--- a/vim/bundle/sparkup/ftplugin/html/sparkup.py Mon Nov 22 14:45:15 2010 -0500
+++ /dev/null Thu Jan 01 00:00:00 1970 +0000
@@ -1,950 +0,0 @@
-#!/usr/bin/env python
-# -*- coding: utf-8 -*-
-
-import sys
-import re
-from optparse import OptionParser
-
-VERSION = "v0.1.99-20100514"
-
-# =============================================================================
-
-class Dialect:
- shortcuts = {}
- synonyms = {}
- required = {}
- short_tags = ()
-
-class HtmlDialect(Dialect):
- shortcuts = {
- 'cc:ie': {
- 'opening_tag': '<!--[if IE]>',
- 'closing_tag': '<![endif]-->'},
- 'cc:ie6': {
- 'opening_tag': '<!--[if lte IE 6]>',
- 'closing_tag': '<![endif]-->'},
- 'cc:ie7': {
- 'opening_tag': '<!--[if lte IE 7]>',
- 'closing_tag': '<![endif]-->'},
- 'cc:noie': {
- 'opening_tag': '<!--[if !IE]><!-->',
- 'closing_tag': '<!--<![endif]-->'},
- 'html:4t': {
- 'expand': True,
- 'opening_tag':
- '<!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01 Transitional//EN" "http://www.w3.org/TR/html4/loose.dtd">\n' +
- '<html lang="en">\n' +
- '<head>\n' +
- ' ' + '<meta http-equiv="Content-Type" content="text/html;charset=UTF-8" />\n' +
- ' ' + '<title></title>\n' +
- '</head>\n' +
- '<body>',
- 'closing_tag':
- '</body>\n' +
- '</html>'},
- 'html:4s': {
- 'expand': True,
- 'opening_tag':
- '<!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01//EN" "http://www.w3.org/TR/html4/strict.dtd">\n' +
- '<html lang="en">\n' +
- '<head>\n' +
- ' ' + '<meta http-equiv="Content-Type" content="text/html;charset=UTF-8" />\n' +
- ' ' + '<title></title>\n' +
- '</head>\n' +
- '<body>',
- 'closing_tag':
- '</body>\n' +
- '</html>'},
- 'html:xt': {
- 'expand': True,
- 'opening_tag':
- '<!DOCTYPE html PUBLIC "-//W3C//DTD XHTML 1.0 Transitional//EN" "http://www.w3.org/TR/xhtml1/DTD/xhtml1-transitional.dtd">\n' +
- '<html xmlns="http://www.w3.org/1999/xhtml" xml:lang="en">\n' +
- '<head>\n' +
- ' ' + '<meta http-equiv="Content-Type" content="text/html;charset=UTF-8" />\n' +
- ' ' + '<title></title>\n' +
- '</head>\n' +
- '<body>',
- 'closing_tag':
- '</body>\n' +
- '</html>'},
- 'html:xs': {
- 'expand': True,
- 'opening_tag':
- '<!DOCTYPE html PUBLIC "-//W3C//DTD XHTML 1.0 Strict//EN" "http://www.w3.org/TR/xhtml1/DTD/xhtml1-strict.dtd">\n' +
- '<html xmlns="http://www.w3.org/1999/xhtml" xml:lang="en">\n' +
- '<head>\n' +
- ' ' + '<meta http-equiv="Content-Type" content="text/html;charset=UTF-8" />\n' +
- ' ' + '<title></title>\n' +
- '</head>\n' +
- '<body>',
- 'closing_tag':
- '</body>\n' +
- '</html>'},
- 'html:xxs': {
- 'expand': True,
- 'opening_tag':
- '<!DOCTYPE html PUBLIC "-//W3C//DTD XHTML 1.1//EN" "http://www.w3.org/TR/xhtml11/DTD/xhtml11.dtd">\n' +
- '<html xmlns="http://www.w3.org/1999/xhtml" xml:lang="en">\n' +
- '<head>\n' +
- ' ' + '<meta http-equiv="Content-Type" content="text/html;charset=UTF-8" />\n' +
- ' ' + '<title></title>\n' +
- '</head>\n' +
- '<body>',
- 'closing_tag':
- '</body>\n' +
- '</html>'},
- 'html:5': {
- 'expand': True,
- 'opening_tag':
- '<!DOCTYPE html>\n' +
- '<html lang="en">\n' +
- '<head>\n' +
- ' ' + '<meta charset="UTF-8" />\n' +
- ' ' + '<title></title>\n' +
- '</head>\n' +
