Foundational thinking in software design
- Posted in
Once you finished with the foundation, it doesn't matter who you call to architect your house or fix problems you might have. Businesses and software are exactly like that. Think hard about your foundation, it will save you a lot of effort later on. I've been working in a lot of different places and was surprised to see they didn't know there are other ways of doing things. I distill the foundational principles one needs for a good software solution and maybe not just software:
- Separation of concerns - processes, components and people should be able to function in isolation. If you can test they work when everything else is shut down, you're good. People should only do what they are good at. Components should do only one thing.
- Cleanliness - keep your code readable rather than efficient, your process flow intuitive, roles and responsibilities clear. Favor convention over documentation, document anything else.
- Knowledge sharing - Allow knowledge to be free and alive in your organization by promoting knowledge sharing, collaborative documentation, searchability.
I am not the greatest of all developers or architects, but I am good. I know how things should be and what they should do in order to reach a goal. When people ask me about software, my greatest gaps are around specific software tools or some algorithm, not the general craft. That is because of several reasons: I enjoy my work, I've been really enthusiastic in my youth and sponged everything I could in order to become better and I've worked in many different types of organizations so I know multiple ways in which people have tried to do this. As I grow older, the last one may be my most valuable skill, but I am yet to find the employer to realize this.
You see, what I've learned from my career so far is that most businesses live in a bubble. Used to not only learn software development as I am working on some task, but also network with other people in the craft from all across the business, I kind of expected every other developer to be like that. Or at least the senior devs, the dev leads and architects, maybe some managers. But no, most of the time, people are stuck in their little position and never stray from it. They may invoke life work balance, or they are just burned out, or they just don't care enough, but more often, they haven't even realized what they are missing. And that's the bubble. A bubble is not a prison, is just a small area in which people stay voluntarily and never get out of.
This is why gaming development is so different from business app development. That is why development in an administrative business with a small software department is so different from development in a software company. It is why development in small shops is so different than in large software companies. Sometimes people, really smart people, have worked for a long time in only one of these ecosystems and they only know that one. They can hardly conceive of different ways to do things.
So this is why I am writing this post, to talk about the foundations of things, that part that separates one from the other, forever, no matter what you do afterwards. And this applies to business, people organization, but especially well to large software projects. You see, if you start your solution badly, it will be bad until you rewrite it. Just like a building with a weak foundation. It doesn't matter you bring the best workers and architects afterwards, they will only build a wonderful house that will fall down when the foundation fails. You want to make a good thing, first plan it through and build the greatest foundation you can think of and afford. It's much easier to change the roof than the foundation.
And you wouldn't believe how many times I've been put in the situation of having to fix the unfixable. "Hey, you're smart, right? We started this thing a million years ago, we thought we would save some money, so we got a bunch of junior developers to do it, and it worked! But then it didn't anymore. Fix it!" And I have to explain it to them: you can't scale duct tape. You can go only so much with a thing held together by paper clips, chewing gum and the occasional hero employee with white hair and hunched back and in his twenties.
Now of course, to an entitled senior engineer like me any software evokes the instinct to rewrite it in their own image. "Also, get some juniors to carve my face into that hill over there!". Sometimes it's just a matter of adapting to the environment, work with what you have. But sometimes you just have to admit things are beyond salvation. Going forward is just a recipe for disaster later on. It's the law of diminishing returns when the original returns were small to begin with. And you wouldn't believe how many people agree with that sentiment, then declare there is nothing that can be done. "They won't give us the budget!" is often muttered. Sometimes it's "We only need this for a few years. After that we start from scratch" and in a few years some "business person" makes a completely uninformed cost and gain analysis and decides building up from existing code is cheaper than starting over. But don't worry, they will suuuurely start from scratch next time.
Sometimes the task of rewriting something is completely daunting. It's not just the size of it, or the feeling that you've achieved nothing if you have to start over to do the same thing. It's the dread that if you make the same thing and it takes less effort and less money and it works better then you must be inferior. And it's true! You sucked! Own it and do better next time. It's not the same thing, it's version 2.0. You now have something that you couldn't have dreamed of when building version 1.0: an overview. You know what you need, not what you think you will need. Your existing project is basically the D&D campaign you've played so many times that it has become a vast landscape, rich with characters and story. You've mapped it all down.
