The best Lisp

:: computer, language, lisp

People sometimes ask which is the best Lisp dialect? That’s a category error, and here’s why.

Programming in Lisp — any Lisp — is about building languages: in Lisp the way you solve a problem is by building a language — a jargon, or a dialect if you like — to talk about the problem and then solving the problem in that language. Lisps are, quite explicitly, language-building languages.

This is, in fact, how people solve large problems in all programming languages: Greenspun’s tenth rule isn’t really a statement about Common Lisp, it’s a statement that all sufficiently large software systems end up having some hacked-together, informally-specified, half-working language in which the problem is actually solved. Often people won’t understand that the thing they’ve built is in fact a language, but that’s what it is. Everyone who has worked on large-scale software will have come across these things: often they are very horrible, and involve much use of language-in-a-string1.

The Lisp difference is two things: when you start solving a problem in Lisp, you know, quite explicitly, that this is what you are going to do; and the language has wonderful tools which let you incrementally build a series of lightweight languages, ending up with one or more languages in which to solve the problem.

So, after that preface, why is this question the wrong one to ask? Well, if you are going to program in Lisp you are going to be building languages, and you want those languages not to be awful. Lisp makes it it far easier to build languages which are not awful, but it doesn’t prevent you doing so if you want to. And again, anyone who has dealt with enough languages built on Lisps will have come across some which are, in fact, awful.

If you are going to build languages then you need to understand how languages work — what makes a language habitable to its human users (the computer does not care with very few exceptions). That means you will need to be a linguist. So the question then is: how do you become a linguist? Well, we know the answer to that, because there are lots of linguists and lots of courses on linguistics. You might say that, well, those people study natural languages, but that’s irrelevant: natural languages have been under evolutionary pressure for a very long time and they’re really good for what they’re designed for (which is not the same as what programming languages are designed for, but the users — humans — are the same).

So, do you become a linguist by learning French? Or German? Or Latin? Or Cuzco Quechua? No, you don’t. You become a linguist by learning enough about enough languages that you can understand how languages work. A linguist isn’t someone who speaks French really well: they’re someone who understands that French is a Romance language, that German isn’t but has many Romance loan words, that English is closer to German than it is French but got a vast injection of Norman French, which in turn wasn’t that close to modern French, that Swiss German has cross-serial dependencies but Hochdeutsch does not and what that means, and so on. A linguist is someone who understands things about the structure of languages: what do you see, what do you never see, how do different languages do equivalent things? And so on.

The way you become a linguist is not by picking a language and learning it: it’s by looking at lots of languages enough to understand how they work.

If you want to learn to program in Lisp, you will need to become a linguist. The very best way to ensure you fail at that is to pick a ‘best’ Lisp and learn that. There is no best Lisp, and in order to program well in any Lisp you must be exposed to as many Lisps and as many other languages as possible.

If you think there’s a distinction between a ‘dialect’, a ‘jargon’ and a ‘language’ then I have news for you: there is. A language is a dialect with a standards committee. (This is stolen from a quote due to Max Weinrich that all linguists know:

אַ שפּראַך איז אַ דיאַלעקט מיט אַן אַרמיי און פֿלאָט

a shprakh iz a dyalekt mit an armey un flot.)

  1. ‘Language-in-a-string’ is where a programming language has another programming language embedded in strings in the outer language. Sometimes programs in that inner programming language will be made up by string concatenation in the outer language. Sometimes that inner language will, in turn, have languages embedded in its strings. It’s a terrible, terrible thing.