Vector and matrices clases are too granular for polymorphism, and the fact is, a 3D vector is NOT a 2D vector.
A 2D vector IS a 3D vector - whose''s z component is always 0.
The equations that govern the behavior of 2D vectors do not apply to 3D vectors - however the 3D generalizations hold true for 2D vectors. So if you used inheritence, you''d have to start with the N case and specialize downward. You could represent every 2D vector as a 3D vector as a nD vector - but for efficentcies sake you want a serperate class that consumes less space and avoids unneccessary calculations.

If you define any constructors, make sure to also define an empty default constructor that does nothing, so that you don''t have to pay initialization cost when it''s unneeded.

You could use a namespace or two if you want to organize it a little.

If you really wanted to make generic vector (and matrix) classes, you could use templates and (2D) iterators. That way the you could write the algorithms once, and they would work with any classes that provided the correct type of iterator. Vector algorithms require random-access iterators, and a matrix algorith would require a new kind of random-access 2D iterator.
If you used inline templates, all the abstraction overhead would compile away to nothing. There are template libraries of such things out there - The Matrix Template Library from Notre Dame comes to mind.

Magmai Kai Holmlor

"Oh, like you''ve never written buggy code" - Lee

"What I see is a system that _could do anything - but currently does nothing !" - Anonymous CEO

My 2 cents:

Look at it more abstractly. It's not that a 3d vector IS_A a 2d vector, or vice versa, it's that a 2d and 3d vector IS_A vector.

Write a pure-virtual class CVector, and then C2DVector and C3DVector which inherit from CVector.

CVector would define the inherent member functions such as dot producting and operator overloading or whatever you want.

same goes for matrices, a 2x2 and 3x3 matrix IS_A matrix, so you could do the same.

that's the way i'd do it anyway.

PS: remember that a 3x3 matrix HAS_A 6 3d vectors, and a 2x2 matrix HAS_A 4 2d vectors. So if you do the vectors first then you can save yourself some code (maybe) if your derived matrix classes HAS_A 2d or 3d vectors.

Edited by - MelvinElvin on December 28, 2001 7:13:27 AM

[edit: oops, nearly toasted your post while trying to write a response, those edit and quote button should be further apart]


Edited by - Magmai Kai Holmlor on December 30, 2001 4:40:23 AM

Magmai''s right. Vectors, matrices and other lowlevel primitives should be excluded from inheritance. There is simply too much overhead associated with virtual calls to justify it.

You can define a few custom operators to prevent typecasts. Like so:

operator Vector3(const Vector2 &v){    return Vector3(v.x, v.y, 0.0f);} 

quote:
Look at it more abstractly. It''s not that a 3d vector IS_A a 2d vector, or vice versa, it''s that a 2d and 3d vector IS_A vector.

True, they are both vectors, but a 2D vector is a 3D vector, however the converse is not true. Also, the computation required for a 3D dot product can be dramaticaly simplified from the general case - and simplified more for 2D (if such a dot product makes any sense at all).

OOP gone horribly wrong:

quote:
Write a pure-virtual class CVector...

Realize that would cause a 50% space increase for the storage of 2D vectors (33% for 3D, etc...)? And I''d hazard about a 20% performance penalty?

For a 100x100 matrix, you might never know the difference...

Magmai Kai Holmlor

"Oh, like you''ve never written buggy code" - Lee

"What I see is a system that _could do anything - but currently does nothing !" - Anonymous CEO