Comments on the draft:

* It makes no sense for the UA to implement an interface. Objects that the UA exposes implement interfaces, not UAs.

* The UA doesn't need to expose an object that implements the NSResolver, only the author does.

* Having an interface doesn't imply behaviour -- e.g. NodeList doesn't imply that NodeList is live. You can have an object that implemnets NodeList and is not live.

* IMHO getElementsBySelector() should return a live list, just like getElementsByTagName.

* I would recommend against supporting namespaces in the first version, for simplicity.

* I would recommend having getElementsBySelector and getElementsBySelectorNS if you wanted to support both, rather than using optional arguments. Some languages don't support method overloading on argument signatures and would need different numbers of arguments anyway.

* There is no actual conformance criteria for what should happen when the getElementsBySelector method is invoked (the spec says "returns" not "must, when invoked, return").

* The spec doesn't say _how_ to resolve the namespaces using the nsresolver argument.

* IMHO the argument to getElementsBySelector should be a "group of selectors" not a "selector" (using Selectors terminology).

* IMHO the method should not raise an exception when the selector contains a pseudo-element. It should would return an empty list.