The Second Language


If you were to ask me to summarize software engineering in one word, “tradeoffs” is the one I’d pick. We make tradeoffs every day: we choose one language over another based on some requirement or constraint, we opt for fast and cheap even though we know sacrificing quality means we’re incurring technical debt, we select a technology that scales to the extent we need at the cost of a steeper learning curve or implementation overhead.

We don’t necessarily consider learning a new language to be a lesson in tradeoffs, but I think the second programming language you learn is exactly that. A programming language fundamentally represents how an engineer or group of engineers thinks about solving problems, and since solving problems is all about tradeoffs, it’s impossible for you to learn a language without coming to grips (consciously or not) with the tradeoffs its creators made (consciously or not).

If you ask your friend who writes Haskell what a monad is (as you eventually must), you might discover that a monad is a monoid in the category of endofunctors, or a programmable semicolon, or a burrito, or some similarly unhelpful combination of words. While any of these might be true, the fact is that none of them tells you what a monad is for. (If you’re curious, this post by James Coglan is the best explanation of what monads are for that I’ve ever read.)

My first languages were JavaScript and Ruby: dynamically typed, object-oriented with a strong functional influence, and nothing preventing you from doing whatever silly things with state you so desired. When I first learned Clojure, I thought the tricky part would be balancing parentheses and composing programs with functions rather than objects. In fact, the hardest part for me was keeping concurrency primitives (agent, atom, and ref) straight; I tried memorizing their definitions and reading documentation, but to no avail. Finally, one day, I made myself go through a couple of examples in The Joy of Clojure. Chapter 11 covers mutation, and in particular goes through the “ideal cases” for each state management primitive.

As I worked through the examples, I discovered that atoms were great for managing independent stateful events (which I later used when swapping a default image for a broken <img /> tag), refs were ideal for coordinating multiple events via transactions (think of transferring money from one bank account to another), and agents were best suited to asynchronous and I/O operations (say, firing an ad beacon). The kicker was that all this was clear as day when I went back and read the blog posts and documentation that had seemed impenetrable to me before, but I couldn’t wrap my head around it until I stopped caring about what things were and focused on what they were for. And once I understood that, I also began to understand the tradeoffs between these tools: the benefits and drawbacks of using each.

Learning a second programming language is useful for all sorts of reasons, but I think the main benefit is this: via an understanding of what things are for (rather than simply what they are), you discover why a language’s inventors and users do things the way they do, and in reaching that understanding, you gain a deeper understanding of the tradeoffs involved.