To what extent is good code self-documenting? In one of the most entertaining sessions I saw at the recent PASS summit, Jeremiah Peschka (blog | twitter) got a laugh out of a sleepy post-lunch audience with the following remark:
“Some developers say good code is self-documenting; I say, get off my team“
I silently applauded the sentiment. It’s not that all comments are useful, but that I mistrust the basic premise that “my code is so clearly written, it doesn’t need any comments”. I’ve read many pieces describing the road to self-documenting code, and my problem with most of them is that they feed the myth that comments in code are a sign of weakness. They aren’t; in fact, used correctly I’d say they are essential.
Regardless of how far intelligent naming can get you in describing what the code does, or how well any accompanying unit tests can explain to your fellow developers why it works that way, it’s no excuse not to document fully the public interfaces to your code. Maybe I just mixed with the wrong crowd while learning my favorite language, but when I open a stored procedure I lose the will even to read it unless I see a big Phil Factor– or Jeff Moden-style header summarizing in plain English what the code does, how it fits in to the broader application, and a usage example. This public interface describes the high-level process and should explain the role of the code, clearly, for fellow developers, language non-experts, and even any non-technical stake holders in the project.
When you step into the body of the code, the low-level details, then I agree that the rules are somewhat different; especially when code is subject to frequent refactoring that can quickly render comments redundant or misleading. At their worst, here, inline comments are sticking plaster to cover up the scars caused by poor naming conventions, failure in clarity when mapping a complex domain into code, or just by not entirely understanding the problem (/ this is the clever part).
If you design and refactor your code carefully so that it is as simple as possible, your functions do one thing only, you avoid having two completely different algorithms in the same piece of code, and your functions, classes and variables are intelligently named, then, yes, the need for inline comments should be minimal. And yet, even given this, I’d still argue that many languages (T-SQL certainly being one) just don’t lend themselves to readability when performing even moderately-complex tasks. If the algorithm is complex, I still like to see the occasional helpful comment.
Please, therefore, be as liberal as you see fit in the detail of the comments you apply to this editorial, for like code it is bound to increase its’ clarity and usefulness.