- '<body>',
- 'closing_tag':
- '</body>\n' +
- '</html>'},
- 'input:button': {
- 'name': 'input',
- 'attributes': { 'class': 'button', 'type': 'button', 'name': '', 'value': '' }
- },
- 'input:password': {
- 'name': 'input',
- 'attributes': { 'class': 'text password', 'type': 'password', 'name': '', 'value': '' }
- },
- 'input:radio': {
- 'name': 'input',
- 'attributes': { 'class': 'radio', 'type': 'radio', 'name': '', 'value': '' }
- },
- 'input:checkbox': {
- 'name': 'input',
- 'attributes': { 'class': 'checkbox', 'type': 'checkbox', 'name': '', 'value': '' }
- },
- 'input:file': {
- 'name': 'input',
- 'attributes': { 'class': 'file', 'type': 'file', 'name': '', 'value': '' }
- },
- 'input:text': {
- 'name': 'input',
- 'attributes': { 'class': 'text', 'type': 'text', 'name': '', 'value': '' }
- },
- 'input:submit': {
- 'name': 'input',
- 'attributes': { 'class': 'submit', 'type': 'submit', 'value': '' }
- },
- 'input:hidden': {
- 'name': 'input',
- 'attributes': { 'type': 'hidden', 'name': '', 'value': '' }
- },
- 'script:src': {
- 'name': 'script',
- 'attributes': { 'src': '' }
- },
- 'script:jquery': {
- 'name': 'script',
- 'attributes': { 'src': 'http://ajax.googleapis.com/ajax/libs/jquery/1.3.2/jquery.min.js' }
- },
- 'script:jsapi': {
- 'name': 'script',
- 'attributes': { 'src': 'http://www.google.com/jsapi' }
- },
- 'script:jsapix': {
- 'name': 'script',
- 'text': '\n google.load("jquery", "1.3.2");\n google.setOnLoadCallback(function() {\n \n });\n'
- },
- 'link:css': {
- 'name': 'link',
- 'attributes': { 'rel': 'stylesheet', 'type': 'text/css', 'href': '', 'media': 'all' },
- },
- 'link:print': {
- 'name': 'link',
- 'attributes': { 'rel': 'stylesheet', 'type': 'text/css', 'href': '', 'media': 'print' },
- },
- 'link:favicon': {
- 'name': 'link',
- 'attributes': { 'rel': 'shortcut icon', 'type': 'image/x-icon', 'href': '' },
- },
- 'link:touch': {
- 'name': 'link',
- 'attributes': { 'rel': 'apple-touch-icon', 'href': '' },
- },
- 'link:rss': {
- 'name': 'link',
- 'attributes': { 'rel': 'alternate', 'type': 'application/rss+xml', 'title': 'RSS', 'href': '' },
- },
- 'link:atom': {
- 'name': 'link',
- 'attributes': { 'rel': 'alternate', 'type': 'application/atom+xml', 'title': 'Atom', 'href': '' },
- },
- 'meta:ie7': {
- 'name': 'meta',
- 'attributes': { 'http-equiv': 'X-UA-Compatible', 'content': 'IE=7' },
- },
- 'meta:ie8': {
- 'name': 'meta',
- 'attributes': { 'http-equiv': 'X-UA-Compatible', 'content': 'IE=8' },
- },
- 'form:get': {
- 'name': 'form',
- 'attributes': { 'method': 'get' },
- },
- 'form:g': {
- 'name': 'form',
- 'attributes': { 'method': 'get' },
- },
- 'form:post': {
- 'name': 'form',
- 'attributes': { 'method': 'post' },
- },
- 'form:p': {
- 'name': 'form',
- 'attributes': { 'method': 'post' },
- },
- }
- synonyms = {
- 'checkbox': 'input:checkbox',
- 'check': 'input:checkbox',
- 'input:c': 'input:checkbox',
- 'button': 'input:button',
- 'input:b': 'input:button',
- 'input:h': 'input:hidden',
- 'hidden': 'input:hidden',
- 'submit': 'input:submit',
- 'input:s': 'input:submit',
- 'radio': 'input:radio',
- 'input:r': 'input:radio',
- 'text': 'input:text',
- 'passwd': 'input:password',
