),
(r'"(\\\\|\\"|[^"])*"', String.Double),
(r"'(\\\\|\\'|[^'])*'", String.Single),
(r"0[xX][0-9a-fA-F]+[Ll]?", Number),
(r"\b[0-9]+\b", Number),
(r'(true|false|null)\b', Keyword.Constant),
(r'\(', Punctuation, '#push'),
(r'\)', Punctuation, '#pop')
]
}
def analyse_text(text):
rv = 0.0
if re.search(r'#\{?macro\}?\(.*?\).*?#\{?end\}?', text):
rv += 0.25
if re.search(r'#\{?if\}?\(.+?\).*?#\{?end\}?', text):
rv += 0.15
if re.search(r'#\{?foreach\}?\(.+?\).*?#\{?end\}?', text):
rv += 0.15
if re.search(r'\$\{?[a-zA-Z_][a-zA-Z0-9_]*(\([^)]*\))?'
r'(\.[a-zA-Z0-9_]+(\([^)]*\))?)*\}?', text):
rv += 0.01
return rv
class VelocityHtmlLexer(DelegatingLexer):
"""
Subclass of the `VelocityLexer` that highlights unlexer data
with the `HtmlLexer`.
"""
name = 'HTML+Velocity'
aliases = ['html+velocity']
alias_filenames = ['*.html','*.fhtml']
mimetypes = ['text/html+velocity']
def __init__(self, **options):
super(VelocityHtmlLexer, self).__init__(HtmlLexer, VelocityLexer,
**options)
class VelocityXmlLexer(DelegatingLexer):
"""
Subclass of the `VelocityLexer` that highlights unlexer data
with the `XmlLexer`.
"""
name = 'XML+Velocity'
aliases = ['xml+velocity']
alias_filenames = ['*.xml','*.vm']
mimetypes = ['application/xml+velocity']
def __init__(self, **options):
super(VelocityXmlLexer, self).__init__(XmlLexer, VelocityLexer,
**options)
def analyse_text(text):
rv = VelocityLexer.analyse_text(text) - 0.01
if looks_like_xml(text):
rv += 0.5
return rv
class DjangoLexer(RegexLexer):
"""
Generic `django <http://www.djangoproject.com/documentation/templates/>`_
and `jinja <http://wsgiarea.pocoo.org/jinja/>`_ template lexer.
It just highlights django/jinja code between the preprocessor directives,
other data is left untouched by the lexer.
"""
name = 'Django/Jinja'
aliases = ['django', 'jinja']
mimetypes = ['application/x-django-templating', 'application/x-jinja']
flags = re.M | re.S
tokens = {
'root': [
(r'[^{]+', Other),
(r'\{\{', Comment.Preproc, 'var'),
# jinja/django comments
(r'\{[*#].*?[*#]\}', Comment),
# django comments
(r'(\{%)(-?\s*)(comment)(\s*-?)(%\})(.*?)'
r'(\{%)(-?\s*)(endcomment)(\s*-?)(%\})',
bygroups(Comment.Preproc, Text, Keyword, Text, Comment.Preproc,
Comment, Comment.Preproc, Text, Keyword, Text,
Comment.Preproc)),
# raw jinja blocks
(r'(\{%)(-?\s*)(raw)(\s*-?)(%\})(.*?)'
r'(\{%)(-?\s*)(endraw)(\s*-?)(%\})',
bygroups(Comment.Preproc, Text, Keyword, Text, Comment.Preproc,
Text, Comment.Preproc, Text, Keyword, Text,
Comment.Preproc)),
# filter blocks
(r'(\{%)(-?\s*)(filter)(\s+)([a-zA-Z_][a-zA-Z0-9_]*)',
bygroups(Comment.Preproc, Text, Keyword, Text, Name.Function),
'block'),
(r'(\{%)(-?\s*)([a-zA-Z_][a-zA-Z0-9_]*)',
bygroups(Comment.Preproc, Text, Keyword), 'block'),
(r'\{', Other)
],
'varnames': [
(r'(\|)(\s*)([a-zA-Z_][a-zA-Z0-9_]*)',
bygroups(Operator, Text, Name.Function)),
(r'(is)(\s+)(not)?(\s+)?([a-zA-Z_][a-zA-Z0-9_]*)',
bygroups(Keyword, Text, Keyword, Text, Name.Function)),
(r'(_|true|false|none|True|False|None)\b', Keyword.Pseudo),
(r'(in|as|reversed|recursive|not|and|or|is|if|else|import|'
r'with(?:(?:out)?\s*context)?|scoped|ignore\s+missing)\b',
Keyword),
(r'(loop|block|super|forloop)\b', Name.Builtin),
(r'[a-zA-Z][a-zA-Z0-9_-]*', Name.Variable),
(r'\.[a-zA-Z0-9_]+', Name.Variable),
(r':?"(\\\\|\\"|[^"])*"', String.Double),
(r":?'(\\\\|\\'|[^'])*'", String.Single),
(r'([{}()\[\]+\-*/,:~]|[><=]=?)', Operator),
(r"[0-9](\.[0-9]*)?(eE[+-][0-9])?[flFLdD]?|"
r"0[xX][0-9a-fA-F]+[Ll]?", Number),
],
'var': [
(r'\s+', Text),
(r'(-?)(\}\})', bygroups(Text, Comment.Preproc), '#pop'),
include('varnames')
],
'block': [
(r'\s+', Text),
(r'(-?)(%\})', bygroups(Text, Comment.Preproc), '#pop'),
include('varnames'),
(r'.', Punctuation)
]
}
def analyse_text(text):
rv = 0.0
if re.search(r'\{%\s*(block|extends)', text) is not None:
rv += 0.4
if re.search(r'\{%\s*if\s*.*?%\}', text) is not None:
rv += 0.1
if re.search(r'\{\{.*?\}\}', text) is not None:
rv += 0.1
return rv
class MyghtyLexer(RegexLexer):
"""
Generic `myghty templates`_ lexer. Code that isn't Myghty
markup is yielded as `Token.Other`.
