[whatwg] Attributes vs. Elements

Maciej Stachowiak mjs at apple.com
Mon Mar 12 11:25:07 PDT 2007



The recent discussion of possibly making the href attribute global  
brings to mind a broader issue. To what extent should semantics and  
behavior belong to specific elements, and to what extent should they  
be carried by global attributes that can apply to any element?

XHTML2 moves a lot of semantics and behavior from elements to global  
attributes. For example, href can turn any element into a hyperlink,  
and src can turn any element into an image.

HTML5 has so far not taken this approach - elements generally remain  
distinct, and new behaviors and semantics appear in the form of new  
elements.


In my opinion, it is usually better to put primary semantics on  
elements rather than attributes. Here are a few reasons, in  
approximate decreasing order of importance:

- An element is generally seen as a separate thing in itself, whereas  
an attribute is just an attribute. So it is natural to think of an  
element having more semantic strength than an attribute.

- Semantics that has associated behavior often has an associated DOM  
API. For example, the A element has a number of attributes related to  
its use as a hyperlink source, to programmatically retrieve or change  
only part of the href URI. If any element could have the href  
attribute, the options become omitting the API, adding it to all  
elements, or (yuck) making it conditionally present on all elements  
based on presence or absence of the attribute.

- It's easier (and, in many implementations, more efficient) to style  
by tag name rather than by presence or absence of an attribute.

- Typically, browser implementations have an implementation class per  
element, but not a class per attribute, so putting primary semantics  
and behavior on an element instead of an attribute promotes clean  
implementation. Otherwise, all behavior ends up in the HTMLElement  
base class.


Finally, I'd like to conclude with this reductio ad absurdum of the  
XHTML2 approach. If assigning behavior and semantics to attributes is  
so much better, why not just have a single <elt> element:

<elt role="paragraph">My cat is really cute: <elt  
src="mycat.jpeg">picture of my cat</elt>. Check out <elt  
href="story.html">this <elt role="emphasis">hilarious</elt> story  
about her</elt>.</elt>

I find the HTML approach much more readable and more semantically clear:

<p>My cat is really cute: <img src="mycat.jpg" alt="picture of my  
cat">. Check out <a href="story.html">this <em>hilarious</em> story  
about her</a>.</p>

Regards,
Maciej




More information about the whatwg mailing list