So here's the fun Catch-22 of profile websites and game development logs: you want to show off your best, but a game in prototype is by definition an unfinished mess. But unless you talk about it, the tumbleweeds in your blog roll, and then you have a completely different problem...

Anyway, here's what I've been doing recently: How many of you would be interested to hear about a 3D procedural boss fighting game?

As many of you may know, Blender files can be imported directly by Unity as models (teeechnically it's just telling the blender file to export itself as FBX, then importing that, but it's defacto Blender -> Unity for the end user). Honestly, I find it impossible not to want to use Blender as my level / modelling tool rather than trying to build levels out of prefabs in Unity. But with the default importer, there's a catch; instancing.

So, some time ago I wrote about the mayhem involved in getting a non-axially aligned character controller to function in Unity.

There were tears, there were shenanigans. By the end of it, there was a functional end product... but with more experience, it's time to look back on it with a more critical eye. For one thing, wouldn't it be amazing if you could just use standard Unity Rigidbody and Colliders to solve your character movement?

It is, in fact, possible, and without all the shaking and falling through mesh colliders that - at least for me - plagued my earlier attempts at solving with that method in the past.

Read on!

So I recently got shot in the foot by an unexpected issue; custom render effects that work just fine in the Unity Editor, but mysteriously disappear when you run them in a standalone build - with a special thanks to Mark Backler for helping catch the issue (which should teach me to test a build before sending it off for someone to look at, even if it's only been in development a week...).

Doubly troublesome in my case given one of the effects (2D lighting) had core gameplay implications, but fortunately it proved easy enough to track. When a build is produced, amongst other things it creates a new folder <build_name>, containing various files. Key amongst them; a build log.

Popping open mine highlighted this little error:

Some quick googling later highlighted the culprit; Shader.Find(). Long story short, when Unity builds a project, it strips out content it believes isn't being used. In the case of shaders, it will strip out anything that does not meet the following criteria: 'It is used in a material that is used in a scene'. Guess what falls out those bounds? Shaders referenced only in scripts; for example, for image effects.

So. How to dodge around this little conundrum? Fairly simple really; stick your custom image effects shaders in a Resource folder. Ordinarily of course, assets are either loaded via the Resources.Load method or by being assigned in the Inspector. Unity can scan through the latter case and pick out any asset that's being used and thus needs including in the project, and it handles the former case by just automatically including everything in Resource folders blindly. Shader.Find() is one of those odd methods out that can reference things the above techniques would miss, and hence - whilst it will work in the Editor where the game has access to all the assets, even the non-final ones - fail in a full, leaner, standalone build.

Finally, there's also the 'Always Included Shaders' array in the Project Settings > Graphics rollout. This can be useful if, like me, you're re-using some of the Image Effects shaders but not the components themselves (you have to write your own scripts if you want them to run their calculations on a rendertexture rather than the screen) and don't want to touch or rearrange the Standard Assets themselves, which obviously rules out sticking them in a Resource folder. For your own custom Image Effect shaders though I would recommend the Resource folder option over the Always Included array; with Resource, they automatically get included simply by being there, making things much easier to maintain.

If my weapons components all implemented that interface, then in all AI logic, player input handling and so forth, all that code would require is:

And the weapon component - be it a simple projectile launcher, raytracer or magic cat missile spewer - would fire. In short, the input / AI code doesn't need to know how the specific weapon fires, it just needs to know it is a weapon, and be able to tell it to fire. In this way, the weapon components themselves can handle their firing behaviour any way they like, by all the myriad ways.

And this is what interfaces allow you; it is a simple guarantee that any class that implements a given interface (and, most beautiful of all, a single class can implement more than one interface) will contain the functions the interface specifies. For instance, that IWeapon interface guarantees that any implementing class will have a public Fire() methods with zero arguments, returning void. Which is all anything interacting with a weapon component will need to know.

However, the catch rolls in here; all my weapon components also need to be components (classes deriving from MonoBehaviour that you can slap onto gameobjects from the inspector) and an interface does not guarantee this. It just declares what functions and properties will be publicly available. Thus, GetComponent<T>(), which can only search for and return MonoBehaviour objects, fails when it comes to searching for interfaces. If, as above, I wrote:

It would fail. Compiler error.

So how can we save this? We want a public contract; that a Weapon Component will have the method Fire() but without defining what that method does and hampering what we can do with it. Well, mercifully, there's a way around this:

Enter, the abstract class.

(note how I still have it implementing IWeapon, given that it's true and I might as well, though you could probably remove the IWeapon interface at this point)
Now, if I have my weapon components all derive from AbstractWeapon, we ensure they implement IWeapon and we ensure they are a derived class from Monobehaviour, which GetComponent<T>() can work with:

An abstract class and any declared abstract methods are basically a way of saying "There will be a function/property here, and my derived classes will implement it". Much like an interface, it declares properties and functions without declaring how they work, leaving that up to the derived classes (which, like an interface, will have to implement said code by overriding the abstract functions or there'll be a compile error), but it also guarantees the base class (Monobehaviour) and all the code that comes with that.

There is a downside of course; a class can implement any number of interfaces, but can only have one base class; it's still an inheritance tree and thus there will naturally be a 'split' involved somewhere. Fortunately Unity's component based design lets you side-step this a little bit (whilst there will be splits, there's nothing to say you can't just have two different components).

Obviously you can still have some behaviour and code defined in the above abstract class example (which will probably prove advantageous; ie providing protected helper functions derived classes can call on rather that waste time writing the same snippet of code multiple times) - something you would not be able to do with an interface - but in this case, with all the methods marked abstract, it provides exactly the same functionality as an interface with a pre-defined base class. Which is exactly what we want.