For markup such as

(a) <form><template><input></template></form>

or

(b) <template><form><template><input></template></form></template>

the spec currently requires that the <input> be associated with the <form>. But that is incorrect: in case (b), the form and input are in different trees. And in case (a), they are not even in the same document! The existing implementations in WebKit, Blink, and Gecko all make this "mistake".

To correct it, I propose that the paragraph in the section on "Creating and inserting elements" (http://www.w3.org/html/wg/drafts/html/master/syntax.html#creating-and-inserting-elements) which currently reads:

"If an element created by the insert an HTML element algorithm is a form-associated element, and the form element pointer is not null, and the newly created element doesn't have a form attribute, the user agent must associate the newly created element with the form element pointed to by the form element pointer when the element is inserted, instead of running the reset the form owner algorithm."

have an additional clause added, requiring that the element has an ownerDocument with a browsing context. This will keep this special-casing of form association from taking place inside templates. I've implemented this in Blink in https://codereview.chromium.org/17033002/ (this misbehavior was causing some crashes, which is why I'm landing the code before a spec update).

An alternate approach would have been to compare the <form>'s ownerDocument with the <input>, but that would lead to the wrong answer in case (b) above (since both <template>'s contents share the same browsing context-less document).