This feature is now implemented.

Testing of pathological cases would be useful (such as render directives for non-existant or overlapping IDs), however, we must compare the ICEfaces failure in these cases to the stock JSF failure).

It is recommended that developers use <f:ajax> render directives only as a last resort for performance optimization, as stale render directives will result in pages that are extremely difficult to maintain.

The diff is now being applied to just the rendered subtrees:

If isRenderAll() is false, the DOMResponseWriter is flagged for subtree rendering, and uses the old DOM as the current DOM
For each component subtree render, the existing DOM subtree is removed and the new subtree is rendered in its place
The diff is calculated against the recently removed subtree

The assumption is that a given component will render a rooted subtree of markup. Component rendering behavior other than this (such as rendering two elements, or rendering a new element without an ID) would result in an empty or malformed diff result.

The basic sample has been enhanced to include icefaces-subtree.jsf (with both execute and render directives). The ICEfaces pages now also include a timestamp to show which part of the page is updated. For icefaces.jsf, the timestamp is always updated because the full page is rendered. For icefaces-subtree.jsf, the timestamp is not updated because it is not included in the render directive.

Implementation note: it is not necessary to clone the DOM. When partial rendering is used, the replaced fragments of the old DOM can be retained. Modification of the old DOM in-place can be used to create the new DOM. The diff can then be run against the retained fragments.

An initial prototype for subtree rendering has been checked in.

Support for document.getElementById() is provided by identifying the id attribute via

Element.setIdAttributeNode()

for attributes that are named "id".

During partial rendering (isRenderAll() == false) the list of ids is "visited" with an encodeAll() in tree visitor callback. The first complication is that the new DOM is empty (the page has not been rendered yet). This is resolved by cloning the old DOM. When a component is partially rendered, the previous subtree is replaced with an empty version; the problem with this, is that the first action of the component is to render the root of the subtree. To suppress this, a flag is set that suppresses the next startElement().

To improve this implementation, the old DOM should not be cloned: it would be more efficient to keep the subtree fragments generated by the rendered components. As a feature improvement, it should be possible to specify whether a given subtree will have the DOM diff applied or not.

document.getElementById() looks reasonable to add to DOMResponseWriter, so I will prototype the JSF rendering aspect. This bug is slightly higher in priority now since many of the mojarra tests exercise this feature.