*New in Pygments 0.6.*
.. _myghty templates: http://www.myghty.org/
"""
name = 'Myghty'
aliases = ['myghty']
filenames = ['*.myt', 'autodelegate']
mimetypes = ['application/x-myghty']
tokens = {
'root': [
(r'\s+', Text),
(r'(<%(?:def|method))(\s*)(.*?)(>)(.*?)(</%\2\s*>)(?s)',
bygroups(Name.Tag, Text, Name.Function, Name.Tag,
using(this), Name.Tag)),
(r'(<%\w+)(.*?)(>)(.*?)(</%\2\s*>)(?s)',
bygroups(Name.Tag, Name.Function, Name.Tag,
using(PythonLexer), Name.Tag)),
(r'(<&[^|])(.*?)(,.*?)?(&>)',
bygroups(Name.Tag, Name.Function, using(PythonLexer), Name.Tag)),
(r'(<&\|)(.*?)(,.*?)?(&>)(?s)',
bygroups(Name.Tag, Name.Function, using(PythonLexer), Name.Tag)),
(r'</&>', Name.Tag),
(r'(<%!?)(.*?)(%>)(?s)',
bygroups(Name.Tag, using(PythonLexer), Name.Tag)),
(r'(?<=^)#[^\n]*(\n|\Z)', Comment),
(r'(?<=^)(%)([^\n]*)(\n|\Z)',
bygroups(Name.Tag, using(PythonLexer), Other)),
(r"""(?sx)
(.+?) # anything, followed by:
(?:
(?<=\n)(?=[%#]) | # an eval or comment line
(?=</?[%&]) | # a substitution or block or
# call start or end
# - don't consume
(\\\n) | # an escaped newline
\Z # end of string
)""", bygroups(Other, Operator)),
]
}
class MyghtyHtmlLexer(DelegatingLexer):
"""
Subclass of the `MyghtyLexer` that highlights unlexer data
with the `HtmlLexer`.
*New in Pygments 0.6.*
"""
name = 'HTML+Myghty'
aliases = ['html+myghty']
mimetypes = ['text/html+myghty']
def __init__(self, **options):
super(MyghtyHtmlLexer, self).__init__(HtmlLexer, MyghtyLexer,
**options)
class MyghtyXmlLexer(DelegatingLexer):
"""
Subclass of the `MyghtyLexer` that highlights unlexer data
with the `XmlLexer`.
*New in Pygments 0.6.*
"""
name = 'XML+Myghty'
aliases = ['xml+myghty']
mimetypes = ['application/xml+myghty']
def __init__(self, **options):
super(MyghtyXmlLexer, self).__init__(XmlLexer, MyghtyLexer,
**options)
class MyghtyJavascriptLexer(DelegatingLexer):
"""
Subclass of the `MyghtyLexer` that highlights unlexer data
with the `JavascriptLexer`.
*New in Pygments 0.6.*
"""
name = 'JavaScript+Myghty'
aliases = ['js+myghty', 'javascript+myghty']
mimetypes = ['application/x-javascript+myghty',
'text/x-javascript+myghty',
'text/javascript+mygthy']
def __init__(self, **options):
super(MyghtyJavascriptLexer, self).__init__(JavascriptLexer,
MyghtyLexer, **options)
class MyghtyCssLexer(DelegatingLexer):
"""
Subclass of the `MyghtyLexer` that highlights unlexer data
with the `CssLexer`.
*New in Pygments 0.6.*
"""
name = 'CSS+Myghty'
aliases = ['css+myghty']
mimetypes = ['text/css+myghty']
def __init__(self, **options):
super(MyghtyCssLexer, self).__init__(CssLexer, MyghtyLexer,
**options)
class MasonLexer(RegexLexer):
"""
Generic `mason templates`_ lexer. Stolen from Myghty lexer. Code that isn't
Mason markup is HTML.
.. _mason templates: http://www.masonhq.com/
*New in Pygments 1.4.*
"""
name = 'Mason'
aliases = ['mason']
filenames = ['*.m', '*.mhtml', '*.mc', '*.mi', 'autohandler', 'dhandler']
mimetypes = ['application/x-mason']
tokens = {
'root': [
(r'\s+', Text),
(r'(<%doc>)(.*?)(</%doc>)(?s)',
bygroups(Name.Tag, Comment.Multiline, Name.Tag)),
(r'(<%(?:def|method))(\s*)(.*?)(>)(.*?)(</%\2\s*>)(?s)',
bygroups(Name.Tag, Text, Name.Function, Name.Tag,
using(this), Name.Tag)),
(r'(<%\w+)(.*?)(>)(.*?)(</%\2\s*>)(?s)',
bygroups(Name.Tag, Name.Function, Name.Tag,
using(PerlLexer), Name.Tag)),
(r'(<&[^|])(.*?)(,.*?)?(&>)(?s)',
bygroups(Name.Tag, Name.Function, using(PerlLexer), Name.Tag)),
(r'(<&\|)(.*?)(,.*?)?(&>)(?s)',
bygroups(Name.Tag, Name.Function, using(PerlLexer), Name.Tag)),
(r'</&>', Name.Tag),
(r'(<%!?)(.*?)(%>)(?s)',
bygroups(Name.Tag, using(PerlLexer), Name.Tag)),
(r'(?<=^)#[^\n]*(\n|\Z)', Comment),
(r'(?<=^)(%)([^\n]*)(\n|\Z)',
bygroups(Name.Tag, using(PerlLexer), Other)),
(r"""(?sx)
(.+?) # anything, followed by:
(?:
(?<=\n)(?=[%#]) | # an eval or comment line
(?=</?[%&]) | # a substitution or block or
# call start or end
# - don't consume
(\\\n) | # an escaped newline
\Z # end of string
)""", bygroups(using(HtmlLexer), Operator)),
]
}
def analyse_text(text):
rv = 0.0
if re.search('<&', text) is not None:
rv = 1.0
return rv
class MakoLexer(RegexLexer):
"""
Generic `mako templates`_ lexer. Code that isn't Mako
markup is yielded as `Token.Other`.
