Here is a great example of a column I wish I'd never written. It presents a new control structure to deal with the complexity of the Composite/Wrapper pattern used in the user interface code in Visual Works. I wrote the column, sent it off, then had a quick talk with David Liebs about the subject. “So you need to talk to a particular wrapper, eh? Why don't you keep a pointer to the wrapper you need to talk to and give it a useful name? Then you can just send it messages.” Duh…

The other aspect of the column I regret is that it seems to condone adding new control structures to Smalltalk. The meta facilities of Smalltalk are extremely powerful, but 99.99 percent of the time, you shouldn't use them. Oh, they'd get the job done alright, but at what cost?

When you write something like #wrapperSend:, you have to take off your application-programmer hat and put on your language-designer hat. The responsibilities of the language designer are ten times as great as those of an application developer. Whatever you do, you have to implement it and test it, sure, but you also have to document it like a new language feature, extend the programming environment to handle it smoothly, and worst of all, maintain it through all future releases of the image.