A case study between Python and Go. In Go, a common complaint from newly-minted gophers who come from another language is the error handling pattern: result, err := DoSomething() if err != nil { return nil, err } another, err := SomethingElse(result) if err != nil { return nil, err } // We can be a bit more terse if we don't need to save any // variables outside of the if-scope: if err := MoreWork(another); err != nil { return nil, err } ... Any time we write similar-looking code over and over, that’s considered . boilerplate A key principle of programming that we all learn is (DRY). In a way, that’s the entire point of programming: To automate a task that we keep repeating manually. Who wants to keep artisanally sorting text files each morning when we could have a program do it for us faster and more reliably? Similarly, why should we keep writing some code pattern over and over when we should be able to abstract it away into some terser syntax? Don’t Repeat Yourself Proportional work After writing Python code for over a decade and becoming quite proficient at it, I’d take any opportunity to write terse code. It feels good to be clever and , right? more code means more bugs One thing I didn’t notice until I switched to writing Go as my primary language: The done in Python code is hugely variable. proportionality of work Consider these two lines: What do I mean by proportionality? a = some_variable + 42 b = sum((j if j % 2 else 0) for (j, k) in results if k) The second line does more work than the first line. The first line is a simple integer addition, while the second line has a loop and a variable split and two conditional branches and an accumulator on a generator. a lot In Python (and many other languages), it’s very easy to unintentionally hide of work in a single line of built-in syntax. Some lines do almost no work, other lines do a moderate amount of work, other lines do what I’d describe is tens of lines worth of work—a lot of work. It varies a lot, even in beginner code. tons In Go, I’ve found that this variability is much more diminished. A loop is always a loop, an error check is always an error check (rather than implicit exception propagation), a logic branch is always a branch. This is the most reasonably-terse way I could write the above in Go: b := 0 for _, r := range results { j, k := r[0], r[1] if !k || j % 2 != 0 { continue } b += j } Is this good? I believe it is. I believe this code is more proportional to the complexity of the work that it’s doing. It’s also possible to write Python in a similar level of proportionality, why don’t I? When reading proportional code, it’s easier to notice where the interesting bits are. It’s easier to notice where the bulk of the work is being done, and focus our review on that. Maybe this is subjective but I posit that the code’s looks more like what it is. shape We can continue to argue that it’s easier to what the work is actually doing, but that’s not a point I want to make. Instead, I want to make a different point: It’s easier to tweak what the code is doing. understand Opportunity for correction After writing of code in Python, and writing a moderate amount of code in Go, I noticed an interesting phenomenon: I tweak code very differently in Go. a lot Programming is a very iterative process. As we implement our solution, our mental model of the problem evolves. We often go back and change some assumptions, tweak our data structures, remove obsolete crutches and stubs, make error handling more robust, and so on. I noticed that when I wrote Python code, I would often to avoid messing up a beautiful terse set of lines by moving logic to places other than where it might naturally belong. I would unconsciously move complexity from where it belongs to a tangential place that made the tweak more complex than it should have been. go out of my way For example: Rather than refactoring a try/except block into separate appropriately-layered method calls or multiple try/except phases, I’d introduce a variable to maintain error state that gets handled in a finally block. ( .) An example of this pattern in urllib3 When I write Go, I noticed that my tweaks almost always go exactly where they naturally belong. My boilerplate doesn’t stay untouched! For example: I’ll usually write lots of Go boilerplate in my first iteration, but by the time I look at my code closer to release I notice that it’s no longer looking like boilerplate. By the time I’m done—adding more recovery scenarios, better logging, augmenting errors with more context, handle more edge cases—most of my boilerplate gets a lot more interesting. I did not realize that good boilerplate could be fertile ground for iteration. When a friend was reading through the , he was surprised that I implemented my own type. I explained that while Go doesn’t have a built-in it’s just a few lines of boilerplate to make your own on top of a . In fact, I noticed that evolved to be fairly specific to how it was being used. In retrospect, I’m glad that I was tweaking my own implementation iteratively rather than spending time working around whichever limitations a generic library might have had. source code for ssh-chat Set Set, map my version of Set I realize now that disproportional code makes me unintentionally dance it like tiny black holes of complexity. Having boilerplate helps balance out the ease of modifying it with the amount of work that code is actually doing—whether it’s error handling or looping or whatever else. around proportional Proportional code creates less complexity I am not claiming that more lines of proportional code is less buggy than fewer lines of disproportional code. I am not claiming that all boilerplate is good: boilerplate can be disproportional in the other direction, too—where it’s doing far work than we’d expect. less My claim is: For non-trivial projects, proportional code is not cumulatively more lines than disproportional code. When we write disproportional code (or code that varies a lot in proportionality), we shift natural complexity into unnatural places which is an unexpected sacrifice. Our beautiful one-liners are short indeed, but the rest of the project suffers in other places to pay off that purchase. By the time we’re done iterating and fleshing out our project, we might not be winning as much as we’re losing from getting rid of boilerplate indiscriminately. boilerplate can help us iterate in a less constrained way. Proportional
