engine of joy mikel evins

More joy

Programmers inevitably spend a certain amount of our time fighting with our tools. There’s always something wrong with them, always some case that they handle badly, always some work we have to do, not because of the job we’re doing, but because of limitations of the stupid tool.

Programming can be a joy, even programming for a living. I’m lucky in that it’s been a joy for me more often than it hasn’t. But fighting with the tools is not a source of joy. In fact, it’s the opposite. More fighting with tools means less joy.

Less joy is bad.

That’s why I remain a Lisp programmer: the various forms of Lisp have given me by far the best return on investment I ever experienced in terms of joy per unit fighting. I can write code in other languages. I’ve written lots of it. I’ve been paid a lot of money to write a lot of lines of C, C++, Python, shell script, Java, ML, Objective-C…just lots of languages, on lots of jobs, in lots of circumstances. I’m still writing Lisp code because of that joy-to-fighting ratio.

More joy is good.

So what’s my favorite programming language?

Well…

I used to say it was 1992-era Dylan. Dylan was a Lisp that Apple invented. I worked for several years on Apple’s Newton project. Newton was initially written mostly in Dylan, and I got to write a lot of OS code in Dylan. That was a time of high joy in my programming life.

It didn’t last, of course. The story of Newton’s abandonment of Dylan and its other adventures makes entertaining reading, but the short of it is that eventually I had to stop programming in Dylan.

Less joy is bad.

Halfway through 2008, one of the original designers and implementors of Dylan pointed out to me this new Lisp dialect named Clojure:

http://clojure.org/

The reason she noticed it was that it’s a Lisp whose name is a misspelling of the world “closure”, and she’s a partner in a Lisp company named “Clozure Associates”. Fun coincidence!

It turned out that this new Lisp, named Clojure, was pretty good. In fact, the more little projects I ported to it, the happier I became. More joy is good.

Three quarters of a year later, Clojure has become my favorite programming language for personal projects. Whenever I think about writing something, I immediately think about Clojure, and if deployment on the JVM is at all reasonable, then Clojure is probably what I’ll use.

I guess Clojure is my favorite programming language now.

All is not entirely well, though. There’s a lot of joy, but there’s still just a little bit too much fighting.

Every time I do anything significant in Clojure, I inevitably stumble over something that makes me wish I was using Dylan instead.

Bother.

I never used to wish I was using Clojure back in the day, instead of Dylan. Well, of course, Clojure didn’t exist then, but I never wished I was using anything instead of Dylan.

So why don’t I just use Dylan? It’s still around, right?

http://www.opendylan.org/

Well…something happened to Dylan over the past seventeen years. It’s not bad or anything. It just doesn’t excite me that way its ancestral form did. It doesn’t feel like a Lisp anymore. Less joy is bad.

Still, when I use Clojure, I often wish I had Dylan. When I used Dylan, I never wished I had something else. What’s the problem?

Sometimes I’m working on an application that has certain requirements on its data. I’d like to just clearly and concisely write down those requirements. In Dylan (or a zillion other languages) I’d write a type definition (Dylan calls them classes). There’s nothing like that in Clojure.

Now, Clojure is a Lisp. You can write whatever facilities you need, if they’re not there already. So I wrote my own implementation of record types, so that I could write down my data requirements succinctly. Problem solved. Still: more fighting, less joy.

Polymorphic functions are really handy. With polymorphic functions you can easily write abstract protocols and then implement them for various different data types without needing to rewrite any of the protocol code. Very useful, especially when you’re writing some big application that you expect to evolve over time, that you expect to need to encompass new and evolving data types.

Clojure has its own idea about how polymorphic functions should work. I don’t want to get into the details, because it would get long and eye-glazing and I’d probably start to froth at the mouth. Let’s just say that Clojure’s creator and some of its other users believe that Clojure’s approach is a better way to approach polymorphism, whereas I think it’s like someone giggling hysterically as they break all your toys and poke sharp sticks in your eyes.

More fighting; less joy.

So I wrote my own generic functions subsystem. Now I can make APIs that don’t randomly burst into flame when someone writes code that uses them. Once again, it’s nice that Clojure is a Lisp: it gives me the tools I need to patch things like this. Still, less joy continues to be bad.

One of the great things about Clojure is that it’s designed to be hosted on the JVM, and to interoperate seamlessly with Java. That’s quite a useful feature, and Clojure’s Java interface is really easy and convenient. So easy and convenient, in fact, that it’s probably easier to use Java libraries from Clojure than it is to use them from Java.

On the other hand, Clojure sort of has more than one type system, and each of the type systems is sort of not really comprehensive. I mean, on the one hand, every Clojure value is in fact a Java value, but, on the other hand, knowing that two different values are both instances of PersistentHashMap may not be all that helpful if each has different :type metatdata, and your application includes some code somewhere that distinguishes them on that basis.

What exactly is the type of Clojure value? Well, there’s its Java class, which may or may not be a helpful thing to know, depending on context. Then there’s its metadata, if it’s a kind of value that can have metadata, and if it is, there might or might not be a :type in that metadata, and that type might or might not be something your code can use for something useful. It sort of feels like Common Lisp’s types before CLOS came along and absorbed all types into its grand galactic scheme.

And what if you want to make your own types? Well, there isn’t really any Clojure facility for doing that, but it’s a Lisp, right? You can make whatever kind of type system you want. Of course, the more type systems you have, the more you have to keep track of, and none of them are really comprehensive over all possible Clojure values. Well, except Java’s. You can always fall back to Java classes. Except that if you want to define new ones you have to kind of jump through a few hoops.

You know what I’m going to say: more fighting, less joy.

No doubt someone will skim this and assume I’m bashing Clojure. I’m not bashing Clojure. It’s my favorite programming language right now. I just noticed that, hey, I keep wishing it was something else. That’s weird. The last time I was doing a lot of work in my favorite programming language, I didn’t ever wish it was something else. What’s up with that?

Maybe it’s that Clojure is still very young and is evolving rapidly (except, hmm…Dylan was really young back then and evolving very rapidly, too…)

Or maybe it’s that Clojure approximates the Lisp I really want, without actually being it. If so, then Dylan was apparently a closer approximation, but one that grew farther away over time, rather than closer.

I guess I’ll have to do some thinking about what both Dylan and Clojure are approximations of.