This began life as a post here but it became long enough and IMHO important enough, I am going to clean up the spelling a bit and repost it here. It was in response to a fairly new developer with a question about class design.
This is one of those things that is nearly impossible to explain in a post, or hell, short of a small book.
But generally ( and yes, there are exceptions, thousands and thousands of exceptions ) you want to decouple your objects from each other as greatly as possible. You also want to keep code as self contained as possible, again, with thousands of exceptions.
A lot of the time when you are designing your classes ask yourself "If I changed this class, how many other aspects would need to be changed as well?" Then go about breaking those dependencies where possible. If two classes don’t really need to know about each other, break that dependency. If two classes share 90% of their functionality, derive from a common base class.
On the other hand, you have to let common sense be a factor. One really common reason to abstract away your classes into engines and various components is so you can make switches later. For example, if you are using SDL, but want to be able to switch over to SFML at a later date, it’s common to make a Renderer class, and an intermediary class over all the various objects like Bitmaps and Sprites, that both SDL and SFML provide so you can transparently change Renderers without making a code change. Reality is though, for 90+% of projects, this is just an added layer of complication and work that will never ever be used. This is where common sense has to come into play. Yes, from a purely object-oriented perspective, it is a very "pure" design that implementation details are abstracted away. From a practical and getting shit done sense, it’s just a stupid waste of time.
This is where we get to the next sticking point and another area where new developers ( and experienced developers! ) get it really wrong most of the time. Global variables. See, when a new developer learns about global variables they seem like a godsend, as then you frankly don’t have to design classes at all, just make everything globally available! This in a very short while makes extremely unreadable and maintainable code. Other new ( and experienced developers ) have heard about this and had it drilled into their heads that GLOBALS ARE BAD!!!! and react accordingly. Ironically enough, 9 times out of 10 this seems to always result in the same discovery… the Singleton. If you haven’t yet, some day you will discover the singleton pattern and you will think it is the answer to your prayers. It’s not a global, it’s an object!!! Actually no, it really is just a global with a pretty layer of object oriented goop wrapped around it(**). Then you will encounter another group of people that will yell from the mountains SINGLETONS ARE BAD!!!! Which in the end leads you to engineering half a hundred "Manager" or "Adapter" classes to eliminate your dependencies on global variable and singletons. All the while, no work is getting done on your game.
Now it’s time to let you in on a dirty little secret. Globals aren’t bad, they are actually pretty frigging useful and important. The most important thing is, DO NOT ABUSE THEM. That is it. Chances are, you are going to have some form of global object, like perhaps a Game, App or World class, this is completely reasonable. On top of that you are going to run into code that needs to be globally accessed, such as your Audio class or Event class, which is also quite reasonable. It simply does not make sense to pass in an Audio class to say, a GameEvent, because that game event might on occasion need to play a sound. There are quite literally only a handful of things that should be made globally available but if it makes sense to be globally available, make it globally available.
But here is where the design gets tricky again, if you are making something globally available, do not make the relationship two-way, or at least not unless the two-way relationship is completely generic or via an interface. For example, if you have a game engine ( globally available ), and that game engine has a (globally available ) event queue, that at some point a game object needs to add events to, make sure that the event queue has the most absolutely minimal requirement to know anything about the game object and that the game engine itself needs to know absolutely NOTHING about the game object, as this is where your design can start to fall on it’s face.
One last general rule of thumb when dealing with globally available classes, make them static whenever possible, it will lead to less potential hazards. Additionally, when make global objects provide access to other global object ( for example Game::SoundManager ), use accessor methods and const whenever possible. Keep all allocation and de-allocation specifically in the realm of the containing object.
(**) There are actually more to singletons than just object oriented globals. Yes, Singletons are global in scope but they have a couple very important properties. First of which is deferred allocation, meaning the Singleton isn’t using memory until the first time it is used, unlike a traditional static variable. Additionally, a Singleton can be sub-classed as a mechanism of providing a single interface to potential disparate code bases, so for example you could have a Renderer::GetInstance(), which is sub-classed as OpenGLRenderer or DirectXRenderer, which will provide your code with a single global runtime determined instance. If you don’t understand this paragraph, that’s completely fine, I’m just throwing it out there as a caveat so my over simplification of a Singleton isn’t misleading… they do have their uses.