- 'password': 'input:password',
- 'pw': 'input:password',
- 'input:t': 'input:text',
- 'linkcss': 'link:css',
- 'scriptsrc': 'script:src',
- 'jquery': 'script:jquery',
- 'jsapi': 'script:jsapi',
- 'html5': 'html:5',
- 'html4': 'html:4s',
- 'html4s': 'html:4s',
- 'html4t': 'html:4t',
- 'xhtml': 'html:xxs',
- 'xhtmlt': 'html:xt',
- 'xhtmls': 'html:xs',
- 'xhtml11': 'html:xxs',
- 'opt': 'option',
- 'st': 'strong',
- 'css': 'style',
- 'csss': 'link:css',
- 'css:src': 'link:css',
- 'csssrc': 'link:css',
- 'js': 'script',
- 'jss': 'script:src',
- 'js:src': 'script:src',
- 'jssrc': 'script:src',
- }
- short_tags = (
- 'area', 'base', 'basefont', 'br', 'embed', 'hr',
- 'input', 'img', 'link', 'param', 'meta')
- required = {
- 'a': {'href':''},
- 'base': {'href':''},
- 'abbr': {'title': ''},
- 'acronym':{'title': ''},
- 'bdo': {'dir': ''},
- 'link': {'rel': 'stylesheet', 'href': ''},
- 'style': {'type': 'text/css'},
- 'script': {'type': 'text/javascript'},
- 'img': {'src':'', 'alt':''},
- 'iframe': {'src': '', 'frameborder': '0'},
- 'embed': {'src': '', 'type': ''},
- 'object': {'data': '', 'type': ''},
- 'param': {'name': '', 'value': ''},
- 'form': {'action': '', 'method': 'post'},
- 'table': {'cellspacing': '0'},
- 'input': {'type': '', 'name': '', 'value': ''},
- 'base': {'href': ''},
- 'area': {'shape': '', 'coords': '', 'href': '', 'alt': ''},
- 'select': {'name': ''},
- 'option': {'value': ''},
- 'textarea':{'name': ''},
- 'meta': {'content': ''},
- }
-
-class Parser:
- """The parser.
- """
-
- # Constructor
- # -------------------------------------------------------------------------
-
- def __init__(self, options, string, dialect=HtmlDialect()):
- """Constructor.
- """
-
- self.tokens = []
- self.string = string
- self.options = options
- self.dialect = dialect
- self.root = Element(parser=self)
- self.caret = []
- self.caret.append(self.root)
- self._last = []
-
- self._tokenize()
- self._parse()
-
-
- # Methods
- # -------------------------------------------------------------------------
-
- def render(self):
- """Renders.
- Called by [[Router]].
- """
-
- # Get the initial render of the root node
- output = self.root.render()
-
- # Indent by whatever the input is indented with
- indent = re.findall(r"^[\r\n]*(\s*)", self.string)[0]
- output = indent + output.replace("\n", "\n" + indent)
-
- # Strip newline if not needed
- if not self.options.last_newline \
- or self.prefix or self.suffix:
- output = re.sub(r'\n\s*$', '', output)
-
- # TextMate mode
- if self.options.textmate:
- output = self._textmatify(output)
-
- return output
-
- # Protected methods
- # -------------------------------------------------------------------------
-
- def _textmatify(self, output):
- """Returns a version of the output with TextMate placeholders in it.
- """
-
- matches = re.findall(r'(></)|("")|(\n\s+)\n|(.|\s)', output)
- output = ''
- n = 1
- for i in matches:
- if i[0]:
- output += '>$%i</' % n
- n += 1
- elif i[1]:
- output += '"$%i"' % n
- n += 1
- elif i[2]:
- output += i[2] + '$%i\n' % n
- n += 1
- elif i[3]:
- output += i[3]
- output += "$0"
- return output
-
- def _tokenize(self):
- """Tokenizes.
