/compat/markdown2.py

Large files files are truncated, but you can click here to view the full file

#!/usr/bin/env python
# Copyright (c) 2007-2008 ActiveState Corp.
# License: MIT (http://www.opensource.org/licenses/mit-license.php)

r"""A fast and complete Python implementation of Markdown.

[from http://daringfireball.net/projects/markdown/]
> Markdown is a text-to-HTML filter; it translates an easy-to-read /
> easy-to-write structured text format into HTML.  Markdown's text
> format is most similar to that of plain text email, and supports
> features such as headers, *emphasis*, code blocks, blockquotes, and
> links.
>
> Markdown's syntax is designed not as a generic markup language, but
> specifically to serve as a front-end to (X)HTML. You can use span-level
> HTML tags anywhere in a Markdown document, and you can use block level
> HTML tags (like <div> and <table> as well).

Module usage:

    >>> import markdown2
    >>> markdown2.markdown("*boo!*")  # or use `html = markdown_path(PATH)`
    u'<p><em>boo!</em></p>\n'

    >>> markdowner = Markdown()
    >>> markdowner.convert("*boo!*")
    u'<p><em>boo!</em></p>\n'
    >>> markdowner.convert("**boom!**")
    u'<p><strong>boom!</strong></p>\n'

This implementation of Markdown implements the full "core" syntax plus a
number of extras (e.g., code syntax coloring, footnotes) as described on
<http://code.google.com/p/python-markdown2/wiki/Extras>.
"""

cmdln_desc = """A fast and complete Python implementation of Markdown, a
text-to-HTML conversion tool for web writers.
"""

# Dev Notes:
# - There is already a Python markdown processor
#   (http://www.freewisdom.org/projects/python-markdown/).
# - Python's regex syntax doesn't have '\z', so I'm using '\Z'. I'm
#   not yet sure if there implications with this. Compare 'pydoc sre'
#   and 'perldoc perlre'.

__version_info__ = (1, 0, 1, 16) # first three nums match Markdown.pl
__version__ = '1.0.1.16'
__author__ = "Trent Mick"

import os
import sys
from pprint import pprint
import re
import logging
try:
    from hashlib import md5
except ImportError:
    from md5 import md5
import optparse
from random import random, randint
import codecs
from urllib import quote



#---- Python version compat

if sys.version_info[:2] < (2,4):
    from sets import Set as set
    def reversed(sequence):
        for i in sequence[::-1]:
            yield i
    def _unicode_decode(s, encoding, errors='xmlcharrefreplace'):
        return unicode(s, encoding, errors)
else:
    def _unicode_decode(s, encoding, errors='strict'):
        return s.decode(encoding, errors)


#---- globals

DEBUG = False
log = logging.getLogger("markdown")

DEFAULT_TAB_WIDTH = 4


try:
    import uuid
except ImportError:
    SECRET_SALT = str(randint(0, 1000000))
else:
    SECRET_SALT = str(uuid.uuid4())
def _hash_ascii(s):
    #return md5(s).hexdigest()   # Markdown.pl effectively does this.
    return 'md5-' + md5(SECRET_SALT + s).hexdigest()
def _hash_text(s):
    return 'md5-' + md5(SECRET_SALT + s.encode("utf-8")).hexdigest()

# Table of hash values for escaped characters:
g_escape_table = dict([(ch, _hash_ascii(ch))
                       for ch in '\\`*_{}[]()>#+-.!'])



#---- exceptions

class MarkdownError(Exception):
    pass



#---- public api

def markdown_path(path, encoding="utf-8",
                  html4tags=False, tab_width=DEFAULT_TAB_WIDTH,
                  safe_mode=None, extras=None, link_patterns=None,
                  use_file_vars=False):
    fp = codecs.open(path, 'r', encoding)
    text = fp.read()
    fp.close()
    return Markdown(html4tags=html4tags, tab_width=tab_width,
                    safe_mode=safe_mode, extras=extras,
                    link_patterns=link_patterns,
                    use_file_vars=use_file_vars).convert(text)

def markdown(text, html4tags=False, tab_width=DEFAULT_TAB_WIDTH,
             safe_mode=None, extras=None, link_patterns=None,
             use_file_vars=False):
    return Markdown(html4tags=html4tags, tab_width=tab_width,
                    safe_mode=safe_mode, extras=extras,
                    link_patterns=link_patterns,
                    use_file_vars=use_file_vars).convert(text)

class Markdown(object):
    # The dict of "extras" to enable in processing -- a mapping of
    # extra name to argument for the extra. Most extras do not have an
    # argument, in which case the value is None.
    #
    # This can be set via (a) subclassing and (b) the constructor
    # "extras" argument.
    extras = None

    urls = None
    titles = None
    html_blocks = None
    html_spans = None
    html_removed_text = "[HTML_REMOVED]"  # for compat with markdown.py

    # Used to track when we're inside an ordered or unordered list
    # (see _ProcessListItems() for details):
    list_level = 0

    _ws_only_line_re = re.compile(r"^[ \t]+$", re.M)

    def __init__(self, html4tags=False, tab_width=4, safe_mode=None,
                 extras=None, link_patterns=None, use_file_vars=False):
        if html4tags:
            self.empty_element_suffix = ">"
        else:
            self.empty_element_suffix = " />"
        self.tab_width = tab_width