*New in Pygments 0.7.*
.. _mako templates: http://www.makotemplates.org/
"""
name = 'Mako'
aliases = ['mako']
filenames = ['*.mao']
mimetypes = ['application/x-mako']
tokens = {
'root': [
(r'(\s*)(%)(\s*end(?:\w+))(\n|\Z)',
bygroups(Text, Comment.Preproc, Keyword, Other)),
(r'(\s*)(%)([^\n]*)(\n|\Z)',
bygroups(Text, Comment.Preproc, using(PythonLexer), Other)),
(r'(\s*)(##[^\n]*)(\n|\Z)',
bygroups(Text, Comment.Preproc, Other)),
(r'(?s)<%doc>.*?</%doc>', Comment.Preproc),
(r'(<%)([\w\.\:]+)',
bygroups(Comment.Preproc, Name.Builtin), 'tag'),
(r'(</%)([\w\.\:]+)(>)',
bygroups(Comment.Preproc, Name.Builtin, Comment.Preproc)),
(r'<%(?=([\w\.\:]+))', Comment.Preproc, 'ondeftags'),
(r'(<%(?:!?))(.*?)(%>)(?s)',
bygroups(Comment.Preproc, using(PythonLexer), Comment.Preproc)),
(r'(\$\{)(.*?)(\})',
bygroups(Comment.Preproc, using(PythonLexer), Comment.Preproc)),
(r'''(?sx)
(.+?) # anything, followed by:
(?:
(?<=\n)(?=%|\#\#) | # an eval or comment line
(?=\#\*) | # multiline comment
(?=</?%) | # a python block
# call start or end
(?=\$\{) | # a substitution
(?<=\n)(?=\s*%) |
# - don't consume
(\\\n) | # an escaped newline
\Z # end of string
)
''', bygroups(Other, Operator)),
(r'\s+', Text),
],
'ondeftags': [
(r'<%', Comment.Preproc),
(r'(?<=<%)(include|inherit|namespace|page)', Name.Builtin),
include('tag'),
],
'tag': [
(r'((?:\w+)\s*=)(\s*)(".*?")',
bygroups(Name.Attribute, Text, String)),
(r'/?\s*>', Comment.Preproc, '#pop'),
(r'\s+', Text),
],
'attr': [
('".*?"', String, '#pop'),
("'.*?'", String, '#pop'),
(r'[^\s>]+', String, '#pop'),
],
}
class MakoHtmlLexer(DelegatingLexer):
"""
Subclass of the `MakoLexer` that highlights unlexed data
with the `HtmlLexer`.
*New in Pygments 0.7.*
"""
name = 'HTML+Mako'
aliases = ['html+mako']
mimetypes = ['text/html+mako']
def __init__(self, **options):
super(MakoHtmlLexer, self).__init__(HtmlLexer, MakoLexer,
**options)
class MakoXmlLexer(DelegatingLexer):
"""
Subclass of the `MakoLexer` that highlights unlexer data
with the `XmlLexer`.
*New in Pygments 0.7.*
"""
name = 'XML+Mako'
aliases = ['xml+mako']
mimetypes = ['application/xml+mako']
def __init__(self, **options):
super(MakoXmlLexer, self).__init__(XmlLexer, MakoLexer,
**options)
class MakoJavascriptLexer(DelegatingLexer):
"""
Subclass of the `MakoLexer` that highlights unlexer data
with the `JavascriptLexer`.
*New in Pygments 0.7.*
"""
name = 'JavaScript+Mako'
aliases = ['js+mako', 'javascript+mako']
mimetypes = ['application/x-javascript+mako',
'text/x-javascript+mako',
'text/javascript+mako']
def __init__(self, **options):
super(MakoJavascriptLexer, self).__init__(JavascriptLexer,
MakoLexer, **options)
class MakoCssLexer(DelegatingLexer):
"""
Subclass of the `MakoLexer` that highlights unlexer data
with the `CssLexer`.
*New in Pygments 0.7.*
"""
name = 'CSS+Mako'
aliases = ['css+mako']
mimetypes = ['text/css+mako']
def __init__(self, **options):
super(MakoCssLexer, self).__init__(CssLexer, MakoLexer,
**options)
# Genshi and Cheetah lexers courtesy of Matt Good.
class CheetahPythonLexer(Lexer):
"""
Lexer for handling Cheetah's special $ tokens in Python syntax.
"""
def get_tokens_unprocessed(self, text):
pylexer = PythonLexer(**self.options)
for pos, type_, value in pylexer.get_tokens_unprocessed(text):
if type_ == Token.Error and value == '$':
type_ = Comment.Preproc
yield pos, type_, value
class CheetahLexer(RegexLexer):
"""
Generic `cheetah templates`_ lexer. Code that isn't Cheetah
markup is yielded as `Token.Other`. This also works for
`spitfire templates`_ which use the same syntax.
.. _cheetah templates: http://www.cheetahtemplate.org/
.. _spitfire templates: http://code.google.com/p/spitfire/
"""
name = 'Cheetah'
aliases = ['cheetah', 'spitfire']
filenames = ['*.tmpl', '*.spt']
mimetypes = ['application/x-cheetah', 'application/x-spitfire']
tokens = {
'root': [
(r'(##[^\n]*)$',
(bygroups(Comment))),
(r'#[*](.|\n)*?[*]#', Comment),
(r'#end[^#\n]*(?:#|$)', Comment.Preproc),
(r'#slurp$', Comment.Preproc),
(r'(#[a-zA-Z]+)([^#\n]*)(#|$)',
(bygroups(Comment.Preproc, using(CheetahPythonLexer),
Comment.Preproc))),
# TODO support other Python syntax like $foo['bar']
(r'(\$)([a-zA-Z_][a-zA-Z0-9_\.]*[a-zA-Z0-9_])',
bygroups(Comment.Preproc, using(CheetahPythonLexer))),
(r'(\$\{!?)(.*?)(\})(?s)',
bygroups(Comment.Preproc, using(CheetahPythonLexer),
Comment.Preproc)),
(r'''(?sx)
(.+?) # anything, followed by:
(?:
(?=[#][#a-zA-Z]*) | # an eval comment
(?=\$[a-zA-Z_{]) | # a substitution
\Z # end of string
)
''', Other),
(r'\s+', Text),
],
}
class CheetahHtmlLexer(DelegatingLexer):
"""
Subclass of the `CheetahLexer` that highlights unlexer data
with the `HtmlLexer`.
