False Peaks and Temporary Code

In the day-to-day work of building new software and maintaining old software, we can easily lose sight of the bigger picture. I think we can find perspective when we step back and walk through the evolution of a single piece of software.

For example: first, you are asked for a simple slideshow to showcase a few images handed to you. Just five images and the images won't change.

An easy request! It only takes you a short time to build with some simple jQuery. You show the client, they approve it. You deploy it to production and call it a day.

This example, and all the examples in this blog post, are interactive. Try it out!

The next week, your client comes back with a new request. They don't think the users notice the slideshow can be navigated. They ask for previews of the next and last image, to use for navigation:


So you jump in. It’s an easy enough addition to the pretty simple slideshow widget you've already built. You slap in two images, position them just so, and add a few more lines of jQuery to bind navigation to them. Once again, it’s a quick review with the client and you ship it off to production.

But the next day, there's a bug report from a user. The slideshow works, the thumbnails show the right image, and the new previous/next preview images navigate correctly. However, the features don't work together, because the thumbnail navigation doesn’t change the new left and right preview images you added.

The client also wants the new navigation to act like a carousel and animate.

Now they want to add more photos.

And they want to be able to add, reorder, and change the photos on their own. That will break all the assumptions you made based on a fixed number of photos.

Every step along the way you added one more layer. One small request at a time, the overall complexity increased as those small requests added up. You took each step with the most efficient option you could find, but somehow you ended up with costly bloat. How does this keep happening?


At each step, you took the next best step. Ultimately, this didn't take you where you needed to go.

We don't subscribe to waterfall development practices at Caktus. Agile is a good choice, but as we work through iterations how do we bridge across those sprints to get a larger picture of our path, and how do we make those larger decisions about technical debt and decisions that impact on a larger scale than a single sprint?

Some of the code we write is there to get us somewhere else. Maybe you need to build a prototype to understand the original problem before you can tackle it with a final solution. Sometimes you need to stub out one area of code because your current task is another focus, and you'll come back to finish it or replace it in a future ticket. Many disciplines have these kinds of temporary artifacts, from the scaffolding built by construction crews to sketches artists make before paintings.

Maybe it is harder for software developers because we often don't know what code is temporary ahead of time. A construction crew would never say, "Now that we've built the roof, we really don't need those walls anymore!" but this is what it can often feel like when we refactor or tear down pieces of a project we worked hard on, even if it really is for the best.

My suggestion: become comfortable with tearing down code as part of the iterative process! Extreme Programming calls this rule "Relentlessly Refactor".

We need to think about some of the features we implement as prototypes, even if they've been shipped to end users. We won't always know when new code or features are stop-gaps or prototypes when we're building them, but we may realize later they had been so all along when more information comes to light about where those features need to go next.

Falling into the trap of thinking the work done in sprints is inherently additive is common, but destructive.

If each sprint is about "adding value", we tend to develop a bias towards the addition of artifacts. We consider modification to happen as part of bug fixes, seeing it as correcting a mistake in earlier requirements or code, or as changes stemming from evolving or misunderstood requirements. We may hold a bias against removing artifacts previously added, either within a given sprint or in a later sprint.

Going back to the construction analogy, when you construct a building you create a lot of things that don't end up in the final construction. You build scaffolding for your workers to stand on while the walls are being built up. You build wooden forms to shape concrete, removing them when the foundations and structures are solid.

You fill the building-in-progress with temporary lighting and wiring for equipment until the project is near completion and the permanent electrical and plumbing are hooked up and usable. A construction crew creates a lot of temporary artifacts in the path to creating a permanent structure, and we can learn from this when building software iteratively. We're going to have cases where work that supports what we're completing this sprint isn't necessary or may even be a hindrance in a future sprint. Not everything is permanent, and removing those temporary artifacts isn't a step backward or correcting a mistake. It is just a form of progress.

New Call-to-action
blog comments powered by Disqus



You're already subscribed