Re: [#issue-mt-ns] GRDDL and Processing Order

At 05:18 PM 10/6/2006 +0100, Harry Halpin wrote:

>I'm just going to sort of do an exegesis of Murray's proposal [1], which
>I *think* can be summarized as "A GRDDL transformation applies to the
>document as received by the client from the server with no additional
>processing assumed to be done. "

RDF=Tx(source)

Where RDF is a resource containing resource descriptions,
         source is a resource containing XML and grddl:transformation link(s)
         Tx is an undefined process
                 executed by an undefined agent
                         in an undefined operating environment

I believe that the intent of the author of GRDDL is to avoid making any
restrictive statements with respect to the policies of the processor
or the operating environment. I agree with Dan that there should be
no policy statements in the GRDDL WD that say what that process,
agent or operating environment should or should not do. For example,
I wouldn't imagine that we should require all GRDDL-aware agents to
perform any action at all, other than taking perfunctory notice of the
required markup being present. That is, all the GRDDL WD should say
is how to recognize GRDDL markup and what it means when encountered.

My SCO Help Browser had knowledge of a set of four LINK/REL values
and it provided a UI to exercise those links to find TOC/Index/Prev/Next.

A given XML or XHTML page might be transformable to resource descriptions
through several different transformation algorithms. Which one do I do first?
How would I or could I decide how to layer the information that was extracted?

Those are policy questions for agents or operating environments to answer.

>Then maybe to clear things up: "If additional processing needs to be
>done in order to extract valid RDF, it should be done either by the
>server before it sends the document to the server or as the processing
>should be included in the GRDDL transform itself."

The quoted sentence above is incomprehensible to me. I think that you
are saying that

         "The GRDDL WD does not warrant that any agent will perform any
         operation at all. However, if an agent under your control is 
capable of
         performing any tasks beyond those performed by an XSLT 1.0 
transformation,
         you could consider including validation and XInclude processing 
before sending
         or after receiving the result."

>Adding sentences to those effect in the Spec would be useful, maybe with
>one of the following examples:
>
>  This means a GRDDL transformation would *not* use XInclude to include
>items before running a GRDDL transform, or if the document itself
>contained XSLT elements it would *not* run that XSLT and then run the
>GRDDL transform, but would just run the GRDDL transform. If you wanted
>the include via XInclude items or run another XSLT transformation before
>running GRDDL,  you would use some sort of pipeline language like
>XMLProc as your GRDDL Transformation [2].

The first sentence is illogical.
"GRDDL transformation would *not* ... before running a GRDDL transform"
A GRDDL transformation can't DO anything before it is run. Only an agent can.

A GRDDL transformation is an undefined process. It may do anything.
Why would you assume that the agent would not run the XSLT?
I can imagine an agent which sends all GRDDL transformations
to a pipeline that may perform any number of tasks. And that agent
could also offer a UI to inform of the availability of an annotated version
of the source document, or a table of contents, or ....

I think that Dan is right to avoid making statements about the behaviour
of agents because the GRDDL WD is only intended to explain the mechanism
for linking to transformations, not how an agent or application will utilize
that mechanism.

>     There used to be some issues, originally brought up by Brian McBride
>I believe, that dealt with whether or not XIncludes should resovled, if
>XSLT code in the source document should be run before GRDDL, and the
>such but I can't find them [3] except for issues involving RDF as a
>namespace document [4], which I don't think Murray's suggestion applies to.

Again, I think that those are policies for the agent or the operating 
environment.

Regards,

Murray