A vector wrapper is mainly useful if used as an adapter between vectors from different libraries. It basically removes the need to explicitly construct a vector of type B when you have a vector of type A.
It's roughly equivalent to using a reinterpret_cast on your vectors, but the code tries really hard to make sure the cast is valid in all cases, wrapping it nicely behind a clean interface. Implicit constructors also allow to convert easily from one type to the other.
We all need vectors in our code, and there are quite a few implementations out there. They all do their job, and in terms of interface they are pretty similar to each other, but not *exactly* the same thing.
While this is not an issue in its own, it can become a problem when your code depends on external libraries that will provide their own vector implementation. So you might have your own implementation, MyVector, and the library will give you Ogre::Vector3, which is what you must use when you invoke Ogre's functions. If you also depend on openCV you will have yet another vector around. Microsoft Kinect SDK also imposes you its own, barebone vector implementation.
While through discipline you can manage such cases more or less well, there is one major issue that I couldn't work around: your own library code. I like to write code I think I will re-use at some point in the future in libraries separate from the main program. Odds are at some point you will need some maths that are not covered by the library vector you are using. Unless it's something really particular, I will probably try to write it in some library so that other projects can benefit from my new formula. The good thing about low level libraries is that they have very few dependencies, or none at all, and can thus be carried around and adapted to a variety of projects. It's all nice and good, but how would you add your new function taking a Ogre::Vector3 into a library that depends on nothing without bringing in the whole Ogre3D library?
This is where my class comes into play.
## Implementation ##
The basic idea is that all vectors will somehow give you access to some x, y, z coordinates. You will do that very frequently in your code, maybe on different platforms and in time critical parts of your code, so virtual calls (or equivalent) are not an option. This library is entirely templated, and will let you wrap any type inside a Vec. In fact, you could even just use this wrapper as your vector of choice, using an `std::array<float, 3>` as the wrapped type and forget how it's working under the hood.
Unfortunately the fact that you use `Vec<std::array<float, 3>>` and `Vec<Ogre::Vector3>` also means that you still have several types of vectors in your program. The good news is that conversion is automated (no more `Ogre::Vector3(other_vec.X, other_vec.Y, other_vec.Z)` in your code) and often it will not even be necessary, as casting will suffice.
You still need to type some code in order to get started using the vector wrapper class, which is basically your custom implementation of a struct listing informations about the wrapped type. That's not a lot of code.
## Features ##
### Automated conversion ###
You can assign or construct a `Vec<A>` from a `Vec<B>`, provided they have the same dimensions and the assignemnt operator is able to convert B's scalar type to that of A.
### Access the wrapped type ###
Through the `data()` method you can always obtain a ref to the wrapped type.
### Dimension change ###
Depending on the wrapped type, Vec can provide methods to get a Vec of lower or higher dimensions. For example, `Vec<Ogre::Vector3>` can optionally expose a `xy()` method that is equivalent to `Vec<Ogre::Vector2>(vec3.x(), vec3.y())`.
### Casting ###
You can cast `Vec<A>` to `Vec<B>` if they have the same layout and if `B` is only as big as the coordinates part in `A`.
For example you could conceptually wrap a type A:
struct A {
int something;
float x, y, z;
std::string something_else;
};
and still have it castable to type B:
struct B {
float x, y, z;
};
Vectors must still have the same dimensions. `cast<type>()` is only available when you define `offset_x` (see <u>Usage</u>). Either way, to cast (reinterpret_cast actually, be careful about the strict aliasing rule) you can just do the following:
typedef Vec<A> avec;
typedef Vec<B> bvec;
avec a(0);
some_function(bvec(a)); //creates a temporary
some_function(a.cast<bvec>()); //reinterpret_cast
## Usage ##
In this example we will adapt `std::array<float, N>` and `Ogre::VectorN`. In your real code, write the VectorWrapperInfo specializations for your own wrapped types in some header file, and just include it in your code to use Vec with those types.