- Initializes [[self.tokens]].
- """
-
- string = self.string.strip()
-
- # Find prefix/suffix
- while True:
- match = re.match(r"^(\s*<[^>]+>\s*)", string)
- if match is None: break
- if self.prefix is None: self.prefix = ''
- self.prefix += match.group(0)
- string = string[len(match.group(0)):]
-
- while True:
- match = re.findall(r"(\s*<[^>]+>[\s\n\r]*)$", string)
- if not match: break
- if self.suffix is None: self.suffix = ''
- self.suffix = match[0] + self.suffix
- string = string[:-len(match[0])]
-
- # Split by the element separators
- for token in re.split('(<|>|\+(?!\\s*\+|$))', string):
- if token.strip() != '':
- self.tokens.append(Token(token, parser=self))
-
- def _parse(self):
- """Takes the tokens and does its thing.
- Populates [[self.root]].
- """
-
- # Carry it over to the root node.
- if self.prefix or self.suffix:
- self.root.prefix = self.prefix
- self.root.suffix = self.suffix
- self.root.depth += 1
-
- for token in self.tokens:
- if token.type == Token.ELEMENT:
- # Reset the "last elements added" list. We will
- # repopulate this with the new elements added now.
- self._last[:] = []
-
- # Create [[Element]]s from a [[Token]].
- # They will be created as many as the multiplier specifies,
- # multiplied by how many carets we have
- count = 0
- for caret in self.caret:
- for local_count in range(1, token.multiplier + 1):
- count += 1
- new = Element(token, caret,
- count = count,
- local_count = local_count,
- parser = self)
- self._last.append(new)
- caret.append(new)
-
- # For >
- elif token.type == Token.CHILD:
- # The last children added.
- self.caret[:] = self._last
-
- # For <
- elif token.type == Token.PARENT:
- # If we're the root node, don't do anything
- parent = self.caret[0].parent
- if parent is not None:
- self.caret[:] = [parent]
-
- # Properties
- # -------------------------------------------------------------------------
-
- # Property: dialect
- # The dialect of XML
- dialect = None
-
- # Property: string
- # The string
- string = ''
-
- # Property: tokens
- # The list of tokens
- tokens = []
-
- # Property: options
- # Reference to the [[Options]] instance
- options = None
-
- # Property: root
- # The root [[Element]] node.
- root = None
-
- # Property: caret
- # The current insertion point.
- caret = None
-
- # Property: _last
- # List of the last appended stuff
- _last = None
-
- # Property: indent
- # Yeah
- indent = ''
-
- # Property: prefix
- # (String) The trailing tag in the beginning.
- #
- # Description:
- # For instance, in `<div>ul>li</div>`, the `prefix` is `<div>`.
- prefix = ''
-
- # Property: suffix
- # (string) The trailing tag at the end.
- suffix = ''
-
-# =============================================================================
-
-class Element:
- """An element.
- """
-
- def __init__(self, token=None, parent=None, count=None, local_count=None,
- parser=None, opening_tag=None, closing_tag=None,
- attributes=None, name=None, text=None):
- """Constructor.
-
- This is called by ???.
-
- Description:
- All parameters are optional.
-
- token - (Token) The token (required)
- parent - (Element) Parent element; `None` if root
- count - (Int) The number to substitute for `&` (e.g., in `li.item-&`)
- local_count - (Int) The number to substitute for `$` (e.g., in `li.item-$`)
- parser - (Parser) The parser
-
- attributes - ...
- name - ...
- text - ...
- """
-
- self.children = []
- self.attributes = {}
- self.parser = parser
-
- if token is not None:
- # Assumption is that token is of type [[Token]] and is
- # a [[Token.ELEMENT]].
