Do you have a test demonstrating this? I'm not 100% sure I follow.

Certainly,

 <form><img name="n"/></form><script>alert(document.forms[0]["n"]);</script>

Per that non-normative text afaict, the <img/> won't match.

(I'm assuming the "Returns the form control" part of that text to refer to the 'listed elements' of the form. Which does not have <img/>.)

Ah, I see. Yeah, that's intentional. The non-normative text ignores what happens for non-normative content. <img name> is non-normative.

That doesn't make much sense :-) Where is <img name> called out as being outside of what the spec requires when the algorithms below (in normative text) includes it, but without stating that it is not a requirement to also consider <img name>?
I would expect the non-normative box to be a summary of normative text, at least here.

(A trifling detail given that the text isn't normative, but still.)

(In reply to sof from comment #4)
> Where is <img name> called out as being
> outside of what the spec requires when the algorithms below (in normative
> text) includes it, but without stating that it is not a requirement to also
> consider <img name>?

I don't understand that question.

The idea is that implementors refer to the normative text to implement browsers, and authors refer to the non-normative text when figuring out how the APIs work.

Authors don't need to know anything about what happens with <img name>, since if they're using it they've probably made a mistake, and the validator will help them to convert to using id="" or whatnot. So the non-normative text doesn't mention it. This is one of the main ways the non-normative API descriptions differ from the normative ones; if the non-normative text were pedantically correct, it might as well be normative.

A problem here is/was how the line between normative and not is drawn (where is that defined); just trust the non-normative text as always being accurate?

But I now see that 

 http://www.whatwg.org/specs/web-apps/current-work/#non-conforming-features

enumerates img.name as obsolete. So, we're fine in this regard.

(In reply to sof from comment #6)
> A problem here is/was how the line between normative and not is drawn (where
> is that defined); just trust the non-normative text as always being accurate?

As an implementors, you should just ignore the non-normative text entirely. Especially the domintro blocks, but also anything that says "Note:", to be honest.

And definitely ignore all the examples. And the diagrams. And the text in the indexes, and any tables marked non-normative, and... :-)