On Dec 12, 2010, at 19:45, Amelia A Lewis wrote:
> XML is, by design,
> extensible. To be so, it has to be more draconian, because there are
> no reasonably imputed semantics for unknown tags (or even for unknown
> attributes). Specifying "must ignore" could work for some fields. It
> doesn't for all.
I'm not suggesting vocabulary-aware error recovery. I'm suggesting the tokenizer should always transition in a well-defined way on any input character (like the HTML5 tokenizer does) and the tree builder should discard end tag tokens without a matching item on the stack or if there is a matching item on the stack, pop until that item has been popped. (Dealing with unbound prefixes doesn't have such a nice solution, unfortunately.)
--
Henri Sivonen
http://hsivonen.iki.fi/
