[no subject]

As suggested by Ken, while the tree is being constructed, some parts
of it may be serialized without even the whole of the tree being
formed in memory. This theory makes me feel, the XSLT 2.0 spec is
correct in the current form (as regards to the point we are
discussing).

Also, as mentioned by Mike, that a user might want something like, <x
xsi:type=xs:date> in the output. In this case the default provided by
the processor (i.e., generating the xs: declaration) is helpful.

The spec has to balance, what the processor should do for the user
(for this discussion, generating xs: declaration for the user), and
what the user can do (like doing exclude-result-prefixes in the
stylesheet).

The WG already has had years of discussions finalizing this decision,
and potential implementers agree to this, so I think it should be
fine.


-- 
Regards,
Mukul Gandhi