- self.name = token.name
- self.attributes = token.attributes.copy()
- self.text = token.text
- self.populate = token.populate
- self.expand = token.expand
- self.opening_tag = token.opening_tag
- self.closing_tag = token.closing_tag
-
- # `count` can be given. This will substitude & in classname and ID
- if count is not None:
- for key in self.attributes:
- attrib = self.attributes[key]
- attrib = attrib.replace('&', ("%i" % count))
- if local_count is not None:
- attrib = attrib.replace('$', ("%i" % local_count))
- self.attributes[key] = attrib
-
- # Copy over from parameters
- if attributes: self.attributes = attributes
- if name: self.name = name
- if text: self.text = text
-
- self._fill_attributes()
-
- self.parent = parent
- if parent is not None:
- self.depth = parent.depth + 1
-
- if self.populate: self._populate()
-
- def render(self):
- """Renders the element, along with it's subelements, into HTML code.
-
- [Grouped under "Rendering methods"]
- """
-
- options = self.parser.options
- output = ""
- spaces_count = options.indent_spaces
- spaces = ' ' * spaces_count
- indent = self.depth * spaces
-
- prefix, suffix = ('', '')
- if self.prefix: prefix = self.prefix + "\n"
- if self.suffix: suffix = self.suffix
-
- # Make the guide from the ID (/#header), or the class if there's no
- # ID (/.item)
- # This is for the start-guide, end-guide and post_tag_guides
- guide_string = ''
- if 'id' in self.attributes:
- guide_string += "#%s" % self.attributes['id']
- elif 'class' in self.attributes:
- guide_string += ".%s" % self.attributes['class'].replace(' ', '.')
-
- # Build the post-tag guide (e.g., </div><!-- /#header -->),
- # the start guide, and the end guide.
- guide = ''
- start_guide = ''
- end_guide = ''
-
- if ((self.name == 'div') and \
- (('id' in self.attributes) or ('class' in self.attributes))):
-
- if (options.post_tag_guides):
- guide = "<!-- /%s -->" % guide_string
-
- if (options.start_guide_format):
- format = options.start_guide_format
- try: start_guide = format % guide_string
- except: start_guide = (format + " " + guide_string).strip()
- start_guide = "%s<!-- %s -->\n" % (indent, start_guide)
-
- if (options.end_guide_format):
- format = options.end_guide_format
- try: end_guide = format % guide_string
- except: end_guide = (format + " " + guide_string).strip()
-
- if options.end_guide_newline:
- end_guide = "\n%s<!-- %s -->" % (indent, end_guide)
- else:
- end_guide = "<!-- %s -->" % (end_guide)
-
- # Short, self-closing tags (<br />)
- short_tags = self.parser.dialect.short_tags
-
- # When it should be expanded..
- # (That is, <div>\n...\n</div> or similar -- wherein something must go
- # inside the opening/closing tags)
- if len(self.children) > 0 \
- or self.expand \
- or prefix or suffix \
- or (self.parser.options.expand_divs and self.name == 'div'):
-
- for child in self.children:
- output += child.render()
-
- # For expand divs: if there are no children (that is, `output`
- # is still blank despite above), fill it with a blank line.
- if (output == ''): output = indent + spaces + "\n"
-
- # If we're a root node and we have a prefix or suffix...
- # (Only the root node can have a prefix or suffix.)
- if prefix or suffix:
- output = "%s%s%s%s%s\n" % \
- (indent, prefix, output, suffix, guide)
-
- # Uh..
- elif self.name != '' or \
- self.opening_tag is not None or \
- self.closing_tag is not None:
- output = start_guide + \
- indent + self.get_opening_tag() + "\n" + \
- output + \
- indent + self.get_closing_tag() + \
- guide + end_guide + "\n"
-
-
- # Short, self-closing tags (<br />)
- elif self.name in short_tags:
- output = "%s<%s />\n" % (indent, self.get_default_tag())
-
- # Tags with text, possibly
- elif self.name != '' or \
- self.opening_tag is not None or \
- self.closing_tag is not None:
- output = "%s%s%s%s%s%s%s%s" % \
- (start_guide, indent, self.get_opening_tag(), \
- self.text, \
- self.get_closing_tag(), \
- guide, end_guide, "\n")
-
- return output
-
- def get_default_tag(self):
- """Returns the opening tag (without brackets).
