OK, I have a little function walking the tree like this:

function walkTree(node, func, args) {
    func(node, args);
    node = node.firstChild;
    while (node) {
        walkTree(node, func, args);
        node = node.nextSibling;
    }
}

And another function that would pick up only the text nodes like so:

function selectTextNodes(node, nodes) {
    if (node instanceof Text) {
        nodes.push(node);
    }
}

And finally, using both:

texts = [];
walkTree(body, selectTextNodes, texts);

However, it doesn't fill the list at all!

If I were to modify the test as to use Node.nodeType it would work:

function selectTextNodes(node, nodes) {
    if (node.nodeType == Node.TEXT_NODE) {
        nodes.push(node);
    }
}

On the other hand, in the console it works both ways:

t = window.document.createTextNode("test");
r = (t.nodeType == Node.TEXT_NODE) && (t instanceof Text);

That is, r is true.

Note that, all functions are nested inside another function that receives the body variable. In my case, this is the contentDocument.body of an iframe. There is no x-domain restriction being applied.

Any idea what's going on?

There are different Text interfaces in the different windows. So, if you have a DOM node from your iframe document, it is not an instanceof window.Text, but an instanceof iframe.contentWindow.Text. Afaik, also the availability of the Text interface as a Javascript object is non-standard.

That is why you just should check the nodeType of the elements. But notice that (older?) IE does not support the TEXT_NODE constant on Node, so you will either need to compare with 3 or assign that value as a polyfill to Node.TEXT_NODE.

more