W3C home > Mailing lists > Public > public-html@w3.org > November 2009

Re: Namespace-aware APIs do not need to suck

From: Maciej Stachowiak <mjs@apple.com>
Date: Thu, 05 Nov 2009 02:18:31 -0800
To: Julian Reschke <julian.reschke@gmx.de>
Message-id: <E6F3EBAB-CD5A-4C90-B154-8E4D3A65231D@apple.com>
Cc: public-html@w3.org

On Nov 4, 2009, at 1:26 PM, Julian Reschke wrote:

> In today's panel it was pointed out that people do not "get" that  
> the real identifier for an element type is a tuple (namespaceURI,  
> localName), and thus URI-based extensibility is a problem.
>
> My experience in other spaces that use URI-based extensibility is  
> that this problem goes away once you use an API that actually takes  
> namespaces into account.
>
> To address the "tuple" problem, a very simple approach is to use the  
> so-called "Clark" notation (<http://www.jclark.com/xml/xmlns.htm>),  
> where the tuple (namespaceURI, localName) is simply replaced by the  
> string
>
>  {namespaceURI}localName
>
> Note that a string in Clark notation can always be disambiguated  
> from a string in prefix notation. We took advantage of that in the  
> JCR 2.0 specification where we overloaded the methods that  
> previously took prefixed strings to allow Clark notation as well.
>
> In the Java (!= JS) API world, we also have been successfully using  
> Java interfaces for these kind of interfaces, exposing both the  
> components of the tuple plus the Clark notation, and leaving it to  
> the implementation how to internally represent that.
>
> In the panel it was also pointed out that no proposals have been  
> made to actually enhance DOM L2 with respect to this. On the other  
> hand, it hasn't been exactly clear that the WG actually is  
> interested to pursue this. If we are (unconference session Thursday  
> or Friday), we really should spend some time looking into this. I  
> can't say that I fully understand the restrictions under which we  
> can extend then DOM (IDL?), but I'm certainly willing to contribute.


So you could imagine code like this:

var el1 = document.createElementNS2("{http://www.w3.org/1999/ 
xhtml/}span");
var el2 = document.getElementsByTagNameNS2("{http://www.w3.org/1999/xhtml/ 
}div")[0];
el2.appendChild(el1);

Hmm. The fact that you have a single identifier instead of a tuple  
sucks less. But it seems to me the resulting code is not very easy to  
read, or pleasant to write. You could put the bracey bit in a variable  
like people do with namespace URIs today:

var xhtmlNS = "{http://www.w3.org/1999/xhtml/}";
var el1 = document.createElementNS2(xhtmlNS + "span");
var el2 = document.getElementsByTagNameNS2(xhtmlNS + "div")[0];
el2.appendChild(el1);

But then all you've really done is replace a comma with a plus. I'm  
not convinced that this will aid author understanding.

Furthermore, what you're writing still doesn't match the markup, which  
uses prefixes instead of Clark notation. Thus I think content  
producers and consumers will continue to mistakenly think the qname is  
the semantic unit, instead of the Clark-style namespace+URI string.

Regards,
Maciej
Received on Thursday, 5 November 2009 10:19:12 UTC

This archive was generated by hypermail 2.4.0 : Saturday, 9 October 2021 18:45:02 UTC