There’s something I need to get out of my system, and it’s how we tend to handle design when writing code. Much has been said on this and I’m not sure I will actually contribute to the discussion, but I feel better when I have it written down. This is going to be a lengthy and probably quite meandering post, but it’s been quite a while since I last wrote about the subject so I think it’s my turn again to further pollute the internet.
The trigger to write this was an internal “wiki” page (I should write a post about the evils of Confluence some time ;-)) where a design proposal for a new service is written down. People get invited to review it, or get referred to it, they leave comments all over the place (in-line, at the bottom), and I think that the poor person that wrote it must feel quite bad about all the flaws being pointed out, edge cases missed, etcetera. I call it bikeshedding, especially because of all the drive-by advice by people who are certainly not going to actually implement the service; some of them aren’t even full-time software developers. I’d feel funny if contractors work at my house and the carpenter starts telling the plumber how to do his job…
So, “design”. Documents like the one on our internal wiki are full with diagrams and documentation and called ‘designs’ and I think that that is wrong. They are, at best, sketches of designs. Analogies when talking about software are always wrong and usually dangerous, so let’s give the stage to some to help me bolster this agrument.
Writing software has been likened to an art, a craft, and - alas - in colleges and universities around the world, to science and engineering. In the article I wrote in 2000, I was explicitly neutral and just made an inventory; however, 17 years later I have a pretty firm position that at least the stuff that gets built in the sort of companies that have hired me since then is, or should be, created with the model of craftsmanship in mind: bespoke, one-off, with a nod to other disciplines where needed. This might be different for writing code for a nuclear powerplant (or might have to be different, at least), but this is the situation for all the “internet” companies I worked for during the last couple of decades. You can call yourself a “software engineer” as much as you want, but you are bullshitting, I hope you know it, and in most jurisdictions you’re also not in line with the law. We are craftspeople and I think we should be proud of it. We’re changing the world after all, and I think we should embrace our craft (partially also because it will set the stage for in-house training after secondary education instead of the silly, expensive, and entirely unnecessary route through college and, heavens forbid, university).
So, we’re like craftsmen, we build bespoke stuff and where I say “build” you rightfully ask “where’s the design?”. Let’s pull two more analogies in. Yes, I like to dig my holes really deep.
First, I’ve been following Dick Gabriel since I first met him, and I think (you’re never sure with him) that he thinks that source code is like poetry. I like that angle, and I’ll work out the analogy myself so you don’t blame him for me getting it all wrong: poetry is all about transmitting thoughts and ideas in a very restricted form, and ideally using that form to lift the content into art. Poetry also started, I think, as a way to make things easy to remember, so that bards could recite very long stories without having access to a written form of language. Source code has a lot of that: it is a restricted form of language, and it captures ideas and thoughts of the programmer. Good source code, I think, tells a story and puts the human reader first; having a machine execute it is almost like a side effect. This, to me, is also the correct order, because people will read your code over and over again (to change it), and they must know what the original programmer was thinking, why certain solutions were chosen, et cetera. Good source code is a poem. Where poetry approaches art, I think good source code does, too. Note that this is an entirely normal effect of good craftsmanship - look at some Shaker furniture, for example.
But it is also the design. I think that this makes software so unique - it plays multiple roles, and that makes analogies either wrong, or complex, or both. Software is simply an executable design, fed to a machine that interprets the design and either starts acting on it directly, or transforms it to something that another machine or the same machine can easily execute. Whether you use an interpreter or a compiler is not important in the discussion, the role of your code stays the same: it’s the design. It’s a wonderful craft, software, because as soon as you have finished your design, you’re done - everything to transform your design into something that actually interacts with the world is automated, and very fast. Even if you use, say, C++, the build time is negligible to the amount of time spent fine-tuning the design and nowhere near the time it takes to build physical stuff like buildings or planes or cars. Also, if your product tells you there are design flaws, you don’t need to go through expensive recalls after design changes - you change the design, re-create the product, and are done with correcting the issue in minutes or hours. Truly a wonderful way to work.