        # For compatibility with earlier markdown2.py and with
        # markdown.py's safe_mode being a boolean, 
        #   safe_mode == True -> "replace"
        if safe_mode is True:
            self.safe_mode = "replace"
        else:
            self.safe_mode = safe_mode

        if self.extras is None:
            self.extras = {}
        elif not isinstance(self.extras, dict):
            self.extras = dict([(e, None) for e in self.extras])
        if extras:
            if not isinstance(extras, dict):
                extras = dict([(e, None) for e in extras])
            self.extras.update(extras)
        assert isinstance(self.extras, dict)
        self._instance_extras = self.extras.copy()
        self.link_patterns = link_patterns
        self.use_file_vars = use_file_vars
        self._outdent_re = re.compile(r'^(\t|[ ]{1,%d})' % tab_width, re.M)

    def reset(self):
        self.urls = {}
        self.titles = {}
        self.html_blocks = {}
        self.html_spans = {}
        self.list_level = 0
        self.extras = self._instance_extras.copy()
        if "footnotes" in self.extras:
            self.footnotes = {}
            self.footnote_ids = []

    def convert(self, text):
        """Convert the given text."""
        # Main function. The order in which other subs are called here is
        # essential. Link and image substitutions need to happen before
        # _EscapeSpecialChars(), so that any *'s or _'s in the <a>
        # and <img> tags get encoded.

        # Clear the global hashes. If we don't clear these, you get conflicts
        # from other articles when generating a page which contains more than
        # one article (e.g. an index page that shows the N most recent
        # articles):
        self.reset()

        if not isinstance(text, unicode):
            #TODO: perhaps shouldn't presume UTF-8 for string input?
            text = unicode(text, 'utf-8')

        if self.use_file_vars:
            # Look for emacs-style file variable hints.
            emacs_vars = self._get_emacs_vars(text)
            if "markdown-extras" in emacs_vars:
                splitter = re.compile("[ ,]+")
                for e in splitter.split(emacs_vars["markdown-extras"]):
                    if '=' in e:
                        ename, earg = e.split('=', 1)
                        try:
                            earg = int(earg)
                        except ValueError:
                            pass
                    else:
                        ename, earg = e, None
                    self.extras[ename] = earg

        # Standardize line endings:
        text = re.sub("\r\n|\r", "\n", text)

        # Make sure $text ends with a couple of newlines:
        text += "\n\n"

        # Convert all tabs to spaces.
        text = self._detab(text)

        # Strip any lines consisting only of spaces and tabs.
        # This makes subsequent regexen easier to write, because we can
        # match consecutive blank lines with /\n+/ instead of something
        # contorted like /[ \t]*\n+/ .
        text = self._ws_only_line_re.sub("", text)

        if self.safe_mode:
            text = self._hash_html_spans(text)

        # Turn block-level HTML blocks into hash entries
        text = self._hash_html_blocks(text, raw=True)

        # Strip link definitions, store in hashes.
        if "footnotes" in self.extras:
            # Must do footnotes first because an unlucky footnote defn
            # looks like a link defn:
            #   [^4]: this "looks like a link defn"
            text = self._strip_footnote_definitions(text)
        text = self._strip_link_definitions(text)

        text = self._run_block_gamut(text)

        if "footnotes" in self.extras:
            text = self._add_footnotes(text)

        text = self._unescape_special_chars(text)

        if self.safe_mode:
            text = self._unhash_html_spans(text)

        text += "\n"
        return text

    _emacs_oneliner_vars_pat = re.compile(r"-\*-\s*([^\r\n]*?)\s*-\*-", re.UNICODE)
    # This regular expression is intended to match blocks like this:
    #    PREFIX Local Variables: SUFFIX
    #    PREFIX mode: Tcl SUFFIX
    #    PREFIX End: SUFFIX
    # Some notes:
    # - "[ \t]" is used instead of "\s" to specifically exclude newlines
    # - "(\r\n|\n|\r)" is used instead of "$" because the sre engine does
    #   not like anything other than Unix-style line terminators.
    _emacs_local_vars_pat = re.compile(r"""^
        (?P<prefix>(?:[^\r\n|\n|\r])*?)
        [\ \t]*Local\ Variables:[\ \t]*
        (?P<suffix>.*?)(?:\r\n|\n|\r)
        (?P<content>.*?\1End:)
        """, re.IGNORECASE | re.MULTILINE | re.DOTALL | re.VERBOSE)

    def _get_emacs_vars(self, text):
        """Return a dictionary of emacs-style local variables.

        Parsing is done loosely according to this spec (and according to
        some in-practice deviations from this):
        http://www.gnu.org/software/emacs/manual/html_node/emacs/Specifying-File-Variables.html#Specifying-File-Variables
        """
        emacs_vars = {}
        SIZE = pow(2, 13) # 8kB

        # Search near the start for a '-*-'-style one-liner of variables.
        head = text[:SIZE]
        if "-*-" in head:
            match = self._emacs_oneliner_vars_pat.search(head)
            if match:
                emacs_vars_str = match.group(1)
                assert '\n' not in emacs_vars_str
                emacs_var_strs = [s.strip() for s in emacs_vars_str.split(';')
                                  if s.strip()]
                if len(emacs_var_strs) == 1 and ':' not in emacs_var_strs[0]:
                    # While not in the spec, this form is allowed by emacs:
                    #   -*- Tcl -*-
                    # where the implied "variable" is "mode". This form
                    # is only allowed if there are no other variables.
                    emacs_vars["mode"] = emacs_var_strs[0].strip()
                else:
                    for emacs_var_str in emacs_var_strs:
                        try:
                            variable, value = emacs_var_str.strip().split(':', 1)
                        except ValueError:
                            log.debug("emacs variables error: malformed -*- "
                                      "line: %r", emacs_var_str)
                            continue
                        # Lowercase the variable name because Emacs allows "Mode"
                        # or "mode" or "MoDe", etc.
                        emacs_vars[variable.lower()] = value.strip()

        tail = text[-SIZE:]
        if "Local Variables" in tail:
            match = self._emacs_local_vars_pat.search(tail)
            if match:
                prefix = match.group("prefix")
                suffix = match.group("suffix")
                lines = match.group("content").splitlines(0)
                #print "prefix=%r, suffix=%r, content=%r, lines: %s"\
                #      % (prefix, suffix, match.group("content"), lines)

