Use the HTML5 parser from html5lib when it's available

[kennyluck]

I am not particular happy about the namespace kludge myself, but I'll post this patch here just in case Simon wants to take it. It should also be useful for anyone else who wants to test WeasyPrint's standards-compliance without falling into some HTML parsing trap.

The kludge can be removed whenever one of following applies:

1. lxml supports XPath 2.0's namespace wildcard *:html and cssselect outputs it.
2. cssselect outputs XPath of the form *[localname() = "html"]

. I am not prepared to go into hacking lxml/cssselect so I'll stop here.

[SimonSapin]

The namespacing stuff is a separate issue in cssselect. For the HTML parser, I'd rather not switch silently, but only use html5parser when explicitly requested by the user (and then fail if it can not be imported). I'm open to ideas for the specific API: maybe a subclass or an additional parameter.

[SimonSapin]

Could you provide an HTML test case where html5parser parses differently from lxml.html?

[SimonSapin]

Oh, I had only seen half of the issue. We can easily add support for a different HTML parser in WeasyPrint, but html5parser is useless because of the namespace issues in cssselect. This is fixable in cssselect, just low priority for me right now.

In the meantime, this is a better way to do the same, without patching WeasyPrint:

tree = lxml.html.html5parser.parse('http://example.net')  # Or whatever
for element in tree.iter():
    # Comments and other non-element stuff have a non-string "tag"
    if hasattr(element.tag, 'split'):
        # lxml uses the '{namespaceURI}localname' syntax
        element.tag = element.tag.split('}')[-1]
html = weasyprint.HTML(tree=tree, base_url='http://example.net')

[SimonSapin]

Leaving this open: I still want to support various parsers eventually.