View source for XBL:Dynamically Applied Shadow Trees

By [[MDC:User:Benjamin Smedberg]]

One of the concerns I have with the current XBL2 specification is the way the (cloned) shadow tree is constructed and reconstructed. Instead of having a single shadow tree that is constructed from all the attached bindings, and reconstructed when a binding is attached or removed, I believe that there should be a cloned shadow content subtree for each binding which is then combined dynamically at runtime by the implementation.

There are a couple of motivating factors for this change:

# The current teardown and rebuilding of the shadow tree whenever a binding is attached or detached means that bindings which modify the shadow tree using script would need to repeat such actions every time the teardown/rebuild process completes. This case may not be uncommon in cases where a language (such as XUL) is implemented with XBL, and author bindings are also used. How would a binding such as the [[XBL:Dynamic Use Cases|2dchart binding]] know when to rebuild its shadow content if external activities such as attaching derived bindings could overwrite its changes?
# Keeping each binding's shadow tree separate allows for significant implementation optimizations; in particular the shadow tree could be a shared prototype until/unless it was accessed or manipulated via script. This is particularly important in terms of bindings which are part of a language and used frequently, as it has a direct impact on startup time.