-
- Usage:
- element.get_default_tag()
-
- [Grouped under "Rendering methods"]
- """
-
- output = '%s' % (self.name)
- for key, value in self.attributes.iteritems():
- output += ' %s="%s"' % (key, value)
- return output
-
- def get_opening_tag(self):
- if self.opening_tag is None:
- return "<%s>" % self.get_default_tag()
- else:
- return self.opening_tag
-
- def get_closing_tag(self):
- if self.closing_tag is None:
- return "</%s>" % self.name
- else:
- return self.closing_tag
-
- def append(self, object):
- """Registers an element as a child of this element.
-
- Usage:
- element.append(child)
-
- Description:
- Adds a given element `child` to the children list of this element. It
- will be rendered when [[render()]] is called on the element.
-
- See also:
- - [[get_last_child()]]
-
- [Grouped under "Traversion methods"]
- """
-
- self.children.append(object)
-
- def get_last_child(self):
- """Returns the last child element which was [[append()]]ed
- to this element.
-
- Usage:
- element.get_last_child()
-
- Description:
- This is the same as using `element.children[-1]`.
-
- [Grouped under "Traversion methods"]
- """
-
- return self.children[-1]
-
- def _populate(self):
- """Expands with default items.
-
- This is called when the [[populate]] flag is turned on.
- """
-
- if self.name == 'ul':
- elements = [Element(name='li', parent=self, parser=self.parser)]
-
- elif self.name == 'dl':
- elements = [
- Element(name='dt', parent=self, parser=self.parser),
- Element(name='dd', parent=self, parser=self.parser)]
-
- elif self.name == 'table':
- tr = Element(name='tr', parent=self, parser=self.parser)
- td = Element(name='td', parent=tr, parser=self.parser)
- tr.children.append(td)
- elements = [tr]
-
- else:
- elements = []
-
- for el in elements:
- self.children.append(el)
-
- def _fill_attributes(self):
- """Fills default attributes for certain elements.
-
- Description:
- This is called by the constructor.
-
- [Protected, grouped under "Protected methods"]
- """
-
- # Make sure <a>'s have a href, <img>'s have an src, etc.
- required = self.parser.dialect.required
-
- for element, attribs in required.iteritems():
- if self.name == element:
- for attrib in attribs:
- if attrib not in self.attributes:
- self.attributes[attrib] = attribs[attrib]
-
- # -------------------------------------------------------------------------
-
- # Property: last_child
- # [Read-only]
- last_child = property(get_last_child)
-
- # -------------------------------------------------------------------------
-
- # Property: parent
- # (Element) The parent element.
- parent = None
-
- # Property: name
- # (String) The name of the element (e.g., `div`)
- name = ''
-
- # Property: attributes
- # (Dict) The dictionary of attributes (e.g., `{'src': 'image.jpg'}`)
- attributes = None
-
- # Property: children
- # (List of Elements) The children
- children = None
-
- # Property: opening_tag
- # (String or None) The opening tag. Optional; will use `name` and
- # `attributes` if this is not given.
- opening_tag = None
-
- # Property: closing_tag
- # (String or None) The closing tag
- closing_tag = None
-
- text = ''
- depth = -1
- expand = False
- populate = False
- parser = None
-
- # Property: prefix
- # Only the root note can have this.
- prefix = None
- suffix = None
-
-# =============================================================================
-
-class Token:
- def __init__(self, string, parser=None):
- """Token.
-
- Description:
- string - The string to parse
-
- In the string `div > ul`, there are 3 tokens. (`div`, `>`, and `ul`)
-
- For `>`, it will be a `Token` with `type` set to `Token.CHILD`
- """
-
- self.string = string.strip()
- self.attributes = {}
- self.parser = parser
-
- # Set the type.
- if self.string == '<':
- self.type = Token.PARENT
- elif self.string == '>':
- self.type = Token.CHILD
- elif self.string == '+':
- self.type = Token.SIBLING
- else:
- self.type = Token.ELEMENT
- self._init_element()
-
- def _init_element(self):
- """Initializes. Only called if the token is an element token.
- [Private]
- """
-
- # Get the tag name. Default to DIV if none given.
