Kernighan's Lever
linusakesson.net
linusakesson.net
Advising against writing complex code is not advising against learning.
The person who solves a hard problem correctly using simple code has generally spent more time learning than the person who solves it using complex code.
But I think the main point stands. There's an old saying that doing a 60 minute presentation is easy, doing one in 15 minutes us hard. In other words writing "clever" (complicated) code is easy. Distilling it down to something simple is hard.
So the final result of any coding might be "complex", "simplified from complex", or "optimized from simple".
The first and third iterations are superficially similar, although likely different in quality.
He means code that appears indecipherable at first glance, but then once you see how it works, you're enlightened. Simple and efficient code can be "clever".
Example: convert RGB to HSV. If you look around for a formula, you'll likely find one that starts so:
cmin = min(r, g, b);
cmax = max(r, g, b);
Legit in some cases. But for usual business software, code is for humans (compiler will make machine code intended for the machine)
It is not just performance. A minimal component gives you flexibility: you may make the whole system performant or you may trade extra performance to reach a different goal, such as robustness or composability. It is a more fundamental principle, common to construction in general: a thing should do all it has to do and should not do anything more.
Valid if needed, but exists
Funny thing: in Python code I've had a few occasions where I needed both quotient and remainder of an integer division, so naturally I used `divmod` which under the hood can exploit the exact sort of overlap you describe. I get the impression that relatively few Python programmers are familiar with `divmod` despite it being a builtin. But also it really doesn't end up mattering anyway once you have to slog through all the object-indirection and bytecode-interpretation overhead. (It seems that it's actually slower given the overhead of looking up and calling a function. But I actually feel like invoking `divmod` is more intention-revealing.)
In the original quote, “clever” refers to the syntax, where they way the code was constructed makes it difficult to decipher.
I believe your interpretation (and perhaps the post’s, as well) is about the design. Often to make a very simple, elegant design (what pieces exist and how they interact) you need to think really hard and creatively, aka be clever.
Programming as a discipline has a problem with using vague terms. “Clean” code, “clever” code, “complex” code; what are we trying to convey when we talk about these things?
I came up with a term I like: Mean Time to Comprehension, or MTC. MTC is the average amount of time it takes for a programmer familiar with the given language, syntax, libraries, tooling, and structure to understand a particular block of code. I find that thinking about code in those terms is much more useful than thinking about it in terms of something like “clever”.
(For anyone interested, I wrote a book that explores the rules for writing code that is meant to reduce MTC: The Elements of Code https://elementsofcode.io)
As you learn more techniques and more data structures the "cleverness" creeps into your code. To the degree that the cleverness might have a complexity cost, sometimes the cost may be worth it—perhaps not always though.
Naive-you would have struggled to understand some of the shortcuts and optimizations you are leveraging.
But then still more-experienced you revisits the more clever code with years now to have both written and attempted to debug such code. You may now begin to eschew the "clever" to the degree its cleverness makes the code harder to understand or debug. You might swear off recursive code for example—breaking it into two functions where the outer one runs a loop of some sort that is easier to set a break-point in and unwind a problem you were seeing. Or you might now lean more on services provided by the platform you are programing for so you don't have to have your own image cache, your own thread manager, etc.
I feel like in that last stage, most-experienced you may well be writing code that naive-you could have understood and learned from.
I take away two ideas:
1. Always be learning. I think everyone believes this, but we often come up with plausible reasons to stick to what we know. This is a good reminder that we should fight that impulse and put in the effort to learn.
2. Always be fearless. This, I think, is the key insight. Fear is easy. We fear the unknown, whether they be APIs or someone else's code. We fear errors, particularly when they have real-world consequences. And we fear complexity, because we think we might not be able to deal with it. But the opposite of fear isn't recklessness, it's confidence. We should be confident that we will figure it out. And even if we don't figure it out, we should be confident that we can revert the code. Face your fears and grow.
I think a better assessment of how well you've evolved as a programmer is how simple you can make the code. It takes real intelligence and flair to simplify the problem as much as possible, and then write the code to be boringly simple and easy to follow by a junior developer or AI agent.
If you're wielding increasingly clever programming techniques, then you're evolving in the wrong direction.
Let me tell you about a key method in the root pricing class for the derivs/credit desk of a major international bank that was all very clever ... and wrong ... as was its sole comment ... and not entirely coincidentally that desk has gone and its host brand also...
At my job we're disqualifying candidates who don't use enough unnecessary classes. I didn't use them, but they proceeded with my interview because I happened to use some other tricks that showed good knowledge of C++. I think the candidate who just wrote the code to solve the task was the best solution, but I'm not in charge of hiring.
Without revealing the actual interview task, let's pretend it was to write a program that lowpass filters a .wav file. The answer we're apparently looking for is to read the input into a vector, FFT it, zero out the second half, unFFT it, and write the output file. And you must have a class called FFT, one called File, FrequencyDomainFile, and InverseFFT. Because that's simple logical organization of code, right? Meanwhile, the actual simple way to do it is to open the input and output files, copy the header, and proceed through the file one sample at a time doing a convolution on a ring buffer. This latter way involves less code, less computation, less memory, and is all-around better. If you think the ring buffer is too risky, you can still do a convolution over the whole file loaded into memory, and still come out ahead of the FFT solution.