                # Validate the Local Variables block: proper prefix and suffix
                # usage.
                for i, line in enumerate(lines):
                    if not line.startswith(prefix):
                        log.debug("emacs variables error: line '%s' "
                                  "does not use proper prefix '%s'"
                                  % (line, prefix))
                        return {}
                    # Don't validate suffix on last line. Emacs doesn't care,
                    # neither should we.
                    if i != len(lines)-1 and not line.endswith(suffix):
                        log.debug("emacs variables error: line '%s' "
                                  "does not use proper suffix '%s'"
                                  % (line, suffix))
                        return {}

                # Parse out one emacs var per line.
                continued_for = None
                for line in lines[:-1]: # no var on the last line ("PREFIX End:")
                    if prefix: line = line[len(prefix):] # strip prefix
                    if suffix: line = line[:-len(suffix)] # strip suffix
                    line = line.strip()
                    if continued_for:
                        variable = continued_for
                        if line.endswith('\\'):
                            line = line[:-1].rstrip()
                        else:
                            continued_for = None
                        emacs_vars[variable] += ' ' + line
                    else:
                        try:
                            variable, value = line.split(':', 1)
                        except ValueError:
                            log.debug("local variables error: missing colon "
                                      "in local variables entry: '%s'" % line)
                            continue
                        # Do NOT lowercase the variable name, because Emacs only
                        # allows "mode" (and not "Mode", "MoDe", etc.) in this block.
                        value = value.strip()
                        if value.endswith('\\'):
                            value = value[:-1].rstrip()
                            continued_for = variable
                        else:
                            continued_for = None
                        emacs_vars[variable] = value

        # Unquote values.
        for var, val in emacs_vars.items():
            if len(val) > 1 and (val.startswith('"') and val.endswith('"')
               or val.startswith('"') and val.endswith('"')):
                emacs_vars[var] = val[1:-1]

        return emacs_vars

    # Cribbed from a post by Bart Lateur:
    # <http://www.nntp.perl.org/group/perl.macperl.anyperl/154>
    _detab_re = re.compile(r'(.*?)\t', re.M)
    def _detab_sub(self, match):
        g1 = match.group(1)
        return g1 + (' ' * (self.tab_width - len(g1) % self.tab_width))
    def _detab(self, text):
        r"""Remove (leading?) tabs from a file.

            >>> m = Markdown()
            >>> m._detab("\tfoo")
            '    foo'
            >>> m._detab("  \tfoo")
            '    foo'
            >>> m._detab("\t  foo")
            '      foo'
            >>> m._detab("  foo")
            '  foo'
            >>> m._detab("  foo\n\tbar\tblam")
            '  foo\n    bar blam'
        """
        if '\t' not in text:
            return text
        return self._detab_re.subn(self._detab_sub, text)[0]

    _block_tags_a = 'p|div|h[1-6]|blockquote|pre|table|dl|ol|ul|script|noscript|form|fieldset|iframe|math|ins|del'
    _strict_tag_block_re = re.compile(r"""
        (                       # save in \1
            ^                   # start of line  (with re.M)
            <(%s)               # start tag = \2
            \b                  # word break
            (.*\n)*?            # any number of lines, minimally matching
            </\2>               # the matching end tag
            [ \t]*              # trailing spaces/tabs
            (?=\n+|\Z)          # followed by a newline or end of document
        )
        """ % _block_tags_a,
        re.X | re.M)

    _block_tags_b = 'p|div|h[1-6]|blockquote|pre|table|dl|ol|ul|script|noscript|form|fieldset|iframe|math'
    _liberal_tag_block_re = re.compile(r"""
        (                       # save in \1
            ^                   # start of line  (with re.M)
            <(%s)               # start tag = \2
            \b                  # word break
            (.*\n)*?            # any number of lines, minimally matching
            .*</\2>             # the matching end tag
            [ \t]*              # trailing spaces/tabs
            (?=\n+|\Z)          # followed by a newline or end of document
        )
        """ % _block_tags_b,
        re.X | re.M)

    def _hash_html_block_sub(self, match, raw=False):
        html = match.group(1)
        if raw and self.safe_mode:
            html = self._sanitize_html(html)
        key = _hash_text(html)
        self.html_blocks[key] = html
        return "\n\n" + key + "\n\n"

    def _hash_html_blocks(self, text, raw=False):
        """Hashify HTML blocks

        We only want to do this for block-level HTML tags, such as headers,
        lists, and tables. That's because we still want to wrap <p>s around
        "paragraphs" that are wrapped in non-block-level tags, such as anchors,
        phrase emphasis, and spans. The list of tags we're looking for is
        hard-coded.

        @param raw {boolean} indicates if these are raw HTML blocks in
            the original source. It makes a difference in "safe" mode.
        """
        if '<' not in text:
            return text

        # Pass `raw` value into our calls to self._hash_html_block_sub.
        hash_html_block_sub = _curry(self._hash_html_block_sub, raw=raw)

        # First, look for nested blocks, e.g.:
        #   <div>
        #       <div>
        #       tags for inner block must be indented.
        #       </div>
        #   </div>
        #
        # The outermost tags must start at the left margin for this to match, and
        # the inner nested divs must be indented.
        # We need to do this before the next, more liberal match, because the next
        # match will start at the first `<div>` and stop at the first `</div>`.
        text = self._strict_tag_block_re.sub(hash_html_block_sub, text)

        # Now match more liberally, simply from `\n<tag>` to `</tag>\n`
        text = self._liberal_tag_block_re.sub(hash_html_block_sub, text)

        # Special case just for <hr />. It was easier to make a special
        # case than to make the other regex more complicated.   
        if "<hr" in text:
            _hr_tag_re = _hr_tag_re_from_tab_width(self.tab_width)
            text = _hr_tag_re.sub(hash_html_block_sub, text)

        # Special case for standalone HTML comments:
        if "<!--" in text:
            start = 0
            while True:
                # Delimiters for next comment block.
                try:
                    start_idx = text.index("<!--", start)
                except ValueError, ex:
                    break
                try:
                    end_idx = text.index("-->", start_idx) + 3
                except ValueError, ex:
                    break

                # Start position for next comment block search.
                start = end_idx

                # Validate whitespace before comment.
                if start_idx:
                    # - Up to `tab_width - 1` spaces before start_idx.
                    for i in range(self.tab_width - 1):
                        if text[start_idx - 1] != ' ':
                            break
                        start_idx -= 1
                        if start_idx == 0:
                            break
                    # - Must be preceded by 2 newlines or hit the start of
                    #   the document.
                    if start_idx == 0:
                        pass
                    elif start_idx == 1 and text[0] == '\n':
                        start_idx = 0  # to match minute detail of Markdown.pl regex
                    elif text[start_idx-2:start_idx] == '\n\n':
                        pass
                    else:
                        break

                # Validate whitespace after comment.
                # - Any number of spaces and tabs.
                while end_idx < len(text):
                    if text[end_idx] not in ' \t':
                        break
                    end_idx += 1
                # - Must be following by 2 newlines or hit end of text.
                if text[end_idx:end_idx+2] not in ('', '\n', '\n\n'):
                    continue

                # Escape and hash (must match `_hash_html_block_sub`).
                html = text[start_idx:end_idx]
                if raw and self.safe_mode:
                    html = self._sanitize_html(html)
                key = _hash_text(html)
                self.html_blocks[key] = html
                text = text[:start_idx] + "\n\n" + key + "\n\n" + text[end_idx:]

        if "xml" in self.extras:
            # Treat XML processing instructions and namespaced one-liner
            # tags as if they were block HTML tags. E.g., if standalone
            # (i.e. are their own paragraph), the following do not get 
            # wrapped in a <p> tag:
            #    <?foo bar?>
            #
            #    <xi:include xmlns:xi="http://www.w3.org/2001/XInclude" href="chapter_1.md"/>
            _xml_oneliner_re = _xml_oneliner_re_from_tab_width(self.tab_width)
            text = _xml_oneliner_re.sub(hash_html_block_sub, text)

        return text

    def _strip_link_definitions(self, text):
        # Strips link definitions from text, stores the URLs and titles in
        # hash references.
        less_than_tab = self.tab_width - 1
    
        # Link defs are in the form:
        #   [id]: url "optional title"
        _link_def_re = re.compile(r"""
            ^[ ]{0,%d}\[(.+)\]: # id = \1
              [ \t]*
              \n?               # maybe *one* newline
              [ \t]*
            <?(.+?)>?           # url = \2
              [ \t]*
            (?:
                \n?             # maybe one newline
                [ \t]*
                (?<=\s)         # lookbehind for whitespace
                ['"(]
                ([^\n]*)        # title = \3
                ['")]
                [ \t]*
            )?  # title is optional
            (?:\n+|\Z)
            """ % less_than_tab, re.X | re.M | re.U)
        return _link_def_re.sub(self._extract_link_def_sub, text)

    def _extract_link_def_sub(self, match):
        id, url, title = match.groups()
        key = id.lower()    # Link IDs are case-insensitive
        self.urls[key] = self._encode_amps_and_angles(url)
        if title:
            self.titles[key] = title.replace('"', '&quot;')
        return ""

    def _extract_footnote_def_sub(self, match):
        id, text = match.groups()
        text = _dedent(text, skip_first_line=not text.startswith('\n')).strip()
        normed_id = re.sub(r'\W', '-', id)
        # Ensure footnote text ends with a couple newlines (for some
        # block gamut matches).
        self.footnotes[normed_id] = text + "\n\n"
        return ""

    def _strip_footnote_definitions(self, text):
        """A footnote definition looks like this:

            [^note-id]: Text of the note.

                May include one or more indented paragraphs.

