Where Are the Custodians of the Field?
— Jonathan Blow on Twitter
Software productivity feels to me like it has lost its way. ”High-level” programming languages from the 1990s and 2000s have left a burdensome legacy that engineers are ignoring. Despite having all these grand tools and languages, the actual output of our software, in terms of features, capabilities, and robustness, is at a low point. All the while, the number of lines of code that we write is just continuing to grow. I think that, instead, we should be taking a step back and reevaluating our return on investment.
print("Look, ma! No malloc or free calls!") a_type_error_waiting_to_happen = 3 print("The number is: " + a_type_error_waiting_to_happen)
I think we, as an industry, have yet to come to terms with how “easy to get started” does not mean “easy to maintain, extend, and change.” As it turns out, simple isn’t the same as easy. What is the cost of convenience? All the features that make us so productive might actually have downsides, the likes of which Dropbox and Airbnb are now reeling from. It may be that the first time you see the code sharing capabilities of object inheritance or the convenience of weak-typing coercion, you think to yourself, “Oh wow, I’ll be faster with that. Sign me up, coach.” However, these “treasures” of expression are almost trivial, and they don’t necessarily help you paint the big picture. I’ve witnessed programmers go so far as to argue about whether or not a programming language should have semi-colons, as if not physically typing semi-colons truly helps us to solve hard problems. Just look at this headline quote from the Ruby homepage:
A dynamic, open source programming language with a focus on simplicity and productivity. It has an elegant syntax that is natural to read and easy to write.
The authors chose to focus on the syntactic quality of their programming language as its selling point. Jeez, of all the things. I’ve never heard a practitioner say, “If only the syntax of my programming language were better, I’d be able to actually solve this problem.” And yet, your average Ruby programmer won’t quiet down about how productive they are because of the syntax. Perhaps what this imaginary Ruby programmer really meant was they are able to solve problems better or faster because of the syntax, but I wouldn’t exactly call the Ruby grammar simple. If you’re one of these semi-colon-hating Rubyists, I don’t blame you, but I think you’ve been distracted.
To be fair, humans turn out to be bad at bookkeeping or otherwise boring, repetitive
tasks. I do think that some of the cognitive load of programming needs to be removed so that engineers can focus all their mental faculties
on the important or creative aspects. For instance, do I think we should go back to using conditional jump codes instead of
for loops and
if blocks? Fuck no — you can pry those abstractions from my cold, dead fingers. However, I’m not sure that some of our other conventional
approaches are really helping us. In fact, I claim they often end up actively doing the opposite. We gave up the tedious and memory-unsafe
free functions and gained a garbage collector, but then we realized that those pesky GC pauses spiked our tail
latency. Well, fuck it, let’s just work around these
pause times by following some rigorous rules that end up sounding a lot like
the very tedium and error-proneness of manual memory management that we were trying to escape. Maybe it’s time for a new
Technical purity isn’t everything, though. People have argued indefinitely about the best priorities for business, and probably, when you’re just trying to survive and find a market fit, you can’t afford to think too hard about the mounting technical debt and complexity. In this starry-eyed, Silicon-Valley-VC-funded world, time to market is all that matters, right? Just keep hustling, write more code, and hope you’ll make it. But that still doesn’t explain why “non-profit” technical efforts like programming languages have to repeat these same values. And because of that, what happens to your company in the long run when you’re stuck with the aftermath and a 2-million-line Java app?
I know I’m not the only one clamoring for deeper thought in engineering; some are even proponents of literal slow programming. But seriously, folks, where are the custodians of the field? Is “winning” all that matters? Where is the emphasis on quality? Why aren’t there more people looking around and saying, “we messed up when we turned left at Banana St. and Clown Ave.?” My beloved GitHub, open-source culture is especially guilty of this mentality, having become infected with shiny-logo and superficial getting-started-page syndrome. In this spirit, projects like Elm are pronounced dead because of their slow, annual release cadences, as if frequent releases are the only sign of quality. Yet, Elm has one of the best developer experiences out there with outstanding, helpful compiler errors — so much so, that other languages are taking note because deliberate, non-trivial design in service of developer quality of life matters.
I’ve come across objections to modern high-level programming languages that, at the surface, seem similar to mine. These objections call for “back-to-fundamental” practices and re-embrace languages such as C. They also occasionally utilize sexist rhetoric about what “real men” or “real programmers” should do. However, this kind of thinking is elitist and exclusionary. The issue isn’t that we have become “soft” or ”less of a programmer” by using Python or Ruby. Instead, the issue is we have become so distracted by the high demands of businesses and users that we have stopped meaningfully criticizing the very tools we rely on for productivity. Or, as Jonathan Blow identifies it, we are only “pretending” that these high-level tools are meeting those demands. Constraints and insulation are useful, but I think we are hiding the wrong details.
Approachabilty has been a driving force for a good number of the efforts I’m criticizing, and I’m not criticizing approachabilty itself as a valuable goal. Relatedly, I don’t want to discount the real, ethical issue here of how we make tools and programming languages more open and inclusive because we do need to think about how to empower all engineers. In the past, I’ve seen two approaches to improving inclusivity in programming language usage, neither of which I’ve been satisfied with. The first purports that inclusivity can be accomplished through only community policy and other social efforts. However, this approach is incomplete because it doesn’t recognize the tight marriage of technological and people problems. They’re one and the same; don’t be fooled into thinking that technological choices don’t have social impact or that technology is pristine and standalone. The second attempts to make tools “easier” to understand and lower the intellectual barrier of entry. But, this is some weird, patronizing bullshit. How insulting that we can’t trust people (i.e., the historically disenfranchised) to grasp more technical details so we’ll just make it more “approachable.” Yikes. No, expose all the nuances to professionals that they need to solve problems effectively, be clear and honest about the chosen design constraints, and avoid “easier to get started” for its own sake.
Some of what I’m proposing here might require some pretty high buy-in and cognitive shift. We may have to dispose of “trusted” solutions that have quite a bit of mileage on the tires. Which brings me to my final point: In this journey, be brave, fair traveler! Courage might just be taking a step back, questioning, and being vocal about your concerns, even when your complaints have mountains of historical momentum. Has garbage collection really paid off? Perhaps even small, “solved” problems, like the ancient and venerable Unix file descriptor, deserve a critical eye and not-so-gentle reexamination. Is this really the best abstraction for files and sockets, especially when these files and sockets are, in fact, quite different from each other? Do I end up paying for conflating the two later with the same kinds of complexity that the abstraction was supposed to save me from in the first place? Maybe what we mean when we say “high-level,” isn’t so high-level after all.