lxml.html.diff does HTML comparisons.  These are word-based comparisons.

First, a handy function for normalizing whitespace and doing word wrapping::

    >>> import re, textwrap
    >>> def pwrapped(text):
    ...     text = re.sub(r'[ \n\t\r]+', ' ', text)
    ...     text = textwrap.fill(text)
    ...     print(text)
    >>> def pdiff(text1, text2):
    ...     pwrapped(htmldiff(text1, text2))

Example::

    >>> from lxml.html.diff import htmldiff, html_annotate
    >>> html1 = '<p>This is some test text with some changes and some same stuff</p>'
    >>> html2 = '''<p>This is some test textual writing with some changed stuff 
    ... and some same stuff</p>'''
    >>> pdiff(html1, html2)
    <p>This is some test <ins>textual writing with some changed
    stuff</ins> <del>text with some changes</del> and some same stuff</p>

Style tags are largely ignored in terms of differences, though markup is not eliminated::

    >>> html1 = '<p>Hi <i>you guys</i></p>'
    >>> html2 = '<p>Hi <i>you</i> guys</p>'
    >>> pdiff(html1, html2)
    <p>Hi <i>you</i> guys</p>
    >>> pdiff('text', '<p>text</p>')
    <p>text</p>
    >>> pdiff('<i>Hi guys</i> !!', '<i>Hi guy</i> !!')
    <i>Hi <ins>guy</ins> <del>guys</del> </i> !!
    >>> pdiff('H<i>i</i>', 'Hi')
    <ins>Hi</ins> <del>H<i>i</i></del>
    >>> pdiff('<i>A B</i> C', '<i>A</i> C')
    <i>A <del>B</del> </i> C
    >>> pdiff('<i>A B</i> C', '<i>B</i> C')
    <i> <del>A</del> B</i> C
    >>> pdiff('<p></p>', '<p></p>')
    <p></p>
    >>> pdiff('<p>Hi</p>', '<p>Bye</p>')
    <p><ins>Bye</ins></p> <p><del>Hi</del></p>
    >>> pdiff('<p>Hi Guy</p>', '<p>Bye Guy</p>')
    <p> <ins>Bye</ins> <del>Hi</del> Guy</p>
    >>> pdiff('<p>Hey there</p>', '')
    <ins></ins> <p><del>Hey there</del></p>

Whitespace is ignored, as it's not meaningful in HTML::

    >>> pdiff('<div>Hi\n\nguys</div>', '<div>Hi guy</div>')
    <div>Hi <ins>guy</ins> <del>guys</del> </div>

Movement between paragraphs is ignored, as tag-based changes are generally ignored::
    >>> 
    >>> pdiff('<p>Hello</p><p>World</p>', '<p>Hello World</p>')
    <p>Hello World</p>

As a special case, changing the href of a link is displayed, and
images are treated like words:

    >>> pdiff('<a href="http://yahoo.com">search</a>', '<a href="http://google.com">search</a>')
    <a href="http://google.com">search <ins> Link: http://google.com</ins>
    <del> Link: http://yahoo.com</del> </a>
    >>> pdiff('<p>Print this <img src="print.gif"></p>', '<p>Print this</p>')
    <p>Print this <del><img src="print.gif"></del> </p>
    >>> pdiff('<a href="http://yahoo.com">search</a>', '<a href="http://yahoo.com">search</a>')
    <a href="http://yahoo.com">search</a>

Images may sometimes not have 'src' attributes:

    >>> pdiff('<img src="tease"><img><img src="test">', '<img><img src="toast">')
    <del><img src="tease"></del> <img><ins><img src="toast"></ins>
    <del><img src="test"></del>

A test of empty elements:

    >>> pdiff('some <br> text', 'some <br> test')
    some <ins><br> test</ins> <del><br> text</del>

The sixteen combinations::

First "insert start" (del start/middle/end/none):

    >>> pdiff('<b>A B C</b>', '<b>D B C</b')
    <b> <ins>D</ins> <del>A</del> B C</b>
    >>> pdiff('<b>A B C</b>', '<b>D A C</b>')
    <b> <ins>D</ins> A <del>B</del> C</b>
    >>> pdiff('<b>A B C</b>', '<b>D A B</b>')
    <b> <ins>D</ins> A B <del>C</del> </b>
    >>> pdiff('<b>A B C</b>', '<b>D A B C</b>')
    <b> <ins>D</ins> A B C</b>

Next, "insert middle" (del start/middle/end/none):

    >>> pdiff('<b>A B C</b>', '<b>D B C</b>')
    <b> <ins>D</ins> <del>A</del> B C</b>
    >>> pdiff('<b>A B C</b>', '<b>A D C</b>')
    <b>A <ins>D</ins> <del>B</del> C</b>
    >>> pdiff('<b>A B C</b>', '<b>A D B</b>')
    <b>A <ins>D</ins> B <del>C</del> </b>

This one case hits the threshold of our insensitive matching:

    >>> pdiff('<b>A B C</b>', '<b>A D B C</b>')
    <b> <ins>A D</ins> <del>A</del> B C</b>


Then "insert end" (del start/middle/end/none):

    >>> pdiff('<b>A B C</b>', '<b>B C D</b>')
    <b> <del>A</del> B C <ins>D</ins> </b>
    >>> pdiff('<b>A B C</b>', '<b>A C D</b>')
    <b>A <del>B</del> C <ins>D</ins> </b>
    >>> pdiff('<b>A B C</b>', '<b>A B D</b>')
    <b>A B <ins>D</ins> <del>C</del> </b>
    >>> pdiff('<b>A B C</b>', '<b>A B C D</b>')
    <b>A B C <ins>D</ins> </b>