        Where,
        - The 'note-id' can be pretty much anything, though typically it
          is the number of the footnote.
        - The first paragraph may start on the next line, like so:
            
            [^note-id]:
                Text of the note.
        """
        less_than_tab = self.tab_width - 1
        footnote_def_re = re.compile(r'''
            ^[ ]{0,%d}\[\^(.+)\]:   # id = \1
            [ \t]*
            (                       # footnote text = \2
              # First line need not start with the spaces.
              (?:\s*.*\n+)
              (?:
                (?:[ ]{%d} | \t)  # Subsequent lines must be indented.
                .*\n+
              )*
            )
            # Lookahead for non-space at line-start, or end of doc.
            (?:(?=^[ ]{0,%d}\S)|\Z)
            ''' % (less_than_tab, self.tab_width, self.tab_width),
            re.X | re.M)
        return footnote_def_re.sub(self._extract_footnote_def_sub, text)


    _hr_res = [
        re.compile(r"^[ ]{0,2}([ ]?\*[ ]?){3,}[ \t]*$", re.M),
        re.compile(r"^[ ]{0,2}([ ]?\-[ ]?){3,}[ \t]*$", re.M),
        re.compile(r"^[ ]{0,2}([ ]?\_[ ]?){3,}[ \t]*$", re.M),
    ]

    def _run_block_gamut(self, text):
        # These are all the transformations that form block-level
        # tags like paragraphs, headers, and list items.

        text = self._do_headers(text)

        # Do Horizontal Rules:
        hr = "\n<hr"+self.empty_element_suffix+"\n"
        for hr_re in self._hr_res:
            text = hr_re.sub(hr, text)

        text = self._do_lists(text)

        if "pyshell" in self.extras:
            text = self._prepare_pyshell_blocks(text)

        text = self._do_code_blocks(text)

        text = self._do_block_quotes(text)

        # We already ran _HashHTMLBlocks() before, in Markdown(), but that
        # was to escape raw HTML in the original Markdown source. This time,
        # we're escaping the markup we've just created, so that we don't wrap
        # <p> tags around block-level tags.
        text = self._hash_html_blocks(text)

        text = self._form_paragraphs(text)

        return text

    def _pyshell_block_sub(self, match):
        lines = match.group(0).splitlines(0)
        _dedentlines(lines)
        indent = ' ' * self.tab_width
        s = ('\n' # separate from possible cuddled paragraph
             + indent + ('\n'+indent).join(lines)
             + '\n\n')
        return s
        
    def _prepare_pyshell_blocks(self, text):
        """Ensure that Python interactive shell sessions are put in
        code blocks -- even if not properly indented.
        """
        if ">>>" not in text:
            return text

        less_than_tab = self.tab_width - 1
        _pyshell_block_re = re.compile(r"""
            ^([ ]{0,%d})>>>[ ].*\n   # first line
            ^(\1.*\S+.*\n)*         # any number of subsequent lines
            ^\n                     # ends with a blank line
            """ % less_than_tab, re.M | re.X)

        return _pyshell_block_re.sub(self._pyshell_block_sub, text)

    def _run_span_gamut(self, text):
        # These are all the transformations that occur *within* block-level
        # tags like paragraphs, headers, and list items.
    
        text = self._do_code_spans(text)
    
        text = self._escape_special_chars(text)
    
        # Process anchor and image tags.
        text = self._do_links(text)
    
        # Make links out of things like `<http://example.com/>`
        # Must come after _do_links(), because you can use < and >
        # delimiters in inline links like [this](<url>).
        text = self._do_auto_links(text)

        if "link-patterns" in self.extras:
            text = self._do_link_patterns(text)
    
        text = self._encode_amps_and_angles(text)
    
        text = self._do_italics_and_bold(text)
    
        # Do hard breaks:
        text = re.sub(r" {2,}\n", " <br%s\n" % self.empty_element_suffix, text)
    
        return text

    # "Sorta" because auto-links are identified as "tag" tokens.
    _sorta_html_tokenize_re = re.compile(r"""
        (
            # tag
            </?         
            (?:\w+)                                     # tag name
            (?:\s+(?:[\w-]+:)?[\w-]+=(?:".*?"|'.*?'))*  # attributes
            \s*/?>
            |
            # auto-link (e.g., <http://www.activestate.com/>)
            <\w+[^>]*>
            |
            <!--.*?-->      # comment
            |
            <\?.*?\?>       # processing instruction
        )
        """, re.X)
    
    def _escape_special_chars(self, text):
        # Python markdown note: the HTML tokenization here differs from
        # that in Markdown.pl, hence the behaviour for subtle cases can
        # differ (I believe the tokenizer here does a better job because
        # it isn't susceptible to unmatched '<' and '>' in HTML tags).
        # Note, however, that '>' is not allowed in an auto-link URL
        # here.
        escaped = []
        is_html_markup = False
        for token in self._sorta_html_tokenize_re.split(text):
            if is_html_markup:
                # Within tags/HTML-comments/auto-links, encode * and _
                # so they don't conflict with their use in Markdown for
                # italics and strong.  We're replacing each such
                # character with its corresponding MD5 checksum value;
                # this is likely overkill, but it should prevent us from
                # colliding with the escape values by accident.
                escaped.append(token.replace('*', g_escape_table['*'])
                                    .replace('_', g_escape_table['_']))
            else:
                escaped.append(self._encode_backslash_escapes(token))
            is_html_markup = not is_html_markup
        return ''.join(escaped)

    def _hash_html_spans(self, text):
        # Used for safe_mode.

        def _is_auto_link(s):
            if ':' in s and self._auto_link_re.match(s):
                return True
            elif '@' in s and self._auto_email_link_re.match(s):
                return True
            return False

        tokens = []
        is_html_markup = False
        for token in self._sorta_html_tokenize_re.split(text):
            if is_html_markup and not _is_auto_link(token):
                sanitized = self._sanitize_html(token)
                key = _hash_text(sanitized)
                self.html_spans[key] = sanitized
                tokens.append(key)
            else:
                tokens.append(token)
            is_html_markup = not is_html_markup
        return ''.join(tokens)

    def _unhash_html_spans(self, text):
        for key, sanitized in self.html_spans.items():
            text = text.replace(key, sanitized)
        return text

    def _sanitize_html(self, s):
        if self.safe_mode == "replace":
            return self.html_removed_text
        elif self.safe_mode == "escape":
            replacements = [
                ('&', '&amp;'),
                ('<', '&lt;'),
                ('>', '&gt;'),
            ]
            for before, after in replacements:
                s = s.replace(before, after)
            return s
        else:
            raise MarkdownError("invalid value for 'safe_mode': %r (must be "
                                "'escape' or 'replace')" % self.safe_mode)

