I went to OOPSLA 2008 this year, but I didn't do a daily event-log as I have done sometimes in the past. If you felt I let you down... I'm sorry for you.

One of the events I really enjoyed was the Refactoring Workshop that took place on Sunday, coordinated by Danny Dig and Andrew Black. One of the neater demos was a guy who used to work for DevExpress, showing some of the tools they built to make various refactorings more discoverable, and useable.

One of the things that struck me in this workshop as well as some other presentations, is how valuable having a simple syntax is. The guys working on the Eclipse refactorings for Java have an object model of the AST that has 100+ node types. Having been working on the RB formatter lately, with its AST object model which has 11 concrete node types, I can see how fortunate I am. When you have 100+ node types, whatever you're doing with it, whether it is refactoring or pretty-printing, the number of combinations becomes a geometric complexity. That was born out in some of the side discussions. Apparently, while many of the basic refactorings that we have in the Refactoring Browser are enjoyed by the Eclipse/Java folks, there's a long list of bugs of various edge cases where they don't work. This is sad and unfortunate, but it's not surprising when your AST is that complex.

I'm impressed with Smalltalk's unique development history. Most languages are developed in file form first. And as problems are encountered, syntax is layered on, like so many colors of greasy frosting. And then after all of that has started to take root, the tool makers come along and have to make sense of it. At least in the early days, much of the Smalltalk evolution was a balanced breed of evolution in both realms, so they evolved together. I have no doubt that tool development is one of the things that kept the pressure to keep the language simple. When the AST is simple, the ability to put new ideas together is much less hindered. Personally I think the principles abstract. It's not just AST simplicity that is valuable, but any model simplicity. It makes for better tools.