"""
name = 'HTML+Cheetah'
aliases = ['html+cheetah', 'html+spitfire', 'htmlcheetah']
mimetypes = ['text/html+cheetah', 'text/html+spitfire']
def __init__(self, **options):
super(CheetahHtmlLexer, self).__init__(HtmlLexer, CheetahLexer,
**options)
class CheetahXmlLexer(DelegatingLexer):
"""
Subclass of the `CheetahLexer` that highlights unlexer data
with the `XmlLexer`.
"""
name = 'XML+Cheetah'
aliases = ['xml+cheetah', 'xml+spitfire']
mimetypes = ['application/xml+cheetah', 'application/xml+spitfire']
def __init__(self, **options):
super(CheetahXmlLexer, self).__init__(XmlLexer, CheetahLexer,
**options)
class CheetahJavascriptLexer(DelegatingLexer):
"""
Subclass of the `CheetahLexer` that highlights unlexer data
with the `JavascriptLexer`.
"""
name = 'JavaScript+Cheetah'
aliases = ['js+cheetah', 'javascript+cheetah',
'js+spitfire', 'javascript+spitfire']
mimetypes = ['application/x-javascript+cheetah',
'text/x-javascript+cheetah',
'text/javascript+cheetah',
'application/x-javascript+spitfire',
'text/x-javascript+spitfire',
'text/javascript+spitfire']
def __init__(self, **options):
super(CheetahJavascriptLexer, self).__init__(JavascriptLexer,
CheetahLexer, **options)
class GenshiTextLexer(RegexLexer):
"""
A lexer that highlights `genshi <http://genshi.edgewall.org/>`_ text
templates.
"""
name = 'Genshi Text'
aliases = ['genshitext']
mimetypes = ['application/x-genshi-text', 'text/x-genshi']
tokens = {
'root': [
(r'[^#\$\s]+', Other),
(r'^(\s*)(##.*)$', bygroups(Text, Comment)),
(r'^(\s*)(#)', bygroups(Text, Comment.Preproc), 'directive'),
include('variable'),
(r'[#\$\s]', Other),
],
'directive': [
(r'\n', Text, '#pop'),
(r'(?:def|for|if)\s+.*', using(PythonLexer), '#pop'),
(r'(choose|when|with)([^\S\n]+)(.*)',
bygroups(Keyword, Text, using(PythonLexer)), '#pop'),
(r'(choose|otherwise)\b', Keyword, '#pop'),
(r'(end\w*)([^\S\n]*)(.*)', bygroups(Keyword, Text, Comment), '#pop'),
],
'variable': [
(r'(?<!\$)(\$\{)(.+?)(\})',
bygroups(Comment.Preproc, using(PythonLexer), Comment.Preproc)),
(r'(?<!\$)(\$)([a-zA-Z_][a-zA-Z0-9_\.]*)',
Name.Variable),
]
}
class GenshiMarkupLexer(RegexLexer):
"""
Base lexer for Genshi markup, used by `HtmlGenshiLexer` and
`GenshiLexer`.
"""
flags = re.DOTALL
tokens = {
'root': [
(r'[^<\$]+', Other),
(r'(<\?python)(.*?)(\?>)',
bygroups(Comment.Preproc, using(PythonLexer), Comment.Preproc)),
# yield style and script blocks as Other
(r'<\s*(script|style)\s*.*?>.*?<\s*/\1\s*>', Other),
(r'<\s*py:[a-zA-Z0-9]+', Name.Tag, 'pytag'),
(r'<\s*[a-zA-Z0-9:]+', Name.Tag, 'tag'),
include('variable'),
(r'[<\$]', Other),
],
'pytag': [
(r'\s+', Text),
(r'[a-zA-Z0-9_:-]+\s*=', Name.Attribute, 'pyattr'),
(r'/?\s*>', Name.Tag, '#pop'),
],
'pyattr': [
('(")(.*?)(")', bygroups(String, using(PythonLexer), String), '#pop'),
("(')(.*?)(')", bygroups(String, using(PythonLexer), String), '#pop'),
(r'[^\s>]+', String, '#pop'),
],
'tag': [
(r'\s+', Text),
(r'py:[a-zA-Z0-9_-]+\s*=', Name.Attribute, 'pyattr'),
(r'[a-zA-Z0-9_:-]+\s*=', Name.Attribute, 'attr'),
(r'/?\s*>', Name.Tag, '#pop'),
],
'attr': [
('"', String, 'attr-dstring'),
("'", String, 'attr-sstring'),
(r'[^\s>]*', String, '#pop')
],
'attr-dstring': [
('"', String, '#pop'),
include('strings'),
("'", String)
],
'attr-sstring': [
("'", String, '#pop'),
include('strings'),
("'", String)
],
'strings': [
('[^"\'$]+', String),
include('variable')
],
'variable': [
(r'(?<!\$)(\$\{)(.+?)(\})',
bygroups(Comment.Preproc, using(PythonLexer), Comment.Preproc)),
(r'(?<!\$)(\$)([a-zA-Z_][a-zA-Z0-9_\.]*)',
Name.Variable),
]
}
class HtmlGenshiLexer(DelegatingLexer):
"""
A lexer that highlights `genshi <http://genshi.edgewall.org/>`_ and
`kid <http://kid-templating.org/>`_ kid HTML templates.
"""
name = 'HTML+Genshi'
aliases = ['html+genshi', 'html+kid']
alias_filenames = ['*.html', '*.htm', '*.xhtml']
mimetypes = ['text/html+genshi']
def __init__(self, **options):
super(HtmlGenshiLexer, self).__init__(HtmlLexer, GenshiMarkupLexer,
**options)
def analyse_text(text):
rv = 0.0
if re.search('\$\{.*?\}', text) is not None:
rv += 0.2
if re.search('py:(.*?)=["\']', text) is not None:
rv += 0.2
return rv + HtmlLexer.analyse_text(text) - 0.01
class GenshiLexer(DelegatingLexer):
"""
A lexer that highlights `genshi <http://genshi.edgewall.org/>`_ and
`kid <http://kid-templating.org/>`_ kid XML templates.