    _tail_of_inline_link_re = re.compile(r'''
          # Match tail of: [text](/url/) or [text](/url/ "title")
          \(            # literal paren
            [ \t]*
            (?P<url>            # \1
                <.*?>
                |
                .*?
            )
            [ \t]*
            (                   # \2
              (['"])            # quote char = \3
              (?P<title>.*?)
              \3                # matching quote
            )?                  # title is optional
          \)
        ''', re.X | re.S)
    _tail_of_reference_link_re = re.compile(r'''
          # Match tail of: [text][id]
          [ ]?          # one optional space
          (?:\n[ ]*)?   # one optional newline followed by spaces
          \[
            (?P<id>.*?)
          \]
        ''', re.X | re.S)

    def _do_links(self, text):
        """Turn Markdown link shortcuts into XHTML <a> and <img> tags.

        This is a combination of Markdown.pl's _DoAnchors() and
        _DoImages(). They are done together because that simplified the
        approach. It was necessary to use a different approach than
        Markdown.pl because of the lack of atomic matching support in
        Python's regex engine used in $g_nested_brackets.
        """
        MAX_LINK_TEXT_SENTINEL = 3000  # markdown2 issue 24

        # `anchor_allowed_pos` is used to support img links inside
        # anchors, but not anchors inside anchors. An anchor's start
        # pos must be `>= anchor_allowed_pos`.
        anchor_allowed_pos = 0

        curr_pos = 0
        while True: # Handle the next link.
            # The next '[' is the start of:
            # - an inline anchor:   [text](url "title")
            # - a reference anchor: [text][id]
            # - an inline img:      ![text](url "title")
            # - a reference img:    ![text][id]
            # - a footnote ref:     [^id]
            #   (Only if 'footnotes' extra enabled)
            # - a footnote defn:    [^id]: ...
            #   (Only if 'footnotes' extra enabled) These have already
            #   been stripped in _strip_footnote_definitions() so no
            #   need to watch for them.
            # - a link definition:  [id]: url "title"
            #   These have already been stripped in
            #   _strip_link_definitions() so no need to watch for them.
            # - not markup:         [...anything else...
            try:
                start_idx = text.index('[', curr_pos)
            except ValueError:
                break
            text_length = len(text)

            # Find the matching closing ']'.
            # Markdown.pl allows *matching* brackets in link text so we
            # will here too. Markdown.pl *doesn't* currently allow
            # matching brackets in img alt text -- we'll differ in that
            # regard.
            bracket_depth = 0
            for p in range(start_idx+1, min(start_idx+MAX_LINK_TEXT_SENTINEL, 
                                            text_length)):
                ch = text[p]
                if ch == ']':
                    bracket_depth -= 1
                    if bracket_depth < 0:
                        break
                elif ch == '[':
                    bracket_depth += 1
            else:
                # Closing bracket not found within sentinel length.
                # This isn't markup.
                curr_pos = start_idx + 1
                continue
            link_text = text[start_idx+1:p]

            # Possibly a footnote ref?
            if "footnotes" in self.extras and link_text.startswith("^"):
                normed_id = re.sub(r'\W', '-', link_text[1:])
                if normed_id in self.footnotes:
                    self.footnote_ids.append(normed_id)
                    result = '<sup class="footnote-ref" id="fnref-%s">' \
                             '<a href="#fn-%s">%s</a></sup>' \
                             % (normed_id, normed_id, len(self.footnote_ids))
                    text = text[:start_idx] + result + text[p+1:]
                else:
                    # This id isn't defined, leave the markup alone.
                    curr_pos = p+1
                continue

            # Now determine what this is by the remainder.
            p += 1
            if p == text_length:
                return text

            # Inline anchor or img?
            if text[p] == '(': # attempt at perf improvement
                match = self._tail_of_inline_link_re.match(text, p)
                if match:
                    # Handle an inline anchor or img.
                    is_img = start_idx > 0 and text[start_idx-1] == "!"
                    if is_img:
                        start_idx -= 1

                    url, title = match.group("url"), match.group("title")
                    if url and url[0] == '<':
                        url = url[1:-1]  # '<url>' -> 'url'
                    # We've got to encode these to avoid conflicting
                    # with italics/bold.
                    url = url.replace('*', g_escape_table['*']) \
                             .replace('_', g_escape_table['_'])
                    if title:
                        title_str = ' title="%s"' \
                            % title.replace('*', g_escape_table['*']) \
                                   .replace('_', g_escape_table['_']) \
                                   .replace('"', '&quot;')
                    else:
                        title_str = ''
                    if is_img:
                        result = '<img src="%s" alt="%s"%s%s' \
                            % (url.replace('"', '&quot;'),
                               link_text.replace('"', '&quot;'),
                               title_str, self.empty_element_suffix)
                        curr_pos = start_idx + len(result)
                        text = text[:start_idx] + result + text[match.end():]
                    elif start_idx >= anchor_allowed_pos:
                        result_head = '<a href="%s"%s>' % (url, title_str)
                        result = '%s%s</a>' % (result_head, link_text)
                        # <img> allowed from curr_pos on, <a> from
                        # anchor_allowed_pos on.
                        curr_pos = start_idx + len(result_head)
                        anchor_allowed_pos = start_idx + len(result)
                        text = text[:start_idx] + result + text[match.end():]
                    else:
                        # Anchor not allowed here.
                        curr_pos = start_idx + 1
                    continue

