I've become a pretty big fan of XML::LibXML, but I haven't tried it on not-well-formed-XML content before. With the recover option, how much of the funniness in typical HTML can it handle (missing end tags, unquoted attribute values, singleton tags like <br>)?