Want an example? Your browser, right now, can auto fill lots of form fields based on the field name. And your browser retrieves that information from its own little database. Let's just generalize that idea, and furnish client programs with the ability to understand lots more about the service it's exercising.
Where do we get the semantics for Atom? The Atom syntax document goes a long way. Henry Story has constructed an OWL ontology for Atom. This is a good starting point. We can use those property and class names in XHTML
Remember the motivation: We want web services to spread as powerfully as the real WWW has. The Atom group believes RESTful web services will do it. But little in Atom, or any other document driven protocol, resembles the actual web we experience. In Atom, they agree on a document format that carries semantics that all servers and clients have to understand, or at least handle gracefully. That concept has no analogy on the web. When you order airline tickets, do you complete the One Universal Reservation form? No. Each airline or agency has its own form, because each business has its own differentiators.
It's the same in web publishing. Consider Moveable Type and Blogger. To create an entry for MT you fill out this form:
They have two places to enter content! One is the lead, the other is the "Read more...." section. There's no concept like that in Atom. I guess the MT guys will propose an extension. But then every other blogging service on the planet will have to handle the case where somebody submits an "mt:extendedContent" extension element. Blogger doesn't have that concept:
I marked up those screen captures with tags grabbed loosely from the Atom ontology. We can make up more. Here's a microform MT could present (modulo GUI labels and layout):
<form method="post" action="/feed/">
<input type="text" name="title" class="atom:title"/>
<select name="category" class="atom:label">
<textarea name="entry" class="atom:htmlContent"/>
<textarea name="extended" class="atom:htmlContent"/>
<input type="checkbox" name="draft" class="app:control" value="app:draft"/>
Later we can hash out the details of these class attributes. Maybe we do need an MT specific extension for that extended content. But only MT servers, or any service supporting extended content, ever will have to deal with clients that submit that field.
So the microforms idea is: You describe the semantics of the form elements in their class attributes. You program clients to GET forms, understand form elements, and fill them out correctly, getting the information from some kind of database, or from a live user.
If we do it well enough, we can build agents that syndicate stuff to all kinds of servers automatically.
(updated: fixed omitted class attribute in form example).