            # Reference anchor or img?
            else:
                match = self._tail_of_reference_link_re.match(text, p)
                if match:
                    # Handle a reference-style anchor or img.
                    is_img = start_idx > 0 and text[start_idx-1] == "!"
                    if is_img:
                        start_idx -= 1
                    link_id = match.group("id").lower()
                    if not link_id:
                        link_id = link_text.lower()  # for links like [this][]
                    if link_id in self.urls:
                        url = self.urls[link_id]
                        # We've got to encode these to avoid conflicting
                        # with italics/bold.
                        url = url.replace('*', g_escape_table['*']) \
                                 .replace('_', g_escape_table['_'])
                        title = self.titles.get(link_id)
                        if title:
                            title = title.replace('*', g_escape_table['*']) \
                                         .replace('_', g_escape_table['_'])
                            title_str = ' title="%s"' % title
                        else:
                            title_str = ''
                        if is_img:
                            result = '<img src="%s" alt="%s"%s%s' \
                                % (url.replace('"', '&quot;'),
                                   link_text.replace('"', '&quot;'),
                                   title_str, self.empty_element_suffix)
                            curr_pos = start_idx + len(result)
                            text = text[:start_idx] + result + text[match.end():]
                        elif start_idx >= anchor_allowed_pos:
                            result = '<a href="%s"%s>%s</a>' \
                                % (url, title_str, link_text)
                            result_head = '<a href="%s"%s>' % (url, title_str)
                            result = '%s%s</a>' % (result_head, link_text)
                            # <img> allowed from curr_pos on, <a> from
                            # anchor_allowed_pos on.
                            curr_pos = start_idx + len(result_head)
                            anchor_allowed_pos = start_idx + len(result)
                            text = text[:start_idx] + result + text[match.end():]
                        else:
                            # Anchor not allowed here.
                            curr_pos = start_idx + 1
                    else:
                        # This id isn't defined, leave the markup alone.
                        curr_pos = match.end()
                    continue

            # Otherwise, it isn't markup.
            curr_pos = start_idx + 1

        return text 


    _setext_h_re = re.compile(r'^(.+)[ \t]*\n(=+|-+)[ \t]*\n+', re.M)
    def _setext_h_sub(self, match):
        n = {"=": 1, "-": 2}[match.group(2)[0]]
        demote_headers = self.extras.get("demote-headers")
        if demote_headers:
            n = min(n + demote_headers, 6)
        return "<h%d>%s</h%d>\n\n" \
               % (n, self._run_span_gamut(match.group(1)), n)

    _atx_h_re = re.compile(r'''
        ^(\#{1,6})  # \1 = string of #'s
        [ \t]*
        (.+?)       # \2 = Header text
        [ \t]*
        (?<!\\)     # ensure not an escaped trailing '#'
        \#*         # optional closing #'s (not counted)
        \n+
        ''', re.X | re.M)
    def _atx_h_sub(self, match):
        n = len(match.group(1))
        demote_headers = self.extras.get("demote-headers")
        if demote_headers:
            n = min(n + demote_headers, 6)
        return "<h%d>%s</h%d>\n\n" \
               % (n, self._run_span_gamut(match.group(2)), n)

    def _do_headers(self, text):
        # Setext-style headers:
        #     Header 1
        #     ========
        #  
        #     Header 2
        #     --------
        text = self._setext_h_re.sub(self._setext_h_sub, text)

        # atx-style headers:
        #   # Header 1
        #   ## Header 2
        #   ## Header 2 with closing hashes ##
        #   ...
        #   ###### Header 6
        text = self._atx_h_re.sub(self._atx_h_sub, text)

        return text


    _marker_ul_chars  = '*+-'
    _marker_any = r'(?:[%s]|\d+\.)' % _marker_ul_chars
    _marker_ul = '(?:[%s])' % _marker_ul_chars
    _marker_ol = r'(?:\d+\.)'

    def _list_sub(self, match):
        lst = match.group(1)
        lst_type = match.group(3) in self._marker_ul_chars and "ul" or "ol"
        result = self._process_list_items(lst)
        if self.list_level:
            return "<%s>\n%s</%s>\n" % (lst_type, result, lst_type)
        else:
            return "<%s>\n%s</%s>\n\n" % (lst_type, result, lst_type)

    def _do_lists(self, text):
        # Form HTML ordered (numbered) and unordered (bulleted) lists.

        for marker_pat in (self._marker_ul, self._marker_ol):
            # Re-usable pattern to match any entire ul or ol list:
            less_than_tab = self.tab_width - 1
            whole_list = r'''
                (                   # \1 = whole list
                  (                 # \2
                    [ ]{0,%d}
                    (%s)            # \3 = first list item marker
                    [ \t]+
                  )
                  (?:.+?)
                  (                 # \4
                      \Z
                    |
                      \n{2,}
                      (?=\S)
                      (?!           # Negative lookahead for another list item marker
                        [ \t]*
                        %s[ \t]+
                      )
                  )
                )
            ''' % (less_than_tab, marker_pat, marker_pat)
        
