RE: ISSUE-76: If we fixed namespaces, does RDFa still have problems?

WRT CURIES, couldn't we just apply the same fixed-prefix rules to them too. In fact, doesn't this end up making CURIES generally simpler?

That is, if we see

<span property="dc:creator">

Then, as far as HTML is concerned, the value of @property really is just "dc:creator". However a tool extracting RDF from this will know that the prefix "dc" always means Dublic Core and can act accordingly.


Similarly, wouldn't unregistered prefixes work just the same way as for node names? That is, if I saw:

<span property="x-apple:whatever">

Then the value of @property is just "x-apple:whatever". A tool that wanted to understand a page semantically could either risk the danger of name clashes and assume that "x-apple" means what it thinks it does, or it could check for the presence of an xmlns declaration saying what "x-apple" means.


-Rob


From: Maciej Stachowiak [mailto:mjs@apple.com]
Sent: Sunday, December 13, 2009 6:22 PM
To: Ennals, Robert
Cc: public-html@w3.org
Subject: Re: ISSUE-76: If we fixed namespaces, does RDFa still have problems?


Personal opinion:

On Dec 10, 2009, at 4:26 PM, Ennals, Robert wrote:


I'd like to draw attention to a point that Sam Ruby made a while back, when we were discussing my proposal for decentralized extensibility [ISSUE-41].

It seems that one of the biggest things that people don't like about RDFa, and thus one of the biggest reasons for creating Microdata, is that RDFa uses namespaces, and namespaces are really hard to use because the meaning of prefixes varies. However, if we adopted my proposal for decentralized extensibility, then prefixes have fixed meanings, and so I believe this problem goes away.

If we can make RDFa have most of the strengths of Microdata, then that would be a good long-term solution, but I don't see that happening fast enough to resolve ISSUE-76.

There does seem to be at least one problem of RDFa that is not obviously resolved via predefined namespace prefixes, namely that it applies namespace prefix resolution to attribute values, resulting in what are effectively "QNames in content" (really CURIEs not QNames, but many of the same issues arise).

It also seems that if you ever use any non-predefined prefixes, then all of the issues with RDFa's use of namespaces (including the somewhat un-XML-ish handling of xmlns declarations) would still be present, if the ability to use non-predefined prefixes was not completely removed.

So, I'm not sure the resolution is that simple. It does seem like a matter worth investigating further, but probably outside the scope of ISSUE-76.

Regards,
Maciej



What do people think the other big problems with RDFa are? How much do they matter? Are there ways of fixing them?

AFAICT the obvious annoyances with RDFa are:
*         Reliance on namespaces - but that can be fixed by the earlier proposal
*         RDFa requires you to declare a type to get a blank node. Does this matter? Could it be easily fixed?
*         RDFa expects @rel when the data is a link, and @property when it is content - but I quite like that
*         RDFa doesn't have a DOM API - but that can be added
*         RDFa requires an extra layer of indirection when putting one node inside another. This is because when a node has a @rel attribute, the node is the subject if the node has a type. Does this matter? Could it be easily fixed?


What am I missing?

To make this discussion more concrete, here is are a couple of examples, using the same data, written in Microdata vs RDFa, assuming that we have a fixed set of prefixes, and that we also have a fixed set of standard types and properties that don't require prefixes.

First, let's do an example where all the names we use are in the global namespace:

Microdata:
  <div itemscope>
    <p>My name is <span itemprop="name">Elizabeth</span></p>
    <img itemprop="image" src="liz.png" alt="My Photo">
   Band:
     <span itemprop="band" itemscope>
        <span itemprop="name">Jazz Band</span> (<span itemprop="size">12</span> players)
    </span>
  </div>

RDFa:
  <div type="person">
    <p>My name is <span property="name">Elizabeth</span></p>
    <img rel="image" src="liz.png" alt="My Photo">
    <span rel="band">    -- extra node needed
      <span type="Band">
        <span property="name">Jazz Band</span> (<span property="size">12</span> players)
      </span>
    </span>
 </div>

Now, an example where we use names from an external vocabulary. In this case, we assume the vocabulary has a registered prefix rather than an experimental prefix:

Microdata:
  <section itemscope itemtype="http://example.com/animals#cat">
     <h1 itemprop="name http://example.com/fn">Hedral</h1>
  </section>

RDFa with registered prefixes:
   <section type="animals:cat">
     <h1 itemprop="name animals:fn">Hedral</h1>
   </section>


When the namespace problem is taken away, what differences are still problematic?


Note that I am not advocating for any particular position. I'm just trying to explore the space.


-Rob