So, my analogies line up as: developing software is a craft, skilled labour to create executable designs that should have the good qualities of poetry. Probably not what your university prof told you in 101 Computer Science, but in my experience this is a useful analogy and it holds up quite well under scrutiny. I urge you to think about it and see where it leads you. At the very least it should be an interesting thought experiment.
So what is the role of this wiki page that someone wrote and called “design for my new servce”? Well, it is not the design. And to have a design of the design sounds to me a tad circular, so I think it is safe to conclude that the label is wrong. The reason, of course, that people label discussions about software as “design” is because they have been imprinted with the misconception that software is the product and then something not-software is needed to play the role of the design.
I do think it is important to think about what you need, and what some constraints and requirements are before starting out to design it, but our craft has designs that are so malleable and feedback from producing executable machine code from these designs is so quick that it is not only possible to start out early, I think it’s mandatory. If you don’t start writing code very soon, you are missing out on all the good bits that it has to offer. Do make some charcoal sketches of what you want, but let that not mislead you into thinking that they are a useful model - at best, they’re thinking aids. At worst, they’re the equivalent of handing someone with writers block a sketchpad and a “Doodling for Dummies” book: a great way to postpone the actual work.
Instead, use the power of increments and iteration to its fullest and let the design tell you what it needs. Don’t waste your time on making nice diagrams if you can write code and let the code tell you what you missed. That feedback is real; the bikeshedding on that wiki page? Theoretical.
Here are some tools to maximize the power of feedback:
Write a test. What does your software need to do? Write it down, run it, make it green. If you write new code, start with a test and nothing else and your final design will be pretty exact. I always regret it when I think I know the problem well enough to start writing code - the best code gets written by tests.
Now, don’t get me wrong - I do not buy that you can start with a blank mind and just write a test and let that drive all of your design. I think you want to make some quick sketches and you should have an idea of where to go. Just make sure you test your ideas, all the time. It will quickly separate the good ideas from the silly ones. I’ve been writing code for 40 years, and I still manage to come up with really silly ideas (I call it “creativity”. Feel free to label it as “stupidity”, though). TDD and refactoring to remove the stupid stuf is your saving grace.
We have the power to postpone decisions and still get useful feedback from early production runs, especially when you start by writing tests. You do not need a database, or a message queue, or deployment scripts, or all that boring stuff around operability to flesh out a design. You can have test dummies, you can replace a database by an in memory structure. Just focus on the business logic, make that nice. Ignore your ideas about Kafka and S3 and Bigquery and Kubernetes and whatnot, just solve the customer’s problem.
Then try running that in a production-like environment and see what you are missing (incoming data is already on that message queue? Ok, we must have a consumer now) and what you’d like (not want to re-read a year worth of events on startup? Ok, so we need to stash some state somewhere). Try to cheat as much as possible, only do the minimum. Have integratin tests tell you what that minimum is.
Treat code as your primary artefact
Because it is your primary artefact. It is your design, both to communicate to fellow human beings and to the machines we use to build and execute our product. Don’t talk about diagrams, talk about code. If you have an “operability-free” business model, you probably have some very concise code you can show to others, review together, and discuss. It’s way more concrete than a diagram, it shows the edge cases you thought about in your test suite, so it should sollicit much better feedback than people going all “what if?” on a diagram. And it’s a minimal design, more fit for debate than for execution, so it should prove easy to change if you change your mind.
Be a poet
Nice code is often good code. Even if it has bugs, these are simple to verify with a test and easy to fix. So even while eventually a machine will have to execute your design, think about the other people (including future you) first and write your code for them. When I once showed Chris Alexander how his ideas about design patterns influenced our practice and walked him through some test driven development I was doing, he exclaimed “so good code has a rhythm!”. I think he was right. Good code has a rhythm, the drum being beat by tests; good code reads like a poem, the metric and rhyme dictated by your language.
Enjoy your craft and be a poet and your code will be good.