But if you do it this way, we think you didn't use enough abstraction so we reject you. Which is insane. Some time after I got this job, I found out I would have also been rejected if not for a few thoughtful comments, which were apparently some of the very few signals that "this guy knows what he's doing and has chosen not to write classes" rather than "this guy doesn't know how classes work."
I think you've unwittingly bought into your hiring team's fallacy that classes are somehow essential to "abstraction". They are not. Wikipedia:
> Abstraction is the process of generalizing rules and concepts from specific examples, literal (real or concrete) signifiers, first principles, or other methods. The result of the process, an abstraction, is a concept that acts as a common noun for all subordinate concepts and connects any related concepts as a group, field, or category.[1]
The fundamental abstraction in computer programs is the function. A class is principally a means of combination that sometimes incidentally creates a useful (but relatively complex) abstraction, by modeling some domain object. But the most natural expression of a "generalized rule" is of course the thing that takes some inputs and directly computes an output from them.
Of course, we also abstract when we assign semantics to some part of the program state, for example by using an enumeration rather than an integer. But in that case we are doing it in reverse; we have already noticed that the cases can be generalized as integers, and then explicitly... enumerate what it is that we're generalizing.
(The reason that "FFT" etc. classes are so grating is that the process of that computation hardly makes sense to model; the input and output do, but both of these are just semantic interpretations of a sequence of values. You could staple a runtime "time-domain" or "frequency-domain" type to those sequences; but the pipeline is so simple that there is never a real opportunity for confusion, nor reason for runtime introspection. I almost wonder if the hiring team comes from a Java background, where classes are required to hold the code?)
If I were writing the convolution, it would still probably involve quite a few functions, because I like to make my functions as short as feasible, hewing closely to SRP. Perhaps the ring buffer would be a class — because that would allow a good way to separate the logic of accessing the underlying array slots that make the ring buffer work, from the logic of actually using the ring buffer to do the convolution.
(I'm not sure offhand what you'd need to convolve with to get the same result as "zeroing out the second half" of the FFT. I guess a sinc pulse? But the simple convolutions I'd think of doing to implement "low-pass filter" would certainly have a different frequency characteristic.)
We have given extra points to a candidate for having an FFT class even though it should obviously be a function. And the comments clearly indicated that candidate simply thought everything should be a class and was skeptical of things not being classes.
But even in complex applications, there's still truth to the idea that your code will get simpler over time. Mostly because you might come up with better abstractions so that at least the complex bit is more isolated from the rest of the logic. That way, each chunk of code is individually easier to understand, as is the relationship between them, even if the overall complexity is actually higher.
But I wouldn't want my OS written like that. In engineering code, the only benefit of cleverness is better performance, and the risk is unreliability. My previous computer was a lot slower and it already did everything I need, so I'm willing to sacrifice a lot of performance for reliability. Most software is written so wastefully that it's usually possible to make up for the lost performance without cleverness anyway.
Thanks. I somehow ignored the URL and the sidebar, and only now made the connection that OP is by the guy who does all that ridiculous C64 tech demo stuff (especially the music).
It's one of the things people say when they don't like some piece of code, but they also can't justify it with a more in-depth explanation on why the cleverness is unecessary/counter-productive/etc.
Truth is, we need "clever code". Lots of it. Your OS and browser are full of it, and they would suck even more without that. We also need people willing to work on things that are only possible with "clever code".
From this point of view, the idea of the Lever makes sense. The quote also works for criticizing clever code, as long as we follow up with concrete justification (not being abstract about some general god-given rule). In a world where _some clever code is always required_, it makes sense that this quote should work for both scenarios.
Also, a "computer" was a human back then, not a machine.
I'm not clear on if the term "programming" had been invented at that time or not.
> program(v.)
> 1889, "write program notes" (a sense now obsolete); 1896 as "arrange according to program," from program (n.).
> Of computers, "cause to be automatically regulated in a prescribed way" from 1945; this was extended to animals by 1963 in the figurative sense of "to train to behave in a predetermined way;" of humans by 1966. Related: Programmed; programming.
and
> computer(n.)
> 1640s, "one who calculates, a reckoner, one whose occupation is to make arithmetical calculations," agent noun from compute (v.).
> Meaning "calculating machine" (of any type) is from 1897; in modern use, "programmable digital electronic device for performing mathematical or logical operations," 1945 under this name (the thing itself was described by 1937 in a theoretical sense as Turing machine). ENIAC (1946) usually is considered the first.
The term "debug" also dates to 1945 per Etymonline, but Wikipedia also claims
> The term bug, in the sense of defect, dates back at least to 1878 when Thomas Edison wrote "little faults and difficulties" in his inventions as "Bugs".
