Limegarden.net Personal site of Wouter Lindenhof

Solved undesired template specification

A while ago (why do I always start with that) I wrote an blog entry about undesired template specification, what to encounter and how to work around it.

Anyway here is a quick definition of two structures I will be using in the article:

/**
 * I will be using the following structures throughout the article
 */
template <typename T> struct Vector3<T> {
	union {
		struct { T x, y, z; }
		T array[3];
	};
	Vector3<T>() : x(0), y(0), z(0) {}
	Vector3<T>(T nx, T ny, T nz) : x(nx), y(ny), z(nz) {}
};

template <typename T> struct Vector4<T> {
	union {
		struct { T x, y, z, w; }
		T array[4];
	};
	Vector4<T>() : x(0), y(0), z(0), w(0) {}
	Vector3<T>(T nx, T ny, T nz, T nw) : x(nx), y(ny), z(nz), w(nw) {}
};
typedef Vector3<unsigned int> Vector3u;
typedef Vector3<float>        Vector3f;
typedef Vector4<unsigned int> Vector4u;
typedef Vector4<float>        Vector4f;

struct SVertex
{
	Vector4f pos;
	Vector3f normal;
	Vector2f texcoord;
};

Because of Brick (3D random dungeon generator that takes design into account) I have noticed that there is one thing I do failry often:

Vector3f position;
SVertex vertex;

/* Need to draw it, so I store position in vertex */
vertex.pos = position; // ERROR! Trying to assign Vector3f to Vector4f!!

And finally I used defines to do the conversion for me:

#define VEC3TOVEC4(v) Vector4f((v).x, (v).y, (v).z, 0)
#define VEC4TOVEC3(v) Vector3f((v).x, (v).y, (v).z)

/* New code becomes */
vertex.pos = VEC3TOVEC4(position); // Works

Of course the above code is not nice to look at and I find it even plain ugly, but it works. However I don't want to do that in future projects (it feels like a hack), I would need to define something like that for every type (float, double, unsigned int et cetera) and on top of that it generates warnings:

Vector3<double> position;	// Notice it is unsigned!
SVertex vertex;

/* Need to draw it, so I store position in vertex */
vertex.pos = VEC3TOVEC4(position); // Works, but generates warning about losing information

But I wouldn't be writing this post unless I tackled that little issue, and for once I can add that the solution is quite nice as well.

Vector3<double> position;	// Notice it is unsigned!
SVertex vertex;

/* Need to draw it, so I store position in vertex */
vertex.pos.Set(position.array, 3); 		// Works, no errors and no warnings
vertex.pos = Vector4f(position.array, 3);	// Works fine as well

So what did I change?

Well, I used mutliple template (one for the class and for the function/constructor). This looks something like this:

template <typename T> struct Vector4
{
	/* ... */
	template <typename R>
	explicit inline Vector4<T>(const R* values, const unsigned int elements/*=4*/);
	template <typename R>
	inline Vector4<T>& Set(const R* values, const unsigned int elements/*=4*/);
	/* ... */
};

// Implementation
template <typename T> template <typename R>
Vector4<T>::Vector4(const R* values, unsigned int elements)
	: x(elements > 0 ? (T)values[0] : 0), y(elements > 1 ? (T)values[1] : 0)
	, z(elements > 2 ? (T)values[2] : 0), w(elements > 3 ? (T)values[3] : 0)
{
}
template <typename T> template <typename R>
Vector4<T>& Vector4<T>::Set(const R* values, const unsigned int elements)
{
	x = (elements > 0 ? (T)values[0] : 0);
	y = (elements > 1 ? (T)values[1] : 0);
	z = (elements > 2 ? (T)values[2] : 0);
	w = (elements > 3 ? (T)values[3] : 0);
	return *this;
}

If you take a look at the code I think it is quite clear except that you might have some questions.

1. Q: Why do you use [cci_cpp]explicit[/cci_cpp] with the constructor?
   A: Because that prevents the implicit use of constructors. If I would allow it a [cci_cpp]Vector4u[/cci_cpp] could be implicit assigned to [cci_cpp]Vector3f[/cci_cpp], although it would be missing an argument. However I think that when you are converting, you should be somewhat aware of it, especially when it can be expensive.
2. Q: Why have you commented out the default value for [cci_cpp]elements[/cci_cpp]?
   A: Because you don't know how many elements there are in [cci_cpp]values[/cci_cpp] might be (there is a method to find out).
3. Q: So why don't you find out automatically and what is with those conditionals in the constructor?
   A: Those two are related. By telling it explicitly there is a real good chance that the compiler removes the conditionals, so the [cci_cpp] ( check ? true-value : false-value) [/cci_cpp] check will be removed.