Hmm, I only have vague recollections about people reporting problems with lxml and I can't trivially find any supporting evidence with Google so it is possible

Hmm, I only have vague recollections about people reporting problems with lxml and I can't trivially find any supporting evidence with Google so it is possible that it is good enough in almost all cases (since afaik it doesn't actually implement the HTML5 algorithm yet it presumably doesn't work when you need browser-grade compatibility, but such extreme requirements are rare). In any case lxml is a superb library which I would recommend to anyone; personally I often use html5lib with the lxml treebuilder to take advantage of all the nice features in the lxml tree. It is also the best choice I am aware of for fast parsing of possibly broken HTML in python.