Slowness is a Side Effect
Functional programmers use the term ‘side effect’ to refer to the impact a function has on the world around it. Side effects are implicit—they are not represented in the input arguments or output types of the function. It’s considered good functional style to write programs without side effects, or at least with a minimal and predictable number. In imperative programming, we still try to keep side effects restricted to recognizable places: Ruby uses a
However, it’s easy to fall into the trap of believing that the footprint of a function begins and ends with its prototype. The question of whether a function has side effects a simple one: is the world you are in when the function terminates equal to the one you were in when it started? At the level of fast, atomic operations, like
1 + 1, this is obviously true. But as a function gets slower, no matter whether it acts on its environment or not, the world you are in when it terminates will have changed significantly since it started.
What does this mean to those of us who don’t care so much about functional purity? Well, consider the case of defining a capital-I Interface, and the concomitant claim that you can swap out the implementation without changing any properties of the system. This is often used as an argument for incurring technical debt, with the promise that the details of an API, database or service can be ‘abstracted away’.
The fallacy—or false hope—in this argument is exposed when one considers that there’s more to an interface than its component function signatures. The performance and implementation details of various methods have a direct impact on the experience of their use, and this will, in turn, shape the software written against them. When you design an interface, it designs back.
My sole recommendation—and it’s not a panacea—is to take a holistic approach to implementation and interface. Recognize that performance characteristics do have an effect on how you code. Write opinionated software, because portable, generic software is usually slow, buggy software. Finally, take pride in designing robust interfaces and schemas that stand the test of time.