C is this brilliant awful crazy-successful hack, yak-shaved into existence just to bootstrap an entire brand new OS (UNIX) in record time; subsequently Peter-Principled into domains it really has no business being in (basically anything userland). This is why, fifty years on, we’re still fighting memory corruption and security holes and generally doing an awful job of parallel computation and everything else C is not very good at.
Everyone should learn C, if only to understand where we’ve come from, where we are now, and where we should be trying to go in future.
Languages like Lisp and Forth and Smalltalk are fundamentally different beasts to Algol’s descendents.
Most developers, having being raised in a pure C/Algol culture, give no thought to the base complexity and limitations inherent in those languages; they just assume that is how computation is.
What the Lisps do is reveal all that complexity and limitations to be artificial constructs, imposed by the designers of those languages, whether as convenience to implementation (as in C, where features like flow control statements map closely to underlying machine instructions) or simply because that’s what they’re already used to (c.f. 90s scripting languages like Python and JavaScript, born on the back of mature, ingrained C culture).
..
Here’s a question to get you thinking: Why do Algol family languages have hardwired flow control statements? Not because that’s how programming needs to be, but because that’s how they’ve always done it. Often there’s some special-case behavior involved, e.g. deferred evaluation of operands, as in the case of conditional (`if`) statement bodies, and hard-wiring that particular special-case into the language guts is a quick way of addressing that need.
Heck, even traditional Lisps (1 & 2) are guilty of hardwiring hackery (don’t let its uniform syntax fool you: Lisp is ridden with special forms, including conditionals). You have to go to something like John Schutt’s Kernel language to see that problem solved correctly, by extracting the special-case behavior (lazy evaluation) into a general, reusable feature. Schutt calls it “vau”, but it’s just a way to dictate how and when arguments should be evaluated at the language level, instead of baking that behavior into the language implementation where it can be neither modified nor reused.
What’s the upshot? Well, first, you realize that 90% of what Algol languages treat as core features aren’t “core” at all. They can be pulled out, generalized, and expressed as plain old composable functions.
Suddenly your “core language” collapses from something big and complicated, with dozens of special-case features baked in, each with its own special syntax and behavioral rules, down to something ridiculously simple: an evaluator, a mechanism for composing behaviors (functions with vau support), and maybe some primitive datatypes to get you started (although even those can be pushed out into libraries if you want), … and that’t it. It is the mother of all refactorings, and it reveals the simple truth.
All those features and complexity the Algols had us believe were essential prerequisites to computation and programming… all gone, pushed out to external libraries which the user can freely import, modify, replace, or ignore as they need. Once you realize how small, simple, and elegant computation can (and should) be, you’ll wonder what the hell all that vast intricate pomp and ceremony is really for, and who it really serves.
Programming languages are big and complicated not because that’s how they need to be but because nobody tries to make them small and simple. But as Tony Hoare said:
“There are two methods in software design. One is to make the program so simple, there are obviously no errors. The other is to make it so complicated, there are no obvious errors.”
So if a language already falls in the latter camp, then Dog only imagine all the software subsequently written in it!
This is where C falls outwith its original purpose, and it’s telling that subsequent popular attempts to fix it have done so by adding even more complexity on top, e.g. C++, Java, Swift. (It’s only recently with the likes of Rust that trend has started to reverse, but Rust is still a Complex language compared to a Forth or a Smalltalk.)
..
Now, there is a tendency in Lisp circles to rave about macros as if they’re the True Revelation, but they’re not. The revelation is when you realize you have the ability to reshape a language you are given into the language that you need; a language that precisely and concisely expresses the concepts and behaviors that are of interest and relevance to you in your particular problem space. Lispers describe this as “bottom-up programming”, but the term hardly does it justice.
Find yourself writing “if not TEST do ACTION” a lot? Extract that out into your own first-class “unless TEST do ACTION” command and use that going forward. Want to evaluate commands non-sequentially according to your own [e.g. priority-based] rules? Done. Don’t want loops? Leave them out.
When you write an “Algol” program, the language always controls how your program is evaluated. If you don’t like the way it does it, tough. When you write a “Lisp 3” program, you get to control how your program is evaluated. You aren’t stuck with a fixed language-level set of first-class flow control statements that may or may not meet your needs, plus a weak second-class user-level mechanism for kind of creating your own (C-style functions, with their inherent limitations). There’s only one level: everything is a first-class feature, whether it’s provided by the language or added by you.
..
We often think of C as a “general-purpose language”, but I think this is wrong and unhelpful. We should really think of C as a DSL, one created specifically for bit-twiddling and crude algorithmic number crunching (which is what it’s good at). The true general-purpose languages are the meta-languages, because they have the power to be whatever we want them to be.
Working in C, you always speak in C. In a “Lisp 3”, you can speak in any language you like; you just have to grow that initial minimalist foundation of key building blocks into the language you want, and then off you go. It is the difference between Algorithmic Thinking and Compositional Thinking.
Yes, algorithmic thinking has achieved a lot in the last 50 years, but the software we’ve built that way has a lot of problems—safety, reliability, scalability, learnability, verifiability—that I strongly suspect are inherent to that approach. We really need to try compositional thinking to escape that trap, but we can’t even start to do that while the languages we commonly use (Algol family) are themselves fundamentally crap at compositionality.
--
Lastly, I think Nile (https://news.ycombinator.com/item?id=19844088) requires an honorary mention in parent’s list; not because of the language itself but because of how it’s constructed: not only is it a composed language, but the language it’s written in is a composed language too. If that doesn’t melt your mind at the possibilities, nothing will!