I'm often faced with the problem of storing, in memory, a few (possibly complex) configuration settings loaded from files on the filesystem. I'm wondering if there's a better way to architect a pattern to this problem, however, than what I've been using.
Essentially, my current solution involves three steps.
Build a singleton. Since data is persistent and guaranteed not to change through the runtime of the application, only one object instance should ever be needed.
When the first request for the object is made, create the object and read in from a file.
Expose data with getters.
This has the effect that a lot of my code looks like this:
MyConfiguration.getInstance().getWeightOfBomb()
, which looks rather odd to me.
Is there a better way to handle this in a more semantic fashion?
Additional proposal for noah's answer.
If it is inconvenient for you to write a method for each configuration parameter, then you could use enum for that. Here is what I mean:
}
After that, if you have a new configuration parameter, all you need to do is to add a new enum entry.
You can also extract an interface from Configuration class, of course, moving enum outside.
You could create an interface to represent the configuration:
And a singleton implementation:
And then inject the configuration in your classes:
Dependency Injection. You don't necessarily have to use a DI framework like Spring or Guice but you really want to avoid littering your code with singletons. You can still use a singleton in the implementation, but there is no reason the rest of your code needs to know that it is a singleton. Singletons are huge pain when unit testing and refactoring. Let your code reference an interface instead. e.g.,
If you use a DI framework, just setup you classes to have your
MyConfig
implementation injected. If you don't, then the laziest approach that still has all the benefits is to do something like:Really it's up to you. The important thing is that you can replace
myConfig
on a per instance basis when you later realize that you need the behavior to vary and/or for unit testing.