W3C home > Mailing lists > Public > public-html@w3.org > August 2008

Re: GRDDL

From: Henri Sivonen <hsivonen@iki.fi>
Date: Wed, 6 Aug 2008 11:12:38 +0300
Cc: Toby A Inkster <tai@g5n.co.uk>, Justin James <j_james@mindspring.com>, public-html@w3.org
Message-Id: <1740E74B-3005-4A29-A019-F38CE14432E7@iki.fi>
To: Julian Reschke <julian.reschke@gmx.de>

On Aug 6, 2008, at 10:21, Julian Reschke wrote:

> Henri Sivonen wrote:
>> GRDDL it is [] very different from CSS in a way that is crucial  
>> considering the question of "CSS for semantics" particularly [for]  
>> accessibility:

(Text input problems there.)

>> CSS doesn't throw away the original DOM. Instead, it annotates the  
>> document tree with properties that are updated when the DOM is  
>> updated.
>> GRDDL, on the other hand, transforms the original document tree  
>> into RDF discarding the relationship between the RDF triples and  
>> the original tree nodes once the transformation has completed.  
>> Also, if you want the RDF triples to be updated when the document  
>> tree changes, you need to rerun the whole transformation.  
>> Therefore, GRDDL is not suitable for annotating an in-browser DOM  
>> with accessibility semantics.
>> (ARIA annotates the DOM without a layer of indirection: the ARIA  
>> properties are attached directly to DOM element nodes as attributes.)
>
> Yes.
>
> GRDDL allows extracting RDF out of other vocabularies. The  
> vocabularies aren't changed

I think defining mappings from unchanged vocabularies to RDF is the  
right way to go if you want to process RDF. (As opposed to asking  
everyone to serve RDF whether in RDF/XML, RDFa, N3 or whatever.) So  
that's cool.

> (well, except for the hook pointing to the transform, if you want to  
> count that).

It seems to me that it would be more useful for the extractor to  
contain a catalog of transforms keying off well-known Content-Types  
(or root namespaces for */*+xml types) and targeting the kind of RDF  
vocabularies that the user of the extractor is interested in, since  
under such a model the extractor would work even when the author isn't  
providing the hook. There will always be more HTML pages labeled text/ 
html without GRDDL hooks than HTML pages with GRDDL hooks.

> If you're looking for a technology that keeps the RDF information in  
> place, look for RDFa.

Personally, I prefer mapping HTML-native metadata to RDF over putting  
RDF directly into HTML. I just think it's more useful for the  
extractor to contain the mappings than to ask HTML authors to  
participate and supply programs for a standardized extractor to run in  
a sandbox.

>> I'm not sure if abusing HTML is the right characterization, but the  
>> GRDDL setup violates the The Rule of Least Power TAG Finding.
>> http://www.w3.org/2001/tag/doc/leastPower
>
> I'm not sure how using XSLT 1.0 violates that finding (please  
> elaborate);

XSLT is in a more powerful language category than (scriptless) HTML or  
any of the notations for RDF triples.

> but it's interested to see TAG findings quoted here.

My motivation to bring this up is pointing out that it's not only the  
browsable Web (aka. the Web) that violates Architecture but the  
Semantic Web violates it as well, which hopefully puts HTML5's  
Architecture violations into perspective down the road when we will no  
doubt be discussing Architecture violations.

>> Instead of serving non-scripted HTML with well-known semantics  
>> (<title>foo</title>) at URI u or serving some RDF triples (http://purl.org/dc/elements/1.1/title 
>>  of u is "foo"), GRDDL is about serving a program and a black box  
>> of data and telling the consumer to run the program with the data  
>> as the input in order to obtain RDF triples as the output of the  
>> program.
>
> Yes. But it's not meant to replace those other techniques.
>
>> Would it be an abuse of SVG if an SVG image wasn't served directly,  
>> but instead a script that fetched the SVG file using XHR and  
>> rendered it to <canvas> was served?
>
> Yes, that would be bad.
>
> Not sure what your point is, though.

If "run this PostScript/JavaScript program to see an image" (as  
opposed to serving an image that isn't a program) is "bad", surely  
"run this XSLT program to get triples" is also "bad".

-- 
Henri Sivonen
hsivonen@iki.fi
http://hsivonen.iki.fi/
Received on Wednesday, 6 August 2008 08:13:21 UTC

This archive was generated by hypermail 2.3.1 : Monday, 29 September 2014 09:38:57 UTC