Subtleties of getElementsBySelector

I should point out that my implementation of getElementsBySelector corrects a couple of subtle bugs in the original version. For one thing, the original version searches for nodes of the form #foo by using document.getElementById; this results in finding the outermost node in the entire document whose ID matches, even if searching a subtree. Since ID's are not guaranteed to be unique, this can produce the wrong node.

An even more subtle bug in the original version is that multiple copies of the same node may appear in the result array. For example, the selector '.foo .bar .mumble', when applied to the following tree:

<div class="foo">
    <div class="bar">
        <div class="bar">
            <div class="mumble">mumble</div>
        </div>
    </div>
</div>

...produces two copies of the mumble node, because it finds it by following two different paths. By contrast, my implementation visits each node in the entire document tree exactly once, so it is guaranteed to produce either 0 or 1 copy of each node in the result array.