            # We use a different prefix before nested lists than top-level lists.
            # See extended comment in _process_list_items().
            #
            # Note: There's a bit of duplication here. My original implementation
            # created a scalar regex pattern as the conditional result of the test on
            # $g_list_level, and then only ran the $text =~ s{...}{...}egmx
            # substitution once, using the scalar as the pattern. This worked,
            # everywhere except when running under MT on my hosting account at Pair
            # Networks. There, this caused all rebuilds to be killed by the reaper (or
            # perhaps they crashed, but that seems incredibly unlikely given that the
            # same script on the same server ran fine *except* under MT. I've spent
            # more time trying to figure out why this is happening than I'd like to
            # admit. My only guess, backed up by the fact that this workaround works,
            # is that Perl optimizes the substition when it can figure out that the
            # pattern will never change, and when this optimization isn't on, we run
            # afoul of the reaper. Thus, the slightly redundant code to that uses two
            # static s/// patterns rather than one conditional pattern.

            if self.list_level:
                sub_list_re = re.compile("^"+whole_list, re.X | re.M | re.S)
                text = sub_list_re.sub(self._list_sub, text)
            else:
                list_re = re.compile(r"(?:(?<=\n\n)|\A\n?)"+whole_list,
                                     re.X | re.M | re.S)
                text = list_re.sub(self._list_sub, text)

        return text
    
    _list_item_re = re.compile(r'''
        (\n)?               # leading line = \1
        (^[ \t]*)           # leading whitespace = \2
        (%s) [ \t]+         # list marker = \3
        ((?:.+?)            # list item text = \4
         (\n{1,2}))         # eols = \5
        (?= \n* (\Z | \2 (%s) [ \t]+))
        ''' % (_marker_any, _marker_any),
        re.M | re.X | re.S)

    _last_li_endswith_two_eols = False
    def _list_item_sub(self, match):
        item = match.group(4)
        leading_line = match.group(1)
        leading_space = match.group(2)
        if leading_line or "\n\n" in item or self._last_li_endswith_two_eols:
            item = self._run_block_gamut(self._outdent(item))
        else:
            # Recursion for sub-lists:
            item = self._do_lists(self._outdent(item))
            if item.endswith('\n'):
                item = item[:-1]
            item = self._run_span_gamut(item)
        self._last_li_endswith_two_eols = (len(match.group(5)) == 2)
        return "<li>%s</li>\n" % item

    def _process_list_items(self, list_str):
        # Process the contents of a single ordered or unordered list,
        # splitting it into individual list items.
    
        # The $g_list_level global keeps track of when we're inside a list.
        # Each time we enter a list, we increment it; when we leave a list,
        # we decrement. If it's zero, we're not in a list anymore.
        #
        # We do this because when we're not inside a list, we want to treat
        # something like this:
        #
        #       I recommend upgrading to version
        #       8. Oops, now this line is treated
        #       as a sub-list.
        #
        # As a single paragraph, despite the fact that the second line starts
        # with a digit-period-space sequence.
        #
        # Whereas when we're inside a list (or sub-list), that line will be
        # treated as the start of a sub-list. What a kludge, huh? This is
        # an aspect of Markdown's syntax that's hard to parse perfectly
        # without resorting to mind-reading. Perhaps the solution is to
        # change the syntax rules such that sub-lists must start with a
        # starting cardinal number; e.g. "1." or "a.".
        self.list_level += 1
        self._last_li_endswith_two_eols = False
        list_str = list_str.rstrip('\n') + '\n'
        list_str = self._list_item_re.sub(self._list_item_sub, list_str)
        self.list_level -= 1
        return list_str

    def _get_pygments_lexer(self, lexer_name):
        try:
            from pygments import lexers, util
        except ImportError:
            return None
        try:
            return lexers.get_lexer_by_name(lexer_name)
        except util.ClassNotFound:
            return None

    def _color_with_pygments(self, codeblock, lexer, **formatter_opts):
        import pygments
        import pygments.formatters

        class HtmlCodeFormatter(pygments.formatters.HtmlFormatter):
            def _wrap_code(self, inner):
                """A function for use in a Pygments Formatter which
                wraps in <code> tags.
                """
                yield 0, "<code>"
                for tup in inner:
                    yield tup 
                yield 0, "</code>"

            def wrap(self, source, outfile):
                """Return the source with a code, pre, and div."""
                return self._wrap_div(self._wrap_pre(self._wrap_code(source)))

        formatter = HtmlCodeFormatter(cssclass="codehilite", **formatter_opts)
        return pygments.highlight(codeblock, lexer, formatter)

    def _code_block_sub(self, match):
        codeblock = match.group(1)
        codeblock = self._outdent(codeblock)
        codeblock = self._detab(codeblock)
        codeblock = codeblock.lstrip('\n')  # trim leading newlines
        codeblock = codeblock.rstrip()      # trim trailing whitespace

        if "code-color" in self.extras and codeblock.startswith(":::"):
            lexer_name, rest = codeblock.split('\n', 1)
            lexer_name = lexer_name[3:].strip()
            lexer = self._get_pygments_lexer(lexer_name)
            codeblock = rest.lstrip("\n")   # Remove lexer declaration line.
            if lexer:
                formatter_opts = self.extras['code-color'] or {}
                colored = self._color_with_pygments(codeblock, lexer,
                                                    **formatter_opts)
                return "\n\n%s\n\n" % colored

        codeblock = self._encode_code(codeblock)
        return "\n\n<pre><code>%s\n</code></pre>\n\n" % codeblock

    def _do_code_blocks(self, text):
        """Process Markdown `<pre><code>` blocks."""
        code_block_re = re.compile(r'''
            (?:\n\n|\A)
      …