The Road to Performance Is Littered with Dirty Code Bombs
More often than not, performance tuning a system requires you to alter code. When we need to alter code, every chunk that is overly complex or highly coupled is a dirty code bomb laying in wait to derail the effort. The first casualty of dirty code will be your schedule. If the way forward is smooth it will be easy to predict when you'll finish. Unexpected encounters with dirty code will make it very difficult to make a sane prediction.
Consider the case where you find an execution hot spot. The normal course of action is to reduce the strength of the underlying algorithm. Let's say you respond to your manager's request for an estimate with an answer of 3-4 hours. As you apply the fix you quickly realize that you've broken a dependent part. Since closely related things are often necessarily coupled, this breakage is most likely expected and accounted for. But what happens if fixing that dependency results in other dependent parts breaking? Furthermore, the farther away the dependency is from the origin, the less likely you are to recognize it as such and account for it in your estimate. All of a sudden your 3-4 hour estimate can easily balloon to 3-4 weeks. Often this unexpected inflation in the schedule happens 1 or 2 days at a time. It is not uncommon to see "quick" refactorings eventually taking several months to complete. In these instances, the damage to the credibility and political capital of the responsible team will range from severe to terminal. If only we had a tool to help us identify and measure this risk.
In fact, we have many ways of measuring and controlling the degree and depth of coupling and complexity of our code. Software metrics can be used to count the occurrences of specific features in our code. The values of these counts do correlate with code quality. Two of a number of metrics that measure coupling are fan-in and fan-out. Consider fan-out for classes: It is defined as the number of classes referenced either directly or indirectly from a class of interest. You can think of this as a count of all the classes that must be compiled before your class can be compiled. Fan-in, on the other hand, is a count of all classes that depend upon the class of interest. Knowing fan-out and fan-in we can calculate an instability factor using I = fo / (fi + fo). As I approaches 0, the package becomes more stable. As I approaches 1, the package becomes unstable. Packages that are stable are low risk targets for recoding whereas unstable packages are more likely to be filled with dirty code bombs. The goal in refactoring is to move I closer to 0.
When using metrics one must remember that they are only rules of thumb. Purely on math we can see that increasing fi without changing fo will move I closer to 0. There is, however, a downside to a very large fan-in value in that these class will be more difficult to alter without breaking dependents. Also, without addressing fan-out you're not really reducing your risks so some balance must be applied.
One downside to software metrics is that the huge array of numbers that metrics tools produce can be intimidating to the uninitiated. That said, software metrics can be a powerful tool in our fight for clean code. They can help us to identify and eliminate dirty code bombs before they are a serious risk to a performance tuning exercise.