Flexible Code

Posted on February 23, 2017


Last week, another team lead asked me to take on one of his team’s products. It’s not an unreasonable request; his team’s overworked and oddly understaffed, and this particular product is closer to my team’s domain than to his. Given some of the work we’ll be doing in the future, having that product in our domain could be quite convenient.

What’s interesting is how he sold it.

“This shouldn’t impose much work on you guys, if any. The service has been running in production for years, now, and hasn’t changed in a while. It hasn’t crashed, hasn’t broken any integrations, hasn’t gotten anyone paged at 3am. It’s fairly clean, decoupled from most everything else, and hasn’t needed fixes. It should be just fine.”




If you want your software to be flexible, you have to flex it.

  • Uncle Bob, The Clean Coder

We’ve been working on a “from-scratch” rewrite against horrid legacy systems nobody dares to touch. It sounds like a classic software horror story, the second-system effect (this is actually the fifth system by my count), but it’s coming along fairly well (if at a slog). The previous systems were written to specific contracts (and then maladapted to others); this one’s intended from the ground up as a proper product.

One of the newer developers on the team claims it’s the best code he’s ever seen. And honestly, I’m pretty damn proud of what we’ve done in a complex domain in which even the customers often don’t know what they want.

We’re constantly comparing the new system against the olds, looking for discrepancies between the two (some of which matter, some of which are the olds being absurd). We’re looking for a minimal subset of features needed to migrate the next chunk of clients from the olds to the new.

We’re constantly in the codebase, changing shit, often making major refactorings.


I dug up the code for the service I’ll be inheriting. True to claim, it hasn’t been touched in a while.

It’s not that it’s bad code. It’s not a three-thousand-line VB.NET monstrosity with a single thousand-line god method in the middle. (Given the context, that god’s probably Azathoth.) It’s well-factored into your usual layer-cake architecture. The names mean things.

But, well… it’s obvious that it hasn’t benefitted from what the company’s collectively learned in the years since it was architected.

For example, it relies on reflection, and ORM, and other tools that seemed so promising until we found that they were great ways to turn compile-time type errors into runtime exceptions. Not in monstrous ways; as advertised, its uptime record is stellar. And I expect it will be, as long as nothing changes. When it does, we’ll hope the NHibernate mappings are adequately tested.

None of this is to disparage the system’s authors; they’re people I respect and admire. I respect and admire them because, among other things, they never stop learning how to write better code.


If you’re constantly in a service’s codebase, you’re constantly finding opportunities to make it a little bit better. “Leave the campsite cleaner than you found it.” (Praise that in code reviews, by the way, don’t shit on people who make “unnecessary” changes that improve the whole.)

You’re constantly in a position to look at a gnarly piece of code and shout “aha!” as you get an idea of how to improve it. Maybe that idea doesn’t work out, but it plants a seed in your mind. Maybe you jot down some notes on your idea in the company wiki. Maybe your colleague reads those notes.

You’re constantly bothered by the parts of the code that are full of pain and suck and fail. You’re constantly spending emotional effort trying to be patient because they’re not part of the ticket you’re working on. One day you run out of patience and fix the fucking thing. (Make a painful task frequent enough and it will be made less painful, with a baseball bat if nothing else. “HP LOAD LETTER”? What the fuck.)

Slowly, slowly, the codebase is incorporating everything that you’re learning about the domain, everything that you’re learning about where the business wants the domain to go in the future, and everything that you’re learning about how to write great code.


We have services, in production, that we don’t even know how to build any more. At least, not at a company level. The guy who knows is on vacation.

But they’re stable! They almost never cause trouble!



I came across a thought experiment (which I can’t find in a quick google) in which every feature branch you opened deleted itself after twenty-four hours.

“Oh fuck, that’s horrible! So much time spent in re-work!”

Really? If I spent a day working on a feature branch, it would probably take me an hour or two to reiterate the previous day’s work. Okay, “time wasted”, but I’d probably do a better job along the way because I’d be seeing the problem for a second time. That gives me four or five hours to do new work; more, if my company’s not obnoxious about meetings.

Not really that much of a difference, is it?

People often see refactoring as re-work, but that ignores the benefit of reviewing - re- viewing - the code with fresh eyes and more knowledge. Of having the opportunity to incorporate that new knowledge into the codebase, be it a better understanding of the domain or just a better understanding of that one library we’re using.


I’d like to say we’ve all heard of the Last Responsible Moment, but experience dictates otherwise. Handwaving enormously, it suggests that the longer we can wait before making a decision, and gather knowledge relevant to that decision, the better a decision we’ll make.

But we’re always gathering knowledge. Often our customers are gathering knowledge that makes them realize how different their requirements are than what they put in the RFP. How do you know if your system is flexible enough to accommodate their new understanding of what they need?

Well, if you’ve been flexing it hard, up to your elbows in its guts, on a regular basis, you probably have some idea.

Your product manager will be glad to hear that.