In 2009, I started an internship at a Los Angeles-based audio/video post production studio. The audio engineer I was learning from was an avid musician just like myself, and we quickly bonded over our mutual interest of recording and producing music. One day we started talking about audio equipment, and I mentioned how I thought it was ridiculous that anyone could justify spending thousands of dollars on a vintage piece of outboard equipment. After all, when you’re producing a song it’s not at all uncommon to have 25 or 50 tracks of audio (and sometimes more). How could the tiny bit of vibe that a vintage piece of gear might have be worth the three-fold price increase over a modern reissue of the same model? Especially when it would be just one tiny part of the ensemble?
He smirked. I could tell from his face this was an argument he’d heard before and that he happened to have a well-thought-out rebuttal. He was quick to point out that I wasn’t considering the effect that elevating every track to that level, or alternatively, allowing every track to be of less-than-optimal quality would have on the final product. I was completely ignoring the cumulative effect of striving for the best possible solution at every step.
As software craftspeople, we’re equally affected by this same cumulative effect. Generally speaking, I believe the code we write is a prime example that we’ve recognized that idea. We relentlessly contemplate the best name for classes, variables, etc. because we know that effort will reward us with a system that unquestionably reveals its intent. We scrutinize every design decision we make because we know the maintainability of the system ultimately relies on it. We strive to produce software that cumulates into clean, easily modifiable, performant systems; and we know the path that gets us there is paved with countless small, yet important decisions. I do, however, question whether we, as professional developers, give that same attention to the way in which we develop—all of the seemingly insignificant actions that ultimately result in the software we create.
It seems logical that we wouldn’t. The companies we work for and the consumers of our products don’t care how we develop as long as the things we deliver are on time, on budget, and meet or exceed any quality expectations. However, as producers, we should always be focused on whether the act of producing is as efficient as it could be.
The Seemingly Insignificant
It wasn’t until I started working with other developers that I really started thinking about inefficiencies in development workflows. I remember the first moment it popped into my head… I was pair programming with a colleague and the code we were working on lived near the bottom of the file. I don’t remember exactly why I needed to jump to the top of the file (I probably needed to require something), but I proceeded to type: shift, colon, one, enter. I’m sure my pair grit his teeth a little, but didn't say anything. I did my business at the top of the file, and to get back to the bottom I typed shift, colon, nine, nine, nine, nine, enter. That was obviously too much for him to bear, so he promptly chuckled and informed me that I could get to the bottom of a file simply by typing G.
It’s not that I didn’t know :9999<cr> was a ridiculous way to get to the bottom of a file, I just didn’t care. It was only seven keystrokes and it worked 99.99% of the time. Later that day, I thought about how many times I would actually need to dive to the bottom of a file in a day, a week, or a year. When taking into account the effort required to perform those seemingly insignificant keystrokes over the course of a thirty year career, it seems crazy that I would choose that over sacrificing 60 seconds to Google a better way. Yet, similar (and worse) inefficiencies exist in every developer’s repertoire of bad habits. We complain about the JVM startup time yet spend twice that holding j to traverse 100 lines in Vim. We repeatedly manually rename series of variables instead of spending four hours one Friday afternoon really harnessing the power of regular expressions.
The idea of streamlining your workflow doesn’t end with leveling up your Vim skills. As software professionals we are committed to test-driven development. We reap many benefits from this discipline, one of which is a fast feedback loop. We create fast unit tests that give near-instantaneous feedback when functionality has been fully implemented or a regression has been introduced and we run them repeatedly, often many times per minute. Surprisingly, however, testing workflows still exist that are not conducive to this instantaneous feedback loop.
It’s incredibly easy to switch panes in tmux, hit the up arrow and enter (assuming you’ve previously run the tests), and watch your entire test suite run. It certainly gets the job done, but it eliminates one of the major benefits (in my opinion) of a fast feedback cycle: minimizing mental context switches. The second you have to think about switching windows in tmux or watch a parade of passing tests fly across your screen you’ve made a mental context switch; the problem at hand is no longer front and center in your mind. Worst case scenario, you may even have time to fire up Twitter or check your email.
I firmly believe in taking breaks from your work. It can help you form new ideas and approach existing problems with new perspectives. In my experience, though, these kinds of micro-breaks are not helpful when taken mid-problem solving. They merely serve as distractions causing you to lose your hard-earned mental state.
Solving Your Problems
In addition to the effect this has on your mental state and how you approach problems, it suffers from the same problems as before: wasted time and keystrokes. I have a few custom tmux bindings, but even with those in place it’s eight keystrokes and a hand shift off home row to leave Vim, run my tests, and get back into Vim. Think of how many times in your career you should be running tests—the answer is a lot. Now think of the productivity you could gain by reducing this one thing you do multiple times per day by 75%. How much more often would you run those tests? How many more ideas would you experiment with? How much faster would you finish a feature if running your tests was an instantaneous, mindless reflex instead of a conscious mental activity?
This is, in fact, something I did decide to tackle. A large majority of the work I do is in Ruby, so it probably comes as no surprise that I run RSpec tests countless times per day. I wanted to be able to execute the exact specs I needed with the minimum amount of effort possible, and to accomplish this, I decided to write my own RSpec runner.
The most important thing I wanted out of a runner was to be able to run the specs for the file I currently had open in Vim. When on a spec file, I wanted it to run that file. When on a production code file, I wanted it to run its associated spec file. In addition to this, I wanted a way to run all the specs for the project and a way to run only the spec my cursor was on. Another requirement of the runner was that it support running the specs in Vim (as you would any other shell command) and running them in a different window from Vim (by using a named pipe1). With this new runner I can have both my tests and my editor in view at the same time, run the whole suite with <leader>a, run the current file’s specs with <leader>r, and run the spec under my cursor with <leader>l. There is next-to-zero ceremony involved in running the specs I want to run. It has become a mindless reflex instead of something I shift to, set up, and execute.
I’m not explaining my RSpec runner because I think it's some sort of game-changing test runner. I bring it up simply as an example of a solution to something I saw as a detriment to my development experience. Something I, for a long time, felt wasn’t an issue, but after taking a step back and reflecting on the way in which I was developing, realized would add up to a significant amount of wasted time and effort. It was incredibly simple to write that script and modify my .vimrc accordingly and that small investment of time and effort is still paying dividends.
I feel it's important to note that I’m incredibly far from having a consistently efficient workflow for all situations. There are developers who most definitely have this worked out far better than I do. The beauty of it is, and the reason a post like this is worth writing and can possibly exist, is because there is no end goal. There is not an “ideal” workflow that everyone should be using, partly because personal preferences vary and partly because there’s always room for improvement. We do, however, owe it to ourselves to be constantly critical of the way we do things and to be on a never-ending search for something better; not just a way to create better software, but also a better way to create better software.
Classical guitarists invest thousands of hours stressing over millimeter differences in finger placement or what hand position should be used to contribute to the overall fluidity of the piece being played. As software craftsmen and craftswomen on a path to mastery, we owe a similar attention to detail to our own techniques. Changes to them may feel "expensive" or not worth the investment, but the cumulative effect they have on the development process as a whole can be quite significant.
1I learned about the named pipe technique here.
