[cap-talk] Potting the web-calculus in a paragraph

[Jed at Webstart]

At 03:48 PM 12/9/2005, Karp, Alan H wrote:
>Ian G wrote:
> >
> > I can suggest Pelle's style - he follows a market process
> > and grabs whatever tools he needs to do that.
>
>That's a great way to get something going quickly.  Unfortunately, if
>you don't understand the why's, you often end up making subtle errors.
>These errors may cause bugs that you recognize and fix, or they may
>cause vulnerabilities that you're not aware of.

Or perhaps build dead ends into the architecture that are difficult
or impossible to recover from later.

This is one reason I mentioned the lost object problem.  I think
it's important to deal with up front.

>It's not uncommon when adopting a new viewpoint.
>
>I know it happened to us with Client Utility (e-speak Beta) when we
>introduced "split capabilities".  Originally, we kept the parts
>together, but later that got "optimized", and we ended up subject to a
>kind of confused deputy.  If we'd understood capabilities better, we
>never would have made that particular error.

In my experience the most common sorts of mistakes using capabilities
come from mistakenly believing that they need more features to work
well.  That's one of the things I particularly like about the "Capability
Myths" paper, that it ties together very wide time periods (back to
Dennis and Van Horn) and suggests that a common and simple
"capability" model suffices for all capability uses.  I personally believe
that even "partitioning" is a feature that isn't needed for the base
functioning of the capability model (e.g. as for the Internet/Web), but
I guess that's something we will have to see as these technologies
roll forward.

--Jed http://www.webstart.com/jed/