Then no insert (del start/middle/end):

    >>> pdiff('<b>A B C</b>', '<b>B C</b>')
    <b> <del>A</del> B C</b>
    >>> pdiff('<b>A B C</b>', '<b>A C</b>')
    <b>A <del>B</del> C</b>
    >>> pdiff('<b>A B C</b>', '<b>A B</b>')
    <b>A B <del>C</del> </b>

    >>> pdiff('<b>A B</b> C', '<b>A B</b>')
    <b>A B</b> <del>C</del>
    >>> pdiff('<b>A B</b> <b>C</b>', '<b>A B</b>')
    <b>A B</b> <del><b>C</b></del>
    >>> pdiff('A <p><b>hey there</b> <i>how are you?</i></p>', 'A')
    A <p><del><b>hey there</b> <i>how are you?</i></del></p>
    
Testing a larger document, to make sure there are not weird
unnecessary parallels found:

    >>> pdiff('''
    ... <p>This is a test document with many words in it that goes on
    ... for a while and doesn't have anything do to with the next
    ... document that we match this against</p>''', '''
    ... <p>This is another document with few similarities to the preceding
    ... one, but enough that it may have overlap that could turn into
    ... a confusing series of deletes and inserts.
    ... </p>''')
    <p><ins>This is another document with few similarities to the
    preceding one, but enough that it may have overlap that could turn
    into a confusing series of deletes and inserts. </ins></p>
    <p><del>This is a test document with many words in it that goes on for
    a while and doesn't have anything do to with the next document that we
    match this against</del></p>



Annotation of content can also be done, where every bit of content is
marked up with information about where it came from.

First, some setup; note that html_annotate is called with a sequence
of documents and the annotation associated with that document.  We'll
just use indexes, but you could use author or timestamp information.

    >>> def markup(text, annotation):
    ...     return '<span version="%s">%s</span>' % (annotation, text)
    >>> def panno(*docs):
    ...     pwrapped(html_annotate([(doc, index) for index, doc in enumerate(docs)],
    ...                            markup=markup))

Now, a sequence of documents:

    >>> panno('Hello cruel world', 'Hi cruel world', 'Hi world')
    <span version="1">Hi</span> <span version="0">world</span>
    >>> panno('A similar document', 'A similar document',
    ...       'A similar document here')
    <span version="0">A similar document</span> <span
    version="2">here</span>
    >>> panno('<p>P1 para</p><p>P2 para</p>', '<p>P1 para</p><p>P3 foo</p>')
    <p><span version="0">P1 para</span></p><p><span version="1">P3
    foo</span></p>
    >>> panno('Hello<p>There World</p>','Hello<p>There Town</p>')
    <span version="0">Hello</span><p><span version="0">There</span> <span
    version="1">Town</span></p>
    >>> panno('<p>Hello</p>There World','<p>Hello</p>There Town')
    <p><span version="0">Hello</span></p><span version="0">There</span>
    <span version="1">Town</span>
    >>> panno('<p>Hello</p><p>There World</p>','<p>Hello</p><p>There Town</p>')
    <p><span version="0">Hello</span></p><p><span version="0">There</span>
    <span version="1">Town</span></p>
    >>> panno('<p>Hi <img src="/foo"> You</p>',
    ...       '<p>Hi You</p>',
    ...       '<p>Hi You <img src="/bar"></p>')
    <p><span version="0">Hi You</span> <span version="2"><img
    src="/bar"></span></p>
    >>> panno('<p><a href="/foo">Hey</a></p>',
    ...       '<p><a href="/bar">Hey</a></p>')
    <p><a href="/bar"><span version="0">Hey</span></a></p>
    >>> panno('<p><a href="/foo">Hey You</a></p>',
    ...       '<p><a href="/foo">Hey Guy</a></p>')
    <p><a href="/foo"><span version="0">Hey</span> <span
    version="1">Guy</span></a></p>

Internals
---------


Some utility functions::

    >>> from lxml.html.diff import fixup_ins_del_tags, split_unbalanced
    >>> def pfixup(text):
    ...     print(fixup_ins_del_tags(text).strip())
    >>> pfixup('<ins><p>some text <b>and more text</b> and more</p></ins>')
    <p><ins>some text <b>and more text</b> and more</ins></p>
    >>> pfixup('<p><ins>Hi!</ins> you</p>')
    <p><ins>Hi!</ins> you</p>
    >>> pfixup('<div>Some text <ins>and <p>more text</p></ins> </div>')
    <div>Some text <ins>and </ins><p><ins>more text</ins></p> </div>
    >>> pfixup('''
    ...    <ins><table><tr><td>One table</td><td>More stuff</td></tr></table></ins>''')
    <table><tr><td><ins>One table</ins></td><td><ins>More stuff</ins></td></tr></table>


Testing split_unbalanced::

    >>> split_unbalanced(['<a href="blah">', 'hey', '</a>'])
    ([], ['<a href="blah">', 'hey', '</a>'], [])
    >>> split_unbalanced(['<a href="blah">', 'hey'])
    (['<a href="blah">'], ['hey'], [])
    >>> split_unbalanced(['Hey', '</i>', 'You', '</b>'])
    ([], ['Hey', 'You'], ['</i>', '</b>'])
    >>> split_unbalanced(['So', '</i>', 'Hi', '<b>', 'There', '</b>'])
    ([], ['So', 'Hi', '<b>', 'There', '</b>'], ['</i>'])
    >>> split_unbalanced(['So', '</i>', 'Hi', '<b>', 'There'])
    (['<b>'], ['So', 'Hi', 'There'], ['</i>'])