"""
name = 'Genshi'
aliases = ['genshi', 'kid', 'xml+genshi', 'xml+kid']
filenames = ['*.kid']
alias_filenames = ['*.xml']
mimetypes = ['application/x-genshi', 'application/x-kid']
def __init__(self, **options):
super(GenshiLexer, self).__init__(XmlLexer, GenshiMarkupLexer,
**options)
def analyse_text(text):
rv = 0.0
if re.search('\$\{.*?\}', text) is not None:
rv += 0.2
if re.search('py:(.*?)=["\']', text) is not None:
rv += 0.2
return rv + XmlLexer.analyse_text(text) - 0.01
class JavascriptGenshiLexer(DelegatingLexer):
"""
A lexer that highlights javascript code in genshi text templates.
"""
name = 'JavaScript+Genshi Text'
aliases = ['js+genshitext', 'js+genshi', 'javascript+genshitext',
'javascript+genshi']
alias_filenames = ['*.js']
mimetypes = ['application/x-javascript+genshi',
'text/x-javascript+genshi',
'text/javascript+genshi']
def __init__(self, **options):
super(JavascriptGenshiLexer, self).__init__(JavascriptLexer,
GenshiTextLexer,
**options)
def analyse_text(text):
return GenshiLexer.analyse_text(text) - 0.05
class CssGenshiLexer(DelegatingLexer):
"""
A lexer that highlights CSS definitions in genshi text templates.
"""
name = 'CSS+Genshi Text'
aliases = ['css+genshitext', 'css+genshi']
alias_filenames = ['*.css']
mimetypes = ['text/css+genshi']
def __init__(self, **options):
super(CssGenshiLexer, self).__init__(CssLexer, GenshiTextLexer,
**options)
def analyse_text(text):
return GenshiLexer.analyse_text(text) - 0.05
class RhtmlLexer(DelegatingLexer):
"""
Subclass of the ERB lexer that highlights the unlexed data with the
html lexer.
Nested Javascript and CSS is highlighted too.
"""
name = 'RHTML'
aliases = ['rhtml', 'html+erb', 'html+ruby']
filenames = ['*.rhtml']
alias_filenames = ['*.html', '*.htm', '*.xhtml']
mimetypes = ['text/html+ruby']
def __init__(self, **options):
super(RhtmlLexer, self).__init__(HtmlLexer, ErbLexer, **options)
def analyse_text(text):
rv = ErbLexer.analyse_text(text) - 0.01
if html_doctype_matches(text):
# one more than the XmlErbLexer returns
rv += 0.5
return rv
class XmlErbLexer(DelegatingLexer):
"""
Subclass of `ErbLexer` which highlights data outside preprocessor
directives with the `XmlLexer`.
"""
name = 'XML+Ruby'
aliases = ['xml+erb', 'xml+ruby']
alias_filenames = ['*.xml']
mimetypes = ['application/xml+ruby']
def __init__(self, **options):
super(XmlErbLexer, self).__init__(XmlLexer, ErbLexer, **options)
def analyse_text(text):
rv = ErbLexer.analyse_text(text) - 0.01
if looks_like_xml(text):
rv += 0.4
return rv
class CssErbLexer(DelegatingLexer):
"""
Subclass of `ErbLexer` which highlights unlexed data with the `CssLexer`.
"""
name = 'CSS+Ruby'
aliases = ['css+erb', 'css+ruby']
alias_filenames = ['*.css']
mimetypes = ['text/css+ruby']
def __init__(self, **options):
super(CssErbLexer, self).__init__(CssLexer, ErbLexer, **options)
def analyse_text(text):
return ErbLexer.analyse_text(text) - 0.05
class JavascriptErbLexer(DelegatingLexer):
"""
Subclass of `ErbLexer` which highlights unlexed data with the
`JavascriptLexer`.
"""
name = 'JavaScript+Ruby'
aliases = ['js+erb', 'javascript+erb', 'js+ruby', 'javascript+ruby']
alias_filenames = ['*.js']
mimetypes = ['application/x-javascript+ruby',
'text/x-javascript+ruby',
'text/javascript+ruby']
def __init__(self, **options):
super(JavascriptErbLexer, self).__init__(JavascriptLexer, ErbLexer,
**options)
def analyse_text(text):
return ErbLexer.analyse_text(text) - 0.05
class HtmlPhpLexer(DelegatingLexer):
"""
Subclass of `PhpLexer` that highlights unhandled data with the `HtmlLexer`.
Nested Javascript and CSS is highlighted too.
"""
name = 'HTML+PHP'
aliases = ['html+php']
filenames = ['*.phtml']
alias_filenames = ['*.php', '*.html', '*.htm', '*.xhtml',
'*.php[345]']
mimetypes = ['application/x-php',
'application/x-httpd-php', 'application/x-httpd-php3',
'application/x-httpd-php4', 'application/x-httpd-php5']
def __init__(self, **options):
super(HtmlPhpLexer, self).__init__(HtmlLexer, PhpLexer, **options)
def analyse_text(text):
rv = PhpLexer.analyse_text(text) - 0.01
if html_doctype_matches(text):
rv += 0.5
return rv
class XmlPhpLexer(DelegatingLexer):
"""
Subclass of `PhpLexer` that higlights unhandled data with the `XmlLexer`.
"""
name = 'XML+PHP'
aliases = ['xml+php']
alias_filenames = ['*.xml', '*.php', '*.php[345]']
mimetypes = ['application/xml+php']
def __init__(self, **options):
super(XmlPhpLexer, self).__init__(XmlLexer, PhpLexer, **options)
def analyse_text(text):
rv = PhpLexer.analyse_text(text) - 0.01
if looks_like_xml(text):
rv += 0.4
return rv
class CssPhpLexer(DelegatingLexer):
"""
Subclass of `PhpLexer` which highlights unmatched data with the `CssLexer`.
"""
name = 'CSS+PHP'
aliases = ['css+php']
alias_filenames = ['*.css']
mimetypes = ['text/css+php']
def __init__(self, **options):
super(CssPhpLexer, self).__init__(CssLexer, PhpLexer, **options)
def analyse_text(text):
return PhpLexer.analyse_text(text) - 0.05
class JavascriptPhpLexer(DelegatingLexer):
"""
Subclass of `PhpLexer` which highlights unmatched data with the
`JavascriptLexer`.
"""
name = 'JavaScript+PHP'
aliases = ['js+php', 'javascript+php']
alias_filenames = ['*.js']
mimetypes = ['application/x-javascript+php',
'text/x-javascript+php',
'text/javascript+php']
def __init__(self, **options):
super(JavascriptPhpLexer, self).__init__(JavascriptLexer, PhpLexer,
**options)
def analyse_text(text):
return PhpLexer.analyse_text(text)
class HtmlSmartyLexer(DelegatingLexer):
"""
Subclass of the `SmartyLexer` that highighlights unlexed data with the
`HtmlLexer`.
Nested Javascript and CSS is highlighted too.
"""
name = 'HTML+Smarty'
aliases = ['html+smarty']
alias_filenames = ['*.html', '*.htm', '*.xhtml', '*.tpl']
mimetypes = ['text/html+smarty']
def __init__(self, **options):
super(HtmlSmartyLexer, self).__init__(HtmlLexer, SmartyLexer, **options)
def analyse_text(text):
rv = SmartyLexer.analyse_text(text) - 0.01
if html_doctype_matches(text):
rv += 0.5
return rv
class XmlSmartyLexer(DelegatingLexer):
"""
Subclass of the `SmartyLexer` that highlights unlexed data with the
`XmlLexer`.
"""
name = 'XML+Smarty'
aliases = ['xml+smarty']
alias_filenames = ['*.xml', '*.tpl']
mimetypes = ['application/xml+smarty']
def __init__(self, **options):
super(XmlSmartyLexer, self).__init__(XmlLexer, SmartyLexer, **options)
def analyse_text(text):
rv = SmartyLexer.analyse_text(text) - 0.01
if looks_like_xml(text):
rv += 0.4
return rv
class CssSmartyLexer(DelegatingLexer):
"""
Subclass of the `SmartyLexer` that highlights unlexed data with the
`CssLexer`.
"""
name = 'CSS+Smarty'
aliases = ['css+smarty']
alias_filenames = ['*.css', '*.tpl']
mimetypes = ['text/css+smarty']
def __init__(self, **options):
super(CssSmartyLexer, self).__init__(CssLexer, SmartyLexer, **options)
def analyse_text(text):
return SmartyLexer.analyse_text(text) - 0.05
class JavascriptSmartyLexer(DelegatingLexer):
"""
Subclass of the `SmartyLexer` that highlights unlexed data with the
`JavascriptLexer`.
"""
name = 'JavaScript+Smarty'
aliases = ['js+smarty', 'javascript+smarty']
alias_filenames = ['*.js', '*.tpl']
mimetypes = ['application/x-javascript+smarty',
'text/x-javascript+smarty',
'text/javascript+smarty']
def __init__(self, **options):
super(JavascriptSmartyLexer, self).__init__(JavascriptLexer, SmartyLexer,
**options)
def analyse_text(text):
return SmartyLexer.analyse_text(text) - 0.05
class HtmlDjangoLexer(DelegatingLexer):
"""
Subclass of the `DjangoLexer` that highighlights unlexed data with the
`HtmlLexer`.
Nested Javascript and CSS is highlighted too.
"""
name = 'HTML+Django/Jinja'
aliases = ['html+django', 'html+jinja', 'htmldjango']
alias_filenames = ['*.html', '*.htm', '*.xhtml']
mimetypes = ['text/html+django', 'text/html+jinja']
def __init__(self, **options):
super(HtmlDjangoLexer, self).__init__(HtmlLexer, DjangoLexer, **options)
def analyse_text(text):
rv = DjangoLexer.analyse_text(text) - 0.01
if html_doctype_matches(text):
rv += 0.5
return rv
class XmlDjangoLexer(DelegatingLexer):
"""
Subclass of the `DjangoLexer` that highlights unlexed data with the
`XmlLexer`.
"""
name = 'XML+Django/Jinja'
aliases = ['xml+django', 'xml+jinja']
alias_filenames = ['*.xml']
mimetypes = ['application/xml+django', 'application/xml+jinja']
def __init__(self, **options):
super(XmlDjangoLexer, self).__init__(XmlLexer, DjangoLexer, **options)
def analyse_text(text):
rv = DjangoLexer.analyse_text(text) - 0.01
if looks_like_xml(text):
rv += 0.4
return rv
class CssDjangoLexer(DelegatingLexer):
"""
Subclass of the `DjangoLexer` that highlights unlexed data with the
`CssLexer`.
"""
name = 'CSS+Django/Jinja'
aliases = ['css+django', 'css+jinja']
alias_filenames = ['*.css']
mimetypes = ['text/css+django', 'text/css+jinja']
def __init__(self, **options):
super(CssDjangoLexer, self).__init__(CssLexer, DjangoLexer, **options)
def analyse_text(text):
return DjangoLexer.analyse_text(text) - 0.05
class JavascriptDjangoLexer(DelegatingLexer):
"""
Subclass of the `DjangoLexer` that highlights unlexed data with the
`JavascriptLexer`.
"""
name = 'JavaScript+Django/Jinja'
aliases = ['js+django', 'javascript+django',
'js+jinja', 'javascript+jinja']
alias_filenames = ['*.js']
mimetypes = ['application/x-javascript+django',
'application/x-javascript+jinja',
'text/x-javascript+django',
'text/x-javascript+jinja',
'text/javascript+django',
'text/javascript+jinja']
def __init__(self, **options):
super(JavascriptDjangoLexer, self).__init__(JavascriptLexer, DjangoLexer,
**options)
def analyse_text(text):
return DjangoLexer.analyse_text(text) - 0.05
class JspRootLexer(RegexLexer):
"""
Base for the `JspLexer`. Yields `Token.Other` for area outside of
JSP tags.
