You don’t want many options

A nice little library came up the other day: cprop. Simply put, it’s a swiss army knife of configuration management: it can load configs from classpath, files, DB or environment, it merges configs, provides defaults, does type coercion etc.

Should you use it? It’s easy to say yes, but I say no.

“Yes” path

Choosing something like cprop is usually a no-brainer. After all, you need something to manage configs, and library like that is obviously future-proof. If you don’t need everything at the start, you might decide to use it just for simple things like port numbers and only load them from the environment and nothing more. Using existing library is still better than rolling your own solution.

But projects are long endeavors. Somewhere down the road, you might have a situation: a demand will come up that is most easily addressed by using some other part of cprop. What would you do?

Let’s see: cprop is already in the project, it has the perfect solution, its authors already made all the design decisions and implemented all the code which is just sitting there, waiting, ready to be used. Using more of cprop at that point will be the absolutely rational, most effective decision you can make.

The downside though is that eventually you’ll end up with a really complex configuration system. It’s nobody’s fault, it’s just how project dynamics work.

“No” path

Alternatively, if you choose a much more focused library or decided to roll your own minimal solution, when that demand comes up there’ll be nothing to address it.

This moment is important: you’ll realize that making config more complicated is not free anymore. Because of that, you might decide not to extend your configuration system but build a workaround or introduce a convention. Your config might, as a result, stay simple.

This is backpressure: your home-grown configuration system resists being extended, and it’s a good thing. It keeps you from turning everything into a complicated mess or at least postpones that moment. Value it. Look for ways to restrain complexity creep and don’t get excited about the plethora of options.