What's Important in Coding Conventions
I was having an e-mail exchange with a reader who commented that he was having trouble following my coding conventions. Among other issues, I seemed to arbitrarily capitalize variable names. I explained that, in these columns, I specifically varied my coding conventions from article to article, just to avoid looking like I'm endorsing any particular style (one of the reasons that I don't use C# in most of my columns is I'm a "every curly brace on its own line" kind of guy and I don't want to get into an argument about it).
You can probably assume that anything you see that does turn up regularly in these columns is there because I think it's mandated by Visual Studio Magazine.
But there's another reason for the variation you see: Often the code in these columns is drawn from projects that I'm doing for a client and, since I've just copied the code and obfuscated its origin, the code reflects my client's coding conventions. And that made me notice something: As I move from client to client, I keep changing my coding conventons and, you know what? All the coding conventions I use seem good to me. And that, in turn, got me thinking: What does matter in a coding convention?
There's actually some research around this, if you're interested. The earliest work, I think, is in Gerald Weinberg's The Psychology of Computer Programming but the most complete review I know of is in Steve McConnell's Code Complete.
It turns out that the only thing that makes a difference is consistency (or, if you prefer, simplicity). The more special cases and exceptions that a set of coding conventions include, the more likely it is that the conventions won't be followed correctly by programmers writing the code or won't be understood by programmers reading the code.
The most successful coding conventions (the ones that programmers comply with, implement correctly, and find useful when reading code) are short and have no exceptions. As an example, the Microsoft C# Coding Conventions would probably cover three pages if printed out (and less than a page if you omitted the examples). Given the research available, that seems about right.
Posted by Peter Vogel on 12/09/2014