*New in Pygments 0.7.*
"""
tokens = {
'root': [
(r'<%\S?', Keyword, 'sec'),
# FIXME: I want to make these keywords but still parse attributes.
(r'</?jsp:(forward|getProperty|include|plugin|setProperty|useBean).*?>',
Keyword),
(r'[^<]+', Other),
(r'<', Other),
],
'sec': [
(r'%>', Keyword, '#pop'),
# note: '\w\W' != '.' without DOTALL.
(r'[\w\W]+?(?=%>|\Z)', using(JavaLexer)),
],
}
class JspLexer(DelegatingLexer):
"""
Lexer for Java Server Pages.
*New in Pygments 0.7.*
"""
name = 'Java Server Page'
aliases = ['jsp']
filenames = ['*.jsp']
mimetypes = ['application/x-jsp']
def __init__(self, **options):
super(JspLexer, self).__init__(XmlLexer, JspRootLexer, **options)
def analyse_text(text):
rv = JavaLexer.analyse_text(text) - 0.01
if looks_like_xml(text):
rv += 0.4
if '<%' in text and '%>' in text:
rv += 0.1
return rv
class EvoqueLexer(RegexLexer):
"""
For files using the Evoque templating system.
*New in Pygments 1.1.*
"""
name = 'Evoque'
aliases = ['evoque']
filenames = ['*.evoque']
mimetypes = ['application/x-evoque']
flags = re.DOTALL
tokens = {
'root': [
(r'[^#$]+', Other),
(r'#\[', Comment.Multiline, 'comment'),
(r'\$\$', Other),
# svn keywords
(r'\$\w+:[^$\n]*\$', Comment.Multiline),
# directives: begin, end
(r'(\$)(begin|end)(\{(%)?)(.*?)((?(4)%)\})',
bygroups(Punctuation, Name.Builtin, Punctuation, None,
String, Punctuation)),
# directives: evoque, overlay
# see doc for handling first name arg: /directives/evoque/
#+ minor inconsistency: the "name" in e.g. $overlay{name=site_base}
# should be using(PythonLexer), not passed out as String
(r'(\$)(evoque|overlay)(\{(%)?)(\s*[#\w\-"\'.]+[^=,%}]+?)?'
r'(.*?)((?(4)%)\})',
bygroups(Punctuation, Name.Builtin, Punctuation, None,
String, using(PythonLexer), Punctuation)),
# directives: if, for, prefer, test
(r'(\$)(\w+)(\{(%)?)(.*?)((?(4)%)\})',
bygroups(Punctuation, Name.Builtin, Punctuation, None,
using(PythonLexer), Punctuation)),
# directive clauses (no {} expression)
(r'(\$)(else|rof|fi)', bygroups(Punctuation, Name.Builtin)),
# expressions
(r'(\$\{(%)?)(.*?)((!)(.*?))?((?(2)%)\})',
bygroups(Punctuation, None, using(PythonLexer),
Name.Builtin, None, None, Punctuation)),
(r'#', Other),
],
'comment': [
(r'[^\]#]', Comment.Multiline),
(r'#\[', Comment.Multiline, '#push'),
(r'\]#', Comment.Multiline, '#pop'),
(r'[\]#]', Comment.Multiline)
],
}
class EvoqueHtmlLexer(DelegatingLexer):
"""
Subclass of the `EvoqueLexer` that highlights unlexed data with the
`HtmlLexer`.
*New in Pygments 1.1.*
"""
name = 'HTML+Evoque'
aliases = ['html+evoque']
filenames = ['*.html']
mimetypes = ['text/html+evoque']
def __init__(self, **options):
super(EvoqueHtmlLexer, self).__init__(HtmlLexer, EvoqueLexer,
**options)
class EvoqueXmlLexer(DelegatingLexer):
"""
Subclass of the `EvoqueLexer` that highlights unlexed data with the
`XmlLexer`.
*New in Pygments 1.1.*
"""
name = 'XML+Evoque'
aliases = ['xml+evoque']
filenames = ['*.xml']
mimetypes = ['application/xml+evoque']
def __init__(self, **options):
super(EvoqueXmlLexer, self).__init__(XmlLexer, EvoqueLexer,
**options)
class ColdfusionLexer(RegexLexer):
"""
Coldfusion statements
"""
name = 'cfstatement'
aliases = ['cfs']
filenames = []
mimetypes = []
flags = re.IGNORECASE | re.MULTILINE
tokens = {
'root': [
(r'//.*', Comment),
(r'\+\+|--', Operator),
(r'[-+*/^&=!]', Operator),
(r'<=|>=|<|>', Operator),
(r'mod\b', Operator),
(r'(eq|lt|gt|lte|gte|not|is|and|or)\b', Operator),
(r'\|\||&&', Operator),
(r'"', String.Double, 'string'),
# There is a special rule for allowing html in single quoted
# strings, evidently.
(r"'.*?'", String.Single),
(r'\d+', Number),
(r'(if|else|len|var|case|default|break|switch)\b', Keyword),
(r'([A-Za-z_$][A-Za-z0-9_.]*)(\s*)(\()',
bygroups(Name.Function, Text, Punctuation)),
(r'[A-Za-z_$][A-Za-z0-9_.]*', Name.Variable),
(r'[()\[\]{};:,.\\]', Punctuation),
(r'\s+', Text),
],
'string': [
(r'""', String.Double),
(r'#.+?#', String.Interp),
(r'[^"#]+', String.Double),
(r'#', String.Double),
(r'"', String.Double, '#pop'),
],
}
class ColdfusionMarkupLexer(RegexLexer):
"""
Coldfusion markup only
"""
name = 'Coldfusion'
aliases = ['cf']
filenames = []
mimetypes = []
tokens = {
'root': [
(r'[^<]+', Other),
include('tags'),
(r'<[^<>]*', Other),
],
'tags': [
(r'(?s)<!---.*?--->', Comment.Multiline),
(r'(?s)<!--.*?-->', Comment),
(r'<cfoutput.*?>', Name.Builtin, 'cfoutput'),
(r'(?s)(<cfscript.*?>)(.+?)(</cfscript.*?>)',
bygroups(Name.Builtin, using(ColdfusionLexer), Name.Builtin)),
# negative lookbehind is for strings with embedded >
(r'(?s)(</?cf(?:component|include|if|else|elseif|loop|return|'
r'dbinfo|dump|abort|location|invoke|throw|file|savecontent|'
r'mailpart|mail|header|content|zip|image|lock|argument|try|'
r'catch|break|directory|http|set|function|param)\b)(.*?)((?<!\\)>)',
bygroups(Name.Builtin, using(ColdfusionLexer), Name.Builtin)),
],
'cfoutput': [
(r'[^#<]+', Other),
(r'(#)(.*?)(#)', bygroups(Punctuation, using(ColdfusionLexer),
Punctuation)),
#(r'<cfoutput.*?>', Name.Builtin, '#push'),
(r'</cfoutput.*?>', Name.Builtin, '#pop'),
include('tags'),
(r'(?s)<[^<>]*', Other),
(r'#', Other),
],
}
class ColdfusionHtmlLexer(DelegatingLexer):
"""
Coldfusion markup in html
"""
name = 'Coldfusion HTML'
aliases = ['cfm']
filenames = ['*.cfm', '*.cfml', '*.cfc']
mimetypes = ['application/x-coldfusion']
def __init__(self, **options):
super(ColdfusionHtmlLexer, self).__init__(HtmlLexer, ColdfusionMarkupLexer,
**options)
class SspLexer(DelegatingLexer):
"""
Lexer for Scalate Server Pages.