- name = re.findall(r'^([\w\-:]*)', self.string)[0]
- name = name.lower().replace('-', ':')
-
- # Find synonyms through this thesaurus
- synonyms = self.parser.dialect.synonyms
- if name in synonyms.keys():
- name = synonyms[name]
-
- if ':' in name:
- shortcuts = self.parser.dialect.shortcuts
- if name in shortcuts.keys():
- for key, value in shortcuts[name].iteritems():
- setattr(self, key, value)
- if 'html' in name:
- return
- else:
- self.name = name
-
- elif (name == ''): self.name = 'div'
- else: self.name = name
-
- # Look for attributes
- attribs = []
- for attrib in re.findall(r'\[([^\]]*)\]', self.string):
- attribs.append(attrib)
- self.string = self.string.replace("[" + attrib + "]", "")
- if len(attribs) > 0:
- for attrib in attribs:
- try: key, value = attrib.split('=', 1)
- except: key, value = attrib, ''
- self.attributes[key] = value
-
- # Try looking for text
- text = None
- for text in re.findall(r'\{([^\}]*)\}', self.string):
- self.string = self.string.replace("{" + text + "}", "")
- if text is not None:
- self.text = text
-
- # Get the class names
- classes = []
- for classname in re.findall(r'\.([\$a-zA-Z0-9_\-\&]+)', self.string):
- classes.append(classname)
- if len(classes) > 0:
- try: self.attributes['class']
- except: self.attributes['class'] = ''
- self.attributes['class'] += ' ' + ' '.join(classes)
- self.attributes['class'] = self.attributes['class'].strip()
-
- # Get the ID
- id_match = re.search(r'#([\$a-zA-Z0-9_\-\&]+)', self.string)
- if id_match is not None:
- self.attributes['id'] = id_match.group(1)
-
- # See if there's a multiplier (e.g., "li*3")
- multiplier_match = re.search(r'\*\s*([0-9]+)', self.string)
- if multiplier_match is not None:
- self.multiplier = int(multiplier_match.group(1))
-
- # Populate flag (e.g., ul+)
- flags_match = re.search(r'[\+\!]+$', self.string)
- if flags_match is not None:
- if '+' in flags_match.group(0):
- self.populate = True
- if '!' in flags_match.group(0):
- self.expand = True
-
- def __str__(self):
- return self.string
-
- string = ''
- parser = None
-
- # For elements
- # See the properties of `Element` for description on these.
- name = ''
- attributes = None
- multiplier = 1
- expand = False
- populate = False
- text = ''
- opening_tag = None
- closing_tag = None
-
- # Type
- type = 0
- ELEMENT = 2
- CHILD = 4
- PARENT = 8
- SIBLING = 16
-
-# =============================================================================
-
-def parse_args():
- optparser = OptionParser(version=VERSION,
- description="Expands input into HTML.",
- epilog="Please refer to the manual for more information.")
-
-
- optparser.add_option('--no-guides', action="store_true", help='Deprecated')
- optparser.add_option('--post-tag-guides', action="store_true",
- help='Adds comments at the end of DIV tags')
- optparser.add_option('--textmate', action="store_true",
- help='Adds snippet info (textmate mode)')
- optparser.add_option('--indent-spaces', help='Indent spaces')
- optparser.add_option('--expand-divs', action="store_true",
- help='Automatically expand divs')
- optparser.add_option('--no-last-newline', action="store_false",
- help='Skip the trailing newline')
- optparser.add_option('--start-guide-format', help='To be documented')
- optparser.add_option('--end-guide-format', help='To be documented')
- optparser.add_option('--end-guide-newline', help='To be documented')
-
- optparser.set_defaults(post_tag_guides=False, textmate=False,
- indent_spaces=4, expand_divs=False, last_newline=True,
- start_guide_format="", end_guide_format="", end_guide_newline=True)
-
- # Make sure they're the correct types
- opt_args = optparser.parse_args()
- opt_args[0].indent_spaces = int(opt_args[0].indent_spaces)
- opt_args[0].end_guide_newline = bool(int(opt_args[0].end_guide_newline))
- return opt_args
-
-def main():
- (options, _) = parse_args()
-
- lines = sys.stdin.read()
-
- parser = Parser(options, lines)
-
- output = parser.render()
- sys.stdout.write(output)
-
-if __name__ == "__main__":
- main()