[
Lists Home |
Date Index |
Thread Index
]
Seairth Jacobs wrote:
> This could have been said for the initial adoption of XML as well. If
> people had listened to this arguement back then, we would still have only
> SGML...
IMHO, no we wouldn't. There are two radical departures from SGML in the
premises of XML: the first is the "SGML on the Web" impetus, however dimly
understood it was at the time, and the second is simple well-formedness.
Taken together these two utterly transform--and are both necessary to the
transformation of--SGML from the assumption of functional nodes made
transparent to one another by a priori agreements and from the mutual
transparency conferred by an homogenous network.
On the internetwork, functional nodes are inherently opaque to one another.
If that opacity is erased it is only through individual nodes giving up their
autonomy of process, which is the very essence of the internetwork. Were the
manner in which functional nodes instantiate data for their own purposes and
then execute their characteristic functions to be governed or limited or even
guided by the practices or expectations of other nodes, the result would
effectively be the creation of an homogenous network--that is, a network
characterized by a shared body of a priori assumptions ultimately governing
the individual functioning of each node. Instead, the internetwork is created
by overlaying a universal addressing mechanism on constituent nodes and
networks. Each node gains the power to address every other node directly. Yet
because the functional assumptions of constituent networks are not recast
around a body of shared data structures, datatypes, and the processing which
acceptance of those shared types and structures implies, the nodes of the
internetwork do not and cannot have 'interfaces' to other functional nodes.
The functional nodes of the internetwork can address one another directly,
can put or post documents by address, and can get documents from addresses
where they are authorized to do so. If those documents come with or refer to
DTDs or other schematics, then nodes which use those documents in their
particular processing may validate if they choose to test that the documents
as received follow the content model they proclaim. Such validation, if
performed, amounts to a functional node yielding, in particular instances at
least, its fundamental privilege of determining its own input data in every
instance of its function or, more accurately, delegating the logistics of
that determination to standard processing parameterized by an accepted form
of schema. The inclusion of the well-formedness-only option in XML 1.0
specifically separates testing for conformance of an instance document to a
particular content model from testing of that document's compliance with
fundamental rules of XML 1.0 syntax. Because it is the node consuming a
document which chooses in each instance whether to validate, the practice of
XML 1.0 has from the beginning made clear that decisions of what data to
accept--given always basic compliance with fundamental rules of syntax--lie
with each user of that data.
Historically, SGML systems were implemented, most often idiosyncratically,
within closed systems of homogenous assumptions. The premise on which various
functions of such systems interoperated by sharing the same documents,
similarly understood, went largely unexamined. In the 'data' world this is
analogous to how two-phase commit enables the transactional interoperation of
RDBMSs through the sharing of identical data structures identically
understood by transactional counterparties, down to the form and type of each
of their constituent fields. In such an environment--whether for processing
documents or data--it is easy to believe, and ultimately to design systems on
the belief, that if a document or data validates to an expected model then
functional nodes intended to operate upon a document or data of that model
should perform their expected functions against a given instance precisely
because that instance validates. Too often when I hear people speak of a
'contract' between the creator and the consumer of documents or data, it is
clear that what they really mean is a two-step agreement: the consumer
retains the right to validate each instance against expectations, but the
producer expects that if the instance validates the consumer will accept and
process it. So far as I can determine, the second step has not been
articulated as an explicit general premise, but it has nonetheless been for
twenty years or more the basis of building homogenous systems or of creating
a sort of interoperability which binds individual functional nodes into
systems.
In short, by being premised on straightforward usability on the internetwork
and by offering simple well-formedness at the choice of the document user,
XML 1.0 radically reaffirms the autonomy of each functional node in
determining its own input data requirements and in deciding what, and
whether, to execute against that data. By contrast with that radical
departure, any simplification or modification of XML 1.0 based on 'the
Infoset' or invoking 'the Infoset' as the basis of interoperability vitiates
the interoperability inherent in XML 1.0 and, in practice, will reverse the
premises of XML 1.0 by imagining some OO-like 'interface' to each functional
node of the internetwork and eventually positing some contractual obligation
of that node to perform when its interface is appropriately invoked. XML
1.0-compliant documents are syntactic entities which may be instantiated and
then used in various idiosyncratic ways by various functional nodes. That is
the sole basis of any interoperability which XML 1.0 affords. But, in fact,
it is the most general interoperability obtainable, and one which is vitiated
by any expectation, model or information-item abstraction which might
additionally be imposed on the uses of a given instance document.
Respectfully,
Walter Perry
|
