Monday, July 29, 2002

Greg Reinacker on Sam Ruby on Loosely Coupled Web Services.  ACK. 10:05:55 PM      comment []

Mark Baker: Here's how I see it. HTTP is the protocol, and no other protocols are required in order to get stuff done.  Perhaps Mark has a narrower definition of communcations protocol than I do?  [Sam Ruby] Here we have it, Sam and I agree on something. I hope Mark's statement is from a larger context that I haven't really followed.  There are things like reliable delivery, broadcast, transactions, streaming, store/forward for which support demand can be expressed in WSDL (by declaring and hence demanding support for headers), but many which can not be solely be carried through HTTP but require multi-pipe connectivity or just can't be done using HTTP, at all. Show me an approach to negotiate a 2-phase or 3-phase commit protocol successfully through HTTP and I'll be quiet. 8:23:04 PM      comment []

In referral to my previous post: With the "same semantic metaschema" I actually meant that the metaschemas are at least compatible and this would be enabled by a network of basic metaschemas that define well understood terminology from all sorts of fields and are based on well-known standards (ISO, IEEE, W3C, UN, WHO, etc.) and also corporate and organization sources. It actually doesn't have to be coordinated and organized and can be as chaotic as the web is, because everyone can and should freely link and establish similarities between these things. So, if both parties talk very loosely (coupled) about a "beetle" and the metaschema reference for one comes from the animal space and the other comes from the Volkswagen space, they are probably not the same. But if both talk about some different sorts of funky insects and know about that by tracing their metaschema references back to a common junction point ... well, then there's hope that some smart logic may actually be able to negotiate the least-common-denominator data between them. The Semantic Web Activity at the W3C does the groundwork for these things now; however, none of the examples that I've seen (and that may be because I am ignorant or lazy not to look hard enough) actually annotate XML Schema (they rather annotate RDF) and I think that's something to at least think about.  For reasons of processing speed and efficiency I would like my stuff to stay very tightly coupled, early bound and with specialized, auto-generated parsers/serializers with a high degree of reliance on stable WSDL and XSD at all times; like RPC stubs/proxies.  For interoperability reasons, I would like to have the message based equivalent of a dynamic invocation interface as well, which is strictly late bound and based only on metadata discovery and works on a global scale and with potentially anonymous partners; resembling the DII, Reflection or IDispatch idea, but based on meta-metadata. Schematron + Semantic Annotations + XSLT could be candidates for components to build this - methinks. [And with this, I invite Sam to have the final word on this friendly banter. I will accept anything from "you are completely nuts" to "that's worse than I thought it could get".] 2:01:14 PM      comment []

I said: "Complete agreement may be a problem, but agreement on a composite, well-defined subset of things that both parties are interested in and understand is not."  Sam says: "Pardon me, but did you say both?  BOTH?  BOTH!!!???  Works well 1:1, but how about nnn:mmmmmm?" I say: "When watching TV, my favorite TV station's program people and I seem to have a well-defined subset of common interests. Otherwise I wouldn't watch their stuff so frequently. They don't know me, but there's a 1:couple-of-million 'both' relationships there."  And then Sam says on my recent post: "I was more thinking of other ways.  Ones that do not require existing instance documents to be invalidated when an XML schema is only extended." look! Look! LOOK!!!!! You can keep all instance documents intact (provided you have an instrumented schema), have any schema namespace you like, have schema versioning, or have no namespace and/or no schema (provided you have instrumented elements) and all that, if you just tell me what the things mean that you give me by annotating them with semantics. I am only saying that schema has to do with semantics, because there's no other way to express a semantic association of data in the standardized technology set but "my thing" and "your thing" via namespaces. It's very, very poor, but that's all we have.  When we start linking things like OWL into schemas and document instances, we'll have a much richer way to define and agree on things. I can write a schema that defines elements and attributes all in German and you can write one that has everything in English and even structure it somewhat differently or use other concrete datatypes (say, strings instead of numbers). If we both use the same semantic metaschema for our schema, we'll both understand.   4:54:47 AM      comment []

© Copyright 2002 Clemens Vasters.