>3?) I still wonder about namespaces as attributes, Xalan does it that way,
>Xerces (1.4.2) does not...
I believe the Xerces does both -- it passes the namespace declarations as
attributes _and_ issues the start/end prefix mapping events. (We rely on
those attributes when building the DTM model, in fact, so I know we're
receiving them.)
Xalan's output should also be issuing both, in case the next stage of
processing wants to use the prefix events. If we aren't, that might indeed
be a bug.
(On the other hand, generating the prefix events imposes some additional
performance overhead; every attribute has to be checked to see if it was a
namespace declaration, and if so reported both ways. I'm tempted to suggest
that if we do produce prefix events, we provide a mode which suppresses
them, so folks who know they aren't going to use them can optimize them
away. Serialization is already slower than we want it to be.)
Interesting little issue.
