Paul Graham - Essays

The true test of a language is how well you can discover and solve new problems, not how well you can use it to solve a problem someone else has already formulated. These two are quite different criteria. In art, mediums like embroidery and mosaic work well if you know beforehand what you want to make, but are absolutely lousy if you don't. When you want to discover the image as you make it-- as you have to do with anything as complex as an image of a person, for example-- you need to use a more fluid medium like pencil or ink wash or oil paint. And indeed, the way tapestries and mosaics are made in practice is to make a painting first, then copy it. (The word "cartoon" was originally used to describe a painting intended for this purpose).

What this means is that we are never likely to have accurate comparisons of the relative power of programming languages. We'll have precise comparisons, but not accurate ones. In particular, explicit studies for the purpose of comparing languages, because they will probably use small problems, and will necessarily use predefined problems, will tend to underestimate the power of the more powerful languages.

Reports from the field, though they will necessarily be less precise than "scientific" studies, are likely to be more meaningful. For example, Ulf Wiger of Ericsson did a study that concluded that Erlang was 4-10x more succinct than C++, and proportionately faster to develop software in:

Comparisons between Ericsson-internal development projects indicate similar line/hour productivity, including all phases of software development, rather independently of which language (Erlang, PLEX, C, C++, or Java) was used. What differentiates the different languages then becomes source code volume.

The study also deals explictly with a point that was only implicit in Brooks' book (since he measured lines of debugged code): programs written in more powerful languages tend to have fewer bugs. That becomes an end in itself, possibly more important than programmer productivity, in applications like network switches.

Ultimately, I think you have to go with your gut. What does it feel like to program in the language? I think the way to find (or design) the best language is to become hypersensitive to how well a language lets you think, then choose/design the language that feels best. If some language feature is awkward or restricting, don't worry, you'll know about it.

Such hypersensitivity will come at a cost. You'll find that you can't stand programming in clumsy languages. I find it unbearably restrictive to program in languages without macros, just as someone used to dynamic typing finds it unbearably restrictive to have to go back to programming in a language where you have to declare the type of every variable, and can't make a list of objects of different types.

I'm not the only one. I know many Lisp hackers that this has happened to. In fact, the most accurate measure of the relative power of programming languages might be the percentage of people who know the language who will take any job where they get to use that language, regardless of the application domain.

I think most hackers know what it means for a language to feel restrictive. What's happening when you feel that? I think it's the same feeling you get when the street you want to take is blocked off, and you have to take a long detour to get where you wanted to go. There is something you want to say, and the language won't let you.

What's really going on here, I think, is that a restrictive language is one that isn't succinct enough. The problem is not simply that you can't say what you planned to. It's that the detour the language makes you take is longer. Try this thought experiment. Suppose there were some program you wanted to write, and the language wouldn't let you express it the way you planned to, but instead forced you to write the program in some other way that was shorter. For me at least, that wouldn't feel very restrictive. It would be like the street you wanted to take being blocked off, and the policeman at the intersection directing you to a shortcut instead of a detour. Great!

I think most (ninety percent?) of the feeling of restrictiveness comes from being forced to make the program you write in the language longer than one you have in your head. Restrictiveness is mostly lack of succinctness. So when a language feels restrictive, what that (mostly) means is that it isn't succinct enough, and when a language isn't succinct, it will feel restrictive.

The quote I began with mentions two other qualities, regularity and readability. I'm not sure what regularity is, or what advantage, if any, code that is regular and readable has over code that is merely readable. But I think I know what is meant by readability, and I think it is also related to succinctness.

We have to be careful here to distinguish between the readability of an individual line of code and the readability of the whole program. It's the second that matters. I agree that a line of Basic is likely to be more readable than a line of Lisp. But a program written in Basic is is going to have more lines than the same program written in Lisp (especially once you cross over into Greenspunland). The total effort of reading the Basic program will surely be greater.

total effort = effort per line x number of lines

I'm not as sure that readability is directly proportionate to succinctness as I am that power is, but certainly succinctness is a factor (in the mathematical sense; see equation above) in readability. So it may not even be meaningful to say that the goal of a language is readability, not succinctness; it could be like saying the goal was readability, not readability.

What readability-per-line does mean, to the user encountering the language for the first time, is that source code will look unthreatening . So readability-per-line could be a good marketing decision, even if it is a bad design decision. It's isomorphic to the very successful technique of letting people pay in installments: instead of frightening them with a high upfront price, you tell them the low monthly payment. Installment plans are a net lose for the buyer, though, as mere readability-per-line probably is for the programmer. The buyer is going to make a lot of those low, low payments; and the programmer is going to read a lot of those individually readable lines.

This tradeoff predates programming languages. If you're used to reading novels and newspaper articles, your first experience of reading a math paper can be dismaying. It could take half an hour to read a single page. And yet, I am pretty sure that the notation is not the problem, even though it may feel like it is. The math paper is hard to read because the ideas are hard. If you expressed the same ideas in prose (as mathematicians had to do before they evolved succinct notations), they wouldn't be any easier to read, because the paper would grow to the size of a book.

A number of people have rejected the idea that succinctness = power. I think it would be more useful, instead of simply arguing that they are the same or aren't, to ask: to what extent does succinctness = power? Because clearly succinctness is a large part of what higher-level languages are for. If it is not all they're for, then what else are they for, and how important, relatively, are these other functions?

I'm not proposing this just to make the debate more civilized. I really want to know the answer. When, if ever, is a language too succinct for its own good?

The hypothesis I began with was that, except in pathological examples, I thought succinctness could be considered identical with power. What I meant was that in any language anyone would design, they would be identical, but that if someone wanted to design a language explicitly to disprove this hyphothesis, they could probably do it. I'm not even sure of that, actually.