Wednesday, January 17, 2007

Living Software

Steve Yegge's latest essay, The Pinocchio Problem discusses how liveness, or the QWAN, is what makes software great. Like many great ideas, his is a crystallization of something that one already knows (e.g., Emacs is uniquely amazing) but somehow cannot quite formalize. In retrospect, of course, it seems almost obvious. Now that I've read his essay, I can think of a few other pieces of software that are alive.

One is Erlang, one of the two programming languages in existence that I wish I could use professionally but can't (the other is Common Lisp). Erlang's live code migration feature allows upgrading code in a running system (such as an ATM switch, with tight availability requirements) in an extremely robust way. As far as I know, no other programming language (even ones that support something like this, such as Common Lisp) have sufficient support built in to really be able to do this safely in production systems. It also has a REPL, and is still the only reasonably well-supported language (i.e., one could consider using it without being laughed off the face of the planet) that actually tackles concurrency well.

Another is, well, any serious relational database. Although much maligned, SQL is actually a cool programming language. And the development environment provided by the database is actually a REPL and is much better than your typical dead IDE. Every time I have to do serious database programming, I actually find myself enjoying the interactivity of the experience. Databases tackle some really thorny problems (such as persistence) that most programming languages treat as a secondary concern, unworthy of treatment directly in the language itself. This is one of the major failings of most programming languages. The C# team seems to be at least trying to remedy this with LINQ in C# 3.0, although I don't know too much about it.

So I guess in addition to the qualities that Steve mentions that make software alive, I would add support for concurrency (or, more generally, explicit support for dealing with the passage of time), and persistence. There are probably a few other qualities as well (such as dealing with uncertainty, but it's too late at night to write any more about that).

Avoiding fallback in distributed systems

As previously mentioned , I was recently able to contribute to the Amazon Builders' Library . I'd also like to share another post t...