I think the text in the spec for XMLDocument might explain this:

http://dom.spec.whatwg.org/#xmldocument

"""A document is assumed to be an XML document unless it is flagged as being an HTML document. Whether a document is an HTML document or an XML document affects the behavior of certain APIs."""

So createDocument() returns an XMLDocument if you pass no arguments and an HTMLDocument if you pass the HTML namespace.

So you know createElement will return an HTMLDivElement if responseXML instanceof XMLDocument is false.

Also note that real world implementations absolutely do return an HTMLDocument or a plain Document (now XMLDocument) based on the content type from responseXML. Gecko, Webkit and Blink all will pick one or the other.

The "XML document" and "HTML document" flag thingie is distinct from the interface going around.

As for what document.createElement() does. I recommend reading bug 19431.

(In reply to comment #3)
> The "XML document" and "HTML document" flag thingie is distinct from the
> interface going around.

Right, so responseXML can be an XMLDocument or an HTMLDocument but createElement will produces HTMLElement instances.

> 
> As for what document.createElement() does. I recommend reading bug 19431.

That thread isn't clear to me. How do I know if a random document created by document.implementation.createDocument() will produce an HTMLElement or an Element when I do createElement('div') ?

FWIW, createDocument() always creates a document marked as "XML document".

That bug is about what document.createElement() should do and whether it should depend on the type of document at all. Currently it only has a few exceptions for "HTML document" flagged document objects, but only with regards to casing.

(In reply to comment #4)
> Right, so responseXML can be an XMLDocument or an HTMLDocument but
> createElement will produces HTMLElement instances.

No. responseXML can only be Document. HTMLDocument no longer exists (it has been merged with Document). XMLDocument is only returned by the document.createDocument() API. As Anne said, "HTML document" and "XML document" are different concepts.

> That thread isn't clear to me. How do I know if a random document created by
> document.implementation.createDocument() will produce an HTMLElement or an
> Element when I do createElement('div') ?

Currently createElement('div') always produces an HTMLDivElement, regardless of the document.

(In reply to comment #6)
> Currently createElement('div') always produces an HTMLDivElement, regardless
> of the document.
Per spec, but not per implementations. Not sure which one is wrong.
I think spec, but not sure...

> HTMLDocument no longer exists (it has been merged with Document)

Which no one implements in practice and which is not clearly desirable anyway, right?

(In reply to comment #8)
> > HTMLDocument no longer exists (it has been merged with Document)
> 
> Which no one implements in practice and which is not clearly desirable
> anyway, right?

FWIW, IE9+ only implements Document. It does not have an XMLDocument nor HTMLDocument.

In IE9 the thing returned from responseXML isn't even instanceof Document and is otherwise all sorts of weird (toString on it throws, Object.prototype.toString.call returns "[object Object]", etc, etc).  It certainly doesn't support various Document stuff from the spec like the named getter and write().

So I guess you could argue that IE has merged Document and HTMLDocument by making only HTML documents be Document.... I'm not sure that's useful.

(In reply to comment #10)
> In IE9 the thing returned from responseXML isn't even instanceof Document
> and is otherwise all sorts of weird (toString on it throws,
> Object.prototype.toString.call returns "[object Object]", etc, etc).  It
> certainly doesn't support various Document stuff from the spec like the
> named getter and write().
> 
> So I guess you could argue that IE has merged Document and HTMLDocument by
> making only HTML documents be Document.... I'm not sure that's useful.

That was in IE9... In IE10 the following asserts all passes.

var xhr = new XMLHttpRequest;
xhr.open('GET', 'test.xml', false);
xhr.send();

assert(Object.prototype.toString.call(xhr.responseXML) === '[object Document]');
assert(xhr.responseXML instanceof Document);

Ah, interesting.  Does IE10 support the named getter on such documents?  What does it do with document.write() on them?

(In reply to comment #12)
> Ah, interesting.  Does IE10 support the named getter on such documents? 
> What does it do with document.write() on them?

Based on my testing... The MS guys are now subscribed to this bug so they can probably answer these questions better.

* document.write throws (maybe it works during loading of the doc... I didn't test)

* Named getters works

*** This bug has been marked as a duplicate of bug 25028 ***