ietf-mxcomp
[Top] [All Lists]

RE: XML name space (was: suggested new RRtype experiment)

2004-05-24 14:59:56

On Mon, 2004-05-24 at 10:45, Bob Atkinson wrote:
Please, Doug; it's seems apparent that you misunderstand how XML works,
in particular it's schema resolution and namespace semantics.

On the one hand 'MARID token' you advocate in no way helps solve the
problem unless it's rooted in some namespace, which you disclaim.

It MUST be controlled by the standard and not become a prerogative of
every programmer wishing to publish new definitions.  Registering the
XML definitions as a token published with IANA controls this name
space.  This is the function of IANA.  Allowing any URI to perform this
function allows ad hoc changes that must be understood by record parsers
through some undefined out-of-band communication.

On the other, you continue to appear to state that downloads are
necessarily required, whereas in practice this will never occur. Even in
the face of extensions, a given server will only be able to actually do
anything useful with a certain fixed set of namespaces (and their
associated schemas), namely the ones that it is coded up to semantically
understand. For those namespaces, the server will load up it's XML
parser with just those definitions, and parse away.

The 'certain fixed set of namespaces'  are not defined by way of a
specific specification nor are these definitions registered in a fashion
allowing 'a given server' a means to understand records a priori.  This
instead becomes defined-by-failure or these definitions must treated in
the manner of web browser.

The xsd:any's and the like are defined in the marid schema with
'processContents="lax"' precisely so that servers that lack
understanding of a particular extension can still parse the remainder.

Without a fixed predefined set of definitions, the entire record becomes
forfeit with the first change.  The "example" reference to a name space
consumed 100 bytes alone.  This overhead is simply too great.  How many
such definitions do you expect used?  It is nonsense to suggest this is
a means to incrementally extend record definitions where the entire
query and response is expected to fit within 512 bytes.  The bare bone
template example in the draft required 4k bytes to just describe the
variable structure.  The content of a TXT record is just ASCII text. 
The XML means of expressing data is far too verbose to be practical
within these constraints.  It is clear why this proposal only expressed
a limit of 2k for the record rather than 512 bytes for the entire
query/response.

The system breaks at a whim if a URI is allowed in the DNS TXT record as
a definition reference.  The use of any and all URI XML definition
references MUST be prohibited.  I have explained how this can be done in
a manner fully compatible with XML.  I still think XML is not
appropriate, however.

-Doug