This post is an overview. Use it! To be honest, reaching this point is inevitable, there will always be a moment when a version 2.0 makes more sense than continuing with what you have. But you can change how terrible your software is when you get to it. And for this you need the right foundation. And I can teach you to do that. It's not even hard.
Separation of Concerns
Most important thing: separation of concerns. Components should not be aware of each other. Compare a Lego construction to a brick and mortar one. One you can disassemble and reassemble, adding to it whatever you need, the other you need to tear down and rebuild from zero. Your foundation needs to allow and even enable this. Define clear boundaries that completely separate the flow into independent parts. For example a job description is an interface. It tells the business that if the person occupying a job leaves, another can come and take their place. The place is clearly defined as a boundary that separates a human being from their role in the organization.
Software components, too, need to be abstracted as interfaces in order to be able to swap them around. And I don't mean the exact concept of interface from some programming languages. I mean that as loosely as one can. A web service is an interface, since it abstracts business logic from user interface. A view model is an interface, as it abstracts the user interface logic from its appearance. A website is an interface, as it performs a different task than another that is completely separated. If you can rewrite an authorization component in isolation and then just replace the one you have and the application continues to work as before, that means you have done well.
Separation of concerns should also apply to your work process and the people in it. A software developer should not have to do much outside developing software. A manager should just manage. People should only be in meetings that bring value and should only be in those that actually concern them. If the process becomes too cumbersome, split it up into smaller pieces, hire people to handle each of them. Free the time of your employees to do the job they are best suited for.
One important value you gain from isolating components is testing. In fact, you can use testing as a force towards separation of concerns. If you can test a part of your application in isolation (so all other parts do not need to be working for it), then you have successfully separated concerns. Consider a fictional flow: you get on the bus, you get to the market, you find a vegetable stand, you buy a kilo of tomatoes, you get back to the bus, you come home. Now, if you can successfully test your ability to get on a bus, any bus, to get anywhere the bus is going, in order to test that you can buy tomatoes from the market you just test you can find the stand and buy the tomatoes. Then, if you can test that you can buy anything at any type of stand, you only need to test your ability to find a stand in a market.
It seems obvious, right? It feels that way to me. Even now, writing this post, I am thinking I sound like an idiot trying to seem smart, but I've seen droves of developers who don't even consider this. Businesses who are not even aware of this as a possibility. "We have testing teams to make sure the application is working end to end, we don't need unit testing" or "We have end to end automated testing. For each new feature we write new tests". When you hear this, fight it. Their tests, even if written correctly and maintained perfectly, will take as long as one needs to get on a bus and go to the market. And then the other test will take as long as one need to get on a train and get to the airport. And so on. End to end testing should exist and if you can automate it, great, but it should be sparse, it should function like an occasional audit, not something that supports your confidence in the solution working correctly.
So go for testable, not for tests. Tests often get a bad wrap because someone like me comes and teaches a company to write tests, then they leave and the people in the company either skip testing occasionally or they change bits of the application and don't bother to update the tests. This leads to my next point: clean code.
Cleanliness applies to everything, again. The flow of your solution (note that I am being as general as possible) needs to be as clear as possible, at every level. In software this usually translates in readable code and builds up from that. Someone looking at the code should be able to instantly and easily understand what it does. Some junior developers want to write their code as efficient as possible. They just read somewhere that this method is faster than the other method and want to put that in code. But it boils down to a cost analysis: if they shave one second off a process you run ten times a day, they save one hour per year; if another developer has to spend more than one hour to understand what the code does, the gain means nothing.
Code should be readable before being fast. Comments in code should document decisions, not explain what is going on. Comments should display information from another level than the code's. Component names, object names, method names, variable names should be self explanatory. Configuration structures, property names, property values, they should be intuitive and discoverable.
And there is another aspect to cleanliness. Most development environments have some automated checks for your code. You can add more and even make your own. This results in lists of errors, warnings and notifications. On a flow level, this might translate to people complaining about various things, some in key positions, some not. Unit tests, once you have them, might be passing or failing. It is important to clean that shit up! Do not ignore warnings or even notifications. You think a warning is wrong, find a way to make it go away, not by ignoring it, but by replacing the complaining component, marking it specifically in the code as not a valid warning and document why, run all the tests and make sure they are green or remove the tests that you think are not important (this should not happen usually). The reason is simple: in a sea of ignored warnings you will not see the one that matters.
