CHROMATIC LEAVES


Software as a Process

Written by Eric Rasmussen on April 12, 2013.

There is a fundamental distinction between viewing tasks in terms of their end results (the product), and viewing them in terms of their development (the process). It’s important to be conscious of both, but I’ve found that the product over process mentality is far more common than the reverse, and this way of thinking can have a negative impact on projects.

One of the issues with focusing on the end product is that, inevitably, the needs of your users will change, or the environment around the product will change. If you’ve written the code and documented the code to account only for this ideal product, frozen in time, then as soon as the product loses value, your code and documentation lose value too.

If we accept for a moment that this is truly inevitable, then the psychological benefit to process over product is that our code and supporting docs will be aware of its doomed future, and hypothetically we can better plan for it. As programmers, our natural inclination is usually to throw more code at problems, and it’s definitely worth exploring best practices for writing extensible code. A good starting point is Chris McDonough’s talk on API Design for Library Authors.

But even then, the most well-intentioned and extensible code will eventually succumb to time. One way we can try to account for this is to add that level of awareness to our code documentation. If we assume nothing in our code is future proof no matter how well planned, then one role of the documentation should be helping future readers understand the design process. We want to give them insight into how we arrived at our conclusions.

For fun, let’s pretend you’re working for the venerable Gibberton Industries, and you’re tasked with parsing 3rd party CSV files and loading them into a database. Everyone knows the files you get will never have more than 100-200 rows, which is good because the only CSV library available to you, NiftyCSV, loads entire files into memory in a very inefficient way.

Fast forward five years. You’ve moved on, and now I’ve been brought in to find out why your scripts are failing and no data is making its way to the database. In this new time and context, it’s completely unthinkable that anyone would use NiftyCSV for anything. It’s a barely remembered, unmaintained library, and these days there are plenty of robust options to choose from. I see that the obvious issue is you’re using NiftyCSV to load files into memory when the files are hundreds of thousands of rows a piece, and no one in the business remembers a time when they were any smaller.

When I look to your code for answers, I find this comment:

Use NiftyCSV to parse the files and load them in our database.

Remember that now I’m in the future, and no one writes code the way you wrote it anymore. Reading that comment, I’m wondering if you were mad, incompetent, or, worse, brilliant. Maybe you wrote the solution the way you did for a subtle and mysterious reason that I can’t see, and when I try to address the immediate memory issue, I may inadvertently break something else.

The comment I should see is:

Use NiftyCSV to parse the files and load them in our database. NiftyCSV is the only stable CSV library available, and we only expect CSV files to have 100-200 rows each.

The amount of work required to update the code may be the same in both cases, but in the latter scenario I have the confidence I need to make changes. When your documentation includes this level of awareness and commentary, it lets new developers (or even future versions of you) step into the context in which it was written to better understand how it can be changed.

The required shift in focus from product to process may seem subtle, but it has profound implications for how you think about development and how you document and evolve your codebase over time. It takes very little time to add a sentence or two per module (or other unit of code) to make future maintainers and readers a part of the process.


Tagged: code, soapbox