[regression] dojo.query matching too many elements

[dante]

using a simple html set:

<div id="one">
    <p class="one"><a>one</a></p>
    <div id="two">
        <p class="two"><a>two</a></p>
    </div>
</div>

the query 'div p' returns 3 elements. also 'div p a'. there are clearly only two p and a elements in the page. the queryies 'p' and 'p a' both pass, as they don't contain an embedded match.

failing unit test attached.

[dante]

first two pass, second two fail

[bill]

This works in 1.2, fails in 1.3 (and trunk).

[nic]

The test fails because there are two nested divs.
For example, dojo.query("div p"): the root.getElementsByTagName(query.getTag()) (query.js, line 1033) finds one node, p.two, when root is the inner div and two nodes, p.one and p.two again, when root is the outer div.
So the ret array contains three elements and the _zip function is never called (ret.nozip = true, line 1092).
Maybe the assumption root nodes > 0 => no dup.controls (line 1086) is correct only when there are no ancestor/descendant root nodes.

[James Burke]

If ret.nozip is not needed on line 1092, then perhaps get rid of the if (x > 0) block also. From what I can tell, bag is never assigned otherwise in the filterDown function, and my cursory look at getElementsFunc and "bag" references seem to indicate bag is not modified underneath either. But it would be good to get confirmation from Alex on it.

[alex]

(In [17460]) fix for query system matching too many elements (possibly not checking for uniqueness in a set). Refs #9071. Back-port to 1.3.x branch will fix. !strict

[alex]

(In [17461]) back-porting fix for query() DOM branch matching too many nodes in descendant selectors. Fixes #9071. !strict