To be perfectly clear: by clean code I don't mean code that follows design patterns, I don't mean documentation comments on every property and method, I don't mean color coded sections (although that's nice). What I mean is code clean enough to read without cringing or having to look in ten other places to figure out what it does. If your hotdog falls on that code you should be comfortable enough to pick it up and continue eating it.
Cleanliness should and must be applied to your work process. If the daily meeting is dirty (many people talking about unrelated things) then everybody is wasting time. If the process of finishing a task is not clear, you will have headless chickens instead of professionals trying to complete it. If you have to ask around where to log your hours or who is responsible for a specific job that you need done in order to continue, you need to clean that process. Remove all superfluous things, optimize remaining ones. Remember separation of concerns.
Cleanliness extends to your project folder structure, your solution structure, your organizational structure. It all has to be intuitive. If you declare a principle, it should inform every query and decision, with no exception. "All software development people are at the fifth floor! Ugh... all except Joe". What if you need Joe? What if you don't know that you need Joe, but you still need him? Favor convention over configuration/documentation, document everything else. And that leads me to the final point: knowledge sharing.
To me, knowledge sharing was always natural. In small companies there was always "that guy" who would know everything and couldn't work at all because people came to ask him things. In medium companies there was always some sort of documentation of decisions and project details. In large companies there were platforms like Confluence where people would share structured information, like the complete description of tasks: what they are about, how decisions were made, who is responsible for what, how they were split into specific technical tasks, what problems arose, what the solutions were, etc. And there were always your peers that you could connect to and casually talk about your day.
Imagine my surprise to find myself working in places where you don't know what anyone else is doing, where you don't know what something is and what it is supposed to do, there are no guidelines outside random and out of date Powerpoint files, where I am alone with no team, brought in for problems that need strong decisions in order to fix but no one is willing to make them, and already I have no idea who should even attempt to. I solve a common problem, I want to share the solution, there is no place to do that. People are not even in the same building as me. Emails are come and go and no one has time to read them.
Knowledge should live freely in your company. You should be able to search for anything and find it, be able to understand it, contribute to it, add more stuff. It should be more natural for the people in your company to write a blog post than go for coffee and complain. It should be easier to find and consume information from people that left the company than to get it from colleagues at the desk next to you. And maybe this cannot be generalized to all departments, but it is fucking important: people in the office should never need to open Microsoft Office (or any similar product suite). I can't stress that enough.
You should not need printed documents, so no need for Word. Excel files are great for simple data tasks, but they are not specific. If you need something done repeatedly and you use Excel sheet, it is probably better to build a tool for it. Don't reinvent the wheel now, but use the best tool for the job. And there are better and more modern tools than Powerpoint files, but I will allow the use of them because, in the context of knowledge sharing, everyone should feel free and confident enough to make presentation for the team. My tenet still stands, though: the Powerpoint file would be used in a presentation. Hardly anyone else should need to open it. I mean, there would be a video of the presentation available, right?
Imagine a park. It is sunny, birds are singing, there are people walking on hardened dirt walkways, cyclers biking on their asphalted bike lanes, benches everywhere, with a small notepad attached to them that people can just pick up and read or write their own notes. Places in the park are clearly indicated with helpful arrows: children playground, hotdog stand, toilet, football field, bar, ice ring. Everything is clean, everybody is doing what they do best, all is good. You feel hungry, you see the arrow pointing towards the hotdog stand, you walk there calmly and ask for one. The boy there give you a bun and a wurst. He is new, but he has a colleague that knows exactly how much mustard and ketchup to put on the hotdog. He even asks you if you want curry on it.
Imagine a park. It is sunny, birds are singing. Some walkways start of as asphalt, then continue as dirt. Some stop suddenly or end in a ditch. There is a place that serves hotdogs next to a toilet. You have to ask around to find out where to find it. You get lost several times, as some people don't know either, but they still come with an opinion, or they are just misinformed. You get tired, but you can't sit on a bench, they are all taken and there are so few of them. You have to look both ways several times before you walk to the stand, because of cyclers. You stand in a line, then order a hotdog. The boy there gives you a bun with a wurst in it. You ask for mustard, but the boy is new and it takes him a while to find it after looking for some paper that tells him where it is. You have to dodge a football that was coming at your head. Someone flushes the toilet.
Be the first to post a comment