\.\.(?P<b_blob_id>[0-9A-Fa-f]+)[ ]?(?P<b_mode>.+)?(?:\n|$))?

        (?:^(?P<bin_patch>GIT[ ]binary[ ]patch)(?:\n|$))?

        (?:^---[ ](a/(?P<a_file>.+?)|/dev/null)\t?(?:\n|$))?

        (?:^\+\+\+[ ](b/(?P<b_file>.+?)|/dev/null)\t?(?:\n|$))?

    """, re.VERBOSE | re.MULTILINE)

    _hg_header_re = re.compile(r"""

        ^diff[ ]--git

            [ ]a/(?P<a_path>.+?)[ ]b/(?P<b_path>.+?)\n

        (?:^old[ ]mode[ ](?P<old_mode>\d+)\n

           ^new[ ]mode[ ](?P<new_mode>\d+)(?:\n|$))?

        (?:^similarity[ ]index[ ](?P<similarity_index>\d+)%(?:\n|$))?

        (?:^rename[ ]from[ ](?P<rename_from>.+)\n

           ^rename[ ]to[ ](?P<rename_to>.+)(?:\n|$))?

        (?:^copy[ ]from[ ](?P<copy_from>.+)\n

           ^copy[ ]to[ ](?P<copy_to>.+)(?:\n|$))?

        (?:^new[ ]file[ ]mode[ ](?P<new_file_mode>.+)(?:\n|$))?

        (?:^deleted[ ]file[ ]mode[ ](?P<deleted_file_mode>.+)(?:\n|$))?

        (?:^index[ ](?P<a_blob_id>[0-9A-Fa-f]+)

            \.\.(?P<b_blob_id>[0-9A-Fa-f]+)[ ]?(?P<b_mode>.+)?(?:\n|$))?

        (?:^(?P<bin_patch>GIT[ ]binary[ ]patch)(?:\n|$))?

        (?:^---[ ](a/(?P<a_file>.+?)|/dev/null)\t?(?:\n|$))?

        (?:^\+\+\+[ ](b/(?P<b_file>.+?)|/dev/null)\t?(?:\n|$))?

    """, re.VERBOSE | re.MULTILINE)

    def __init__(self, diff, vcs='hg', diff_limit=None, inline_diff=True):

        """

        :param diff:   a text in diff format

        :param vcs: type of version control hg or git

        :param diff_limit: define the size of diff that is considered "big"

            based on that parameter cut off will be triggered, set to None

            to show full diff

        """

        if not isinstance(diff, basestring):

            raise Exception('Diff must be a basestring got %s instead' % type(diff))

        self._diff = diff

        self.adds = 0

        self.removes = 0

        self.diff_limit = diff_limit

        self.limited_diff = False

        self.vcs = vcs

        self.parsed = self._parse_gitdiff(inline_diff=inline_diff)

    def _get_header(self, diff_chunk):

        """

        Parses a Git diff for a single file (header and chunks) and returns a tuple with:

        1. A dict with meta info:

            a_path, b_path, similarity_index, rename_from, rename_to,

            old_mode, new_mode, new_file_mode, deleted_file_mode,

            a_blob_id, b_blob_id, b_mode, a_file, b_file

        2. An iterator yielding lines with simple HTML markup.

        """

        match = None

        if self.vcs == 'git':

            match = self._git_header_re.match(diff_chunk)

        elif self.vcs == 'hg':

            match = self._hg_header_re.match(diff_chunk)

        if match is None:

            raise Exception('diff not recognized as valid %s diff' % self.vcs)

        meta_info = match.groupdict()

        rest = diff_chunk[match.end():]

        if rest and not rest.startswith('@') and not rest.startswith('literal ') and not rest.startswith('delta '):

            raise Exception('cannot parse %s diff header: %r followed by %r' % (self.vcs, diff_chunk[:match.end()], rest[:1000]))

        return meta_info, diff_lines

    def _parse_gitdiff(self, inline_diff):

        """Parse self._diff and return a list of dicts with meta info and chunks for each file.

        Might set limited_diff.

        Optionally, do an extra pass and to extra markup of one-liner changes.

        """

        _files = [] # list of dicts with meta info and chunks

        starts = [m.start() for m in self._diff_git_re.finditer(self._diff)]

        starts.append(len(self._diff))

        for start, end in zip(starts, starts[1:]):

            if self.diff_limit and end > self.diff_limit:

                self.limited_diff = True

                continue

            head, diff_lines = self._get_header(buffer(self._diff, start, end - start))

            op = None

            stats = {

                'added': 0,

                'deleted': 0,

                'binary': False,

                            'old_lineno':   '...',

                            'new_lineno':   '...',

                            'action':       'context',

                            'line':         line,

                        })

                        line = diff_lines.next()

                if old_line > old_end:

                    raise Exception('error parsing diff - more than %s "-" lines at -%s+%s' % (old_end, old_line, new_line))

                if new_line > new_end:

                    raise Exception('error parsing diff - more than %s "+" lines at -%s+%s' % (new_end, old_line, new_line))

        except StopIteration:

            pass

        if old_line != old_end or new_line != new_end:

            raise Exception('diff processing broken when old %s<>%s or new %s<>%s line %r' % (old_line, old_end, new_line, new_end, line))

        return chunks, added, deleted

    def stat(self):

        """

        Returns tuple of added, and removed lines for this instance

        """

        return self.adds, self.removes

# Used for inline highlighter word split, must match the substitutions in _escaper

_token_re = re.compile(r'()(&amp;|&lt;|&gt;|<u>\t</u>|<u class="cr"></u>| <i></i>|\W+?)')

def _highlight_inline_diff(old, new):

    """

    Highlight simple add/remove in two lines given as info dicts. They are

    modified in place and given markup with <del>/<ins>.

    """

    assert old['action'] == 'del'

    assert new['action'] == 'add'

    oldwords = _token_re.split(old['line'])

    newwords = _token_re.split(new['line'])

    sequence = difflib.SequenceMatcher(None, oldwords, newwords)

    oldfragments, newfragments = [], []

    for tag, i1, i2, j1, j2 in sequence.get_opcodes():

        oldfrag = ''.join(oldwords[i1:i2])

        newfrag = ''.join(newwords[j1:j2])

        if tag != 'equal':

            if oldfrag:

                oldfrag = '<del>%s</del>' % oldfrag

            if newfrag: