Mark Miller riffs off some comments Joshua Bloch has made about the the growing complexity of Java, and cloncludes that they made a basic error early on: by keeping the language "simple" - no closures, no real meta system, etc - they've actually made any growth that comes (generics) a nightmare of complexity:

I think there is a fundamental contradiction in goals between what Bloch says he prefers about Java, and the growing problems of computing. He says that adding features to a language increases its complexity, and he makes a good case for that. Yet, how is Java going to solve new problems in the future? His answer is to add more features to the language. Granted they're focused, and not open-ended to allow further expansion, but they're still feature additions all the same. As is illustrated by Squeak, library-defined structures would actually help reduce the number of features in the language. Even though Squeak can do more things concisely than Java can, the feature set of the language is very small. As a point of comparison, Lisp's language feature set is even smaller, yet it's more capable of handling complex problems than Java. They accomplish this by taking what Java does as a built-in feature, and use message-passing (Squeak), closures (Squeak, Lisp, others), an extensive meta system (Squeak, Lisp, others), named functions (Lisp, Scheme), and macros (Lisp) instead.

Smalltalk has a tiny "feature set", but immense power - any decent sized application starts to resemble a DSL over time, because that power is reserved to the developer. That simply doesn't happen in Java; instead, a decent sized application just grows warts.

Read the whole thing, because Mark makes a number of points worth thinking about.

Technorati Tags: smalltalk, dynamic language, development