peter@mit-amt.MEDIA.MIT.EDU (Peter Schroeder) (08/27/89)
I would like to ask some help regarding a design choice I have to make.
I've been writing a class of 3-vectors for a physical simulation system
with the obvious arithmetic operators. Now, what I really want are row
3-vectors and column 3-vectors and I want to be able to differentiate
between them. I do not want to include a type field, since the vectors are
so fundamental and deep down in the code, that I want them to be as
efficient as possible. On the other hand I want the compiler to flag the
attempted addition of a row vector to a column vector as undefined.
The choice I've made was to derive 2 classes from my base class vect:
rvector and cvector. They do not contain anything but what the base class
already contains. So far so good.
Defining the arithmetic operators on, say, row vectors is now just a matter
of calling the base class operators. However this entails extra copying
since I can call the base class `+' with derived class objects (rvector)
but the returned result is of type vect and still needs to be promoted up
to rvector again. It makes sense that the language requires this, but in my
case it seems silly, because there is no additional data in the derived
class.
Should I be defining rvector as a class with one member, vect? That would
avoid this problem I suppose.
Another thing I've been doing is to define the transpose operator:
inline rvector& T( cvector& cv) { return ( rvector& )cv; }
This way I can do things like T( c ) * c (dot product of cvector c with
itself) but flag c * c.
Am I taking advantage of a loophole here? The reason I did it this way was
to avoid actually creating a new object, just treat the old one
differently.
This is even more of an issue with my matrices. I also define the notion of
the transpose of a matrix by casting it into a tmatrix, which is just
another derived class without anything extra. But now I can have the
compiler differentiate between row major accessing and column major
accessing ( as in transposing a matrix ) when I do a multiply with a
vector, without actually having to make a new matrix and copying everything
across. I just have to define two different functions for matrix * vector.
Is that the right thing to do?
You see, I am new to C++ and would really like to have someone look over my
shoulders in cases like these...
Peter
peter@media-lab.media.mit.edu