> A popular story from the 1940s is from Admiral Grace Hopper.[1] While she was working on a Mark II computer at Harvard University, her associates discovered a moth stuck in a relay that impeded operation and wrote in a log book "First actual case of a bug being found". Although probably a joke, conflating the two meanings of bug (biological and defect), the story indicates that the term was used in the computer field at that time.
So the metaphorical sense previously existed, but was relatively new as applied to computers (since doing anything with computers at all was relatively new). And "computer" did refer to a human, but the modern sense was in the process of being established during the literal-bugs-in-vacuum-tubes era.
You could counter that the word “clever” only applies to hard-to-debug code, but that makes the whole statement rather vacuous, no?
It's worse than that. It might not be you who has to debug it, but someone else. Maybe after you left the company already. Maybe at 3AM after a pager alert in production ..
Caveat: in collaborative/prod contexts you sometimes trade cleverness for maintainability, but if you always do that, you skip the lever.
This article can be summarised in one word: learning. I've noticed over the years that there seems to be a growing divide amongst programmers, between those who believe in learning, and those who don't (and actively try to avoid it); unfortunately the latter has become a majority position, but I still try to show others this article when they don't understand code that I've written and would rather I stoop to their level.
A look around the site at what else he has accomplished, should be enough evidence that he isn't just a charlatan, unlike some others who have made a consulting career out of spouting pompous hot air about methodology.
I am also comfortable with the closing comments that you can't always dumb down your code or you stagnate and never learn new tricks/techniques. It is a good thing to keep in mind.
But I have also seen people waste a lot of their (and others') time trying to be clever in ways which as an outsider from additional context I have I can anticipate won't pan out. And I've let it slide and watched it end "not-well", leading to long unnecessary debugging cycles, missing deadlines, and creating boilerplates of YAGNI abstraction and complexity that didn't "make the easy things easy and the hard things possible" but instead made the easy things complicated.
I myself have been accused of that when trying to design optimal "scalable" architectures up front. And I myself have patched over inherited "clever" things with flaws that I handled by adding yet more incremental "cleverness" when, N years later I wish I had just cut the knot of Gordian complexity on day 1.
I think Kernighan's Law is perhaps best applied as a cautionary question to ask yourselves or others in the journey: are you getting too clever, and can you (and others around you) really debug the cleverness you are pursuing?
Complexity and cleverness may be needed, but have you considered re-approaching the problem from a standpoint that minimizes the need for cleverness?
Put another way, there is cleverness that brings "simplicity of code" that does not bring "simplicity of debugging or maintenance" by yourself or others. It's wise to be aware of that.
I view cleverness as somewhat like "innovation tokens"... you should "pick a small handful" of them strategically but not overuse them. I don't see that caution in a pure statement of "Kernighan's lever".
Also seemingly tacitly ignored in the poster's perspective is any acknowledgement that software is, or can be in a huge chunk of scenarios, a "team sport". It's all fine for you to get more clever by pushing yourself, but if you don't transfer your knowledge/cleverness to the broader development+support group, it isn't good for the organization, and perhaps not even you if you consider your code's value proposition will itself harden and stagnate and get refactored out.
(Of course, for some programmers, that's a virtue; write your code in an obscure language/style so that nobody else will take credit or touch it and mess it up. I literally had an acquaintance who, sensing in me a similar competence (or elitism?), boasted to me about his cleverness in doing this at his workplace. I was intrigued, but silently not impressed.)
Yes Kernighan was trying to pass along advice to future programmers about what he thinks is how to reduce unnecessary effort, and yes at the same time spending time debugging difficult code often/ideally increases your skill and avoiding that effort might mean you miss out on developing those skills. Both things are true, so how does one decide which way to go? Of course it depends on your goals, but it’s also worth asking what the opportunity cost is. What if instead of doing battle with complexity and debuggers, you could instead pick up different skills?
It is possible that people should deliberately ignore Kernighan’s advice, and use clever code in order to gain the skills and experience needed to see that Kernighan was right. ;) It’s also possible that spending that valuable time learning how to scale to larger systems would pay off. Or, for some people, spending less time coding and more time devoted to other pursuits.
The ‘stopped evolving’ comment seems like it might be designed to stir the pot, but the best programmers I’ve ever known tend to work hard at reducing complexity by thinking hard about dependencies and about how to write large systems. They don’t necessarily shy away from high performance tricks or hard problems. It’s possible that what a young programmer means by “clever” and what a very seasoned programmer means by “clever” aren’t the same thing at all. https://www.teamten.com/lawrence/writings/norris-numbers.htm...
> You effortlessly wield clever programming techniques today that would've baffled your younger self. (If not, then I'm afraid you stopped evolving as a programmer long ago.)
... Perhaps if we allow that "clever techniques" can yield simpler results than my former self did.
Millions upon millions of C code, over decades, controlled (and still control) things around you that would kill you, or similar catastrophic failure. Cars, microwaves, industrial machinery, munitions, aircraft systems ... with so few errors attributable to C that I can only think of one prominent example.
So sure, you can get bugs written in C. In practice, the development process is more important to fault-reduction than the language chosen. And yes, I speak from a place of experience, having spent considerable parts of my career in embedded systems.