*New in Pygments 1.4.*
"""
name = 'Scalate Server Page'
aliases = ['ssp']
filenames = ['*.ssp']
mimetypes = ['application/x-ssp']
def __init__(self, **options):
super(SspLexer, self).__init__(XmlLexer, JspRootLexer, **options)
def analyse_text(text):
rv = 0.0
if re.search('val \w+\s*:', text):
rv += 0.6
if looks_like_xml(text):
rv += 0.2
if '<%' in text and '%>' in text:
rv += 0.1
return rv
class TeaTemplateRootLexer(RegexLexer):
"""
Base for the `TeaTemplateLexer`. Yields `Token.Other` for area outside of
code blocks.
*New in Pygments 1.5.*
"""
tokens = {
'root': [
(r'<%\S?', Keyword, 'sec'),
(r'[^<]+', Other),
(r'<', Other),
],
'sec': [
(r'%>', Keyword, '#pop'),
# note: '\w\W' != '.' without DOTALL.
(r'[\w\W]+?(?=%>|\Z)', using(TeaLangLexer)),
],
}
class TeaTemplateLexer(DelegatingLexer):
"""
Lexer for `Tea Templates <http://teatrove.org/>`_.
*New in Pygments 1.5.*
"""
name = 'Tea'
aliases = ['tea']
filenames = ['*.tea']
mimetypes = ['text/x-tea']
def __init__(self, **options):
super(TeaTemplateLexer, self).__init__(XmlLexer,
TeaTemplateRootLexer, **options)
def analyse_text(text):
rv = TeaLangLexer.analyse_text(text) - 0.01
if looks_like_xml(text):
rv += 0.4
if '<%' in text and '%>' in text:
rv += 0.1
return rv
class LassoHtmlLexer(DelegatingLexer):
"""
Subclass of the `LassoLexer` which highlights unhandled data with the
`HtmlLexer`.
Nested JavaScript and CSS is also highlighted.
*New in Pygments 1.6.*
"""
name = 'HTML+Lasso'
aliases = ['html+lasso']
alias_filenames = ['*.html', '*.htm', '*.xhtml', '*.lasso', '*.lasso[89]',
'*.incl', '*.inc', '*.las']
mimetypes = ['text/html+lasso',
'application/x-httpd-lasso',
'application/x-httpd-lasso[89]']
def __init__(self, **options):
super(LassoHtmlLexer, self).__init__(HtmlLexer, LassoLexer, **options)
def analyse_text(text):
rv = LassoLexer.analyse_text(text) - 0.01
if re.search(r'<\w+>', text, re.I):
rv += 0.2
if html_doctype_matches(text):
rv += 0.5
return rv
class LassoXmlLexer(DelegatingLexer):
"""
Subclass of the `LassoLexer` which highlights unhandled data with the
`XmlLexer`.
*New in Pygments 1.6.*
"""
name = 'XML+Lasso'
aliases = ['xml+lasso']
alias_filenames = ['*.xml', '*.lasso', '*.lasso[89]',
'*.incl', '*.inc', '*.las']
mimetypes = ['application/xml+lasso']
def __init__(self, **options):
super(LassoXmlLexer, self).__init__(XmlLexer, LassoLexer, **options)
def analyse_text(text):
rv = LassoLexer.analyse_text(text) - 0.01
if looks_like_xml(text):
rv += 0.4
return rv
class LassoCssLexer(DelegatingLexer):
"""
Subclass of the `LassoLexer` which highlights unhandled data with the
`CssLexer`.
*New in Pygments 1.6.*
"""
name = 'CSS+Lasso'
aliases = ['css+lasso']
alias_filenames = ['*.css']
mimetypes = ['text/css+lasso']
def __init__(self, **options):
options['requiredelimiters'] = True
super(LassoCssLexer, self).__init__(CssLexer, LassoLexer, **options)
def analyse_text(text):
rv = LassoLexer.analyse_text(text) - 0.05
if re.search(r'\w+:.+?;', text):
rv += 0.1
if 'padding:' in text:
rv += 0.1
return rv
class LassoJavascriptLexer(DelegatingLexer):
"""
Subclass of the `LassoLexer` which highlights unhandled data with the
`JavascriptLexer`.
*New in Pygments 1.6.*
"""
name = 'JavaScript+Lasso'
aliases = ['js+lasso', 'javascript+lasso']
alias_filenames = ['*.js']
mimetypes = ['application/x-javascript+lasso',
'text/x-javascript+lasso',
'text/javascript+lasso']
def __init__(self, **options):
options['requiredelimiters'] = True
super(LassoJavascriptLexer, self).__init__(JavascriptLexer, LassoLexer,
**options)
def analyse_text(text):
rv = LassoLexer.analyse_text(text) - 0.05
if 'function' in text:
rv += 0.2
return rv