Sorry for reopening this again:

Ok after rereading the spec and reiterating the problem again I came to the same conclusion as Ganesh and Alex here.

Everything which is in the CDATA blocks in update must be considered to be part of the update.
so if a component author uses some kind of decoration around the <placeholder id="clientId> part then he has to be aware of the implications.

I cannot really throw an error here however if the author does not follow the pattern, due to not being specified and due to having valid usecases to do this (mostly to embed some javascript which should get triggered rightly after the update is done but before the entire ajax processing is finished)
So I will revert my original mojarra identical behavior to the one which I had originally, which is more or less exactly the behavior the spec says and what you can get within mojarra under IE as well (whatever Mojarra had for the fixup of this problem fails on IE anyway)

Werner


On Wed, Apr 14, 2010 at 2:32 PM, Werner Punz <werner.punz@gmail.com> wrote:
Ok I posted an issue with the problem and solution under bug #785 in the spec issue database so that it is not lost.


Werner



On Tue, Apr 13, 2010 at 9:46 PM, Werner Punz <werner.punz@gmail.com> wrote:
Ok probably too late to fix that for myfaces 2.0 but I will get the fixup in for a post 2.0 release (since this is more or less cosmetic for normal root element usecases I dont see any problem by postponing the fix)
with a clear error mesage.
In either case, both Mojarra as well as myfaces are broken in this area, while I have the feeling that Mojarras behavior is more broken by being inconsistent between IE and the rest of the world :-(

Werner



On Tue, Apr 13, 2010 at 8:58 AM, Martin Marinschek <mmarinschek@apache.org> wrote:
On Tue, Apr 13, 2010 at 1:56 AM, Alexander Smirnov <asmirnov@exadel.com> wrote:
> Surely, the API should be available for component developers. Some
> components need to update its own part only - for example, tree has to
> manage expand/collapse events, or table that able to append or delete
> particular rows.

I agree. The API should be there. And for now, we should present an
error for the broken cases.

best regards,

Martin