Affine transforms - matrix algebra

[Martin Desruisseaux]

Le 12/07/12 19:42, Pavel Safrata a écrit :
> Even in this case I don't really like the inconsistency with List 
> which allocates new array if the given one is not sufficient. I think 
> we either need to choose one of the numbers or do
>
> public double[] toArray(MatrixArrayType type[, double[] a)
>
> where MatrixArrayType is an enum with values MAT_2D, MAT_3D, MAT_Full, 
> requesting 6, 12, 16 numbers respectively. How does that sound?

Maybe I lack a little bit of JavaFX knowledge here, but how do we known 
for which axes are the 2D-relevant elements? Do the (x,y) plane is 
special enough in the JavaFX infrastructure for stating that the Affine 
2D elements will never be (x,z) or (y,z)? Otherwise instead of MAT_2D we 
could define MAT_XY, MAT_XZ and MAT_YZ.

But maybe there is an other issue: when I look at the matrices 
documented in the javadoc of every Transform subclasses except Affine 
(and maybe Rotate when the axis of rotation is not Z), I see X and Y 
axes that are independent of Z (i.e. the coefficients in the third 
column are zero for the first (X) and second (Y) rows). It make sense 
for those Transforms to extract a sub-transform for only the (x,y) 
plane, because transformations on that plane do not depend on the z value.

But in the Affine class, the mxz and myz coefficients can be anything. 
Consequently the transformations on the (x,y) plane may not be 
independent of z. If we extract the 2D part of that matrix, we would 
need to specify in the javadoc that this two-dimensional transform is 
valid only for z=0. For other values of z, the translation terms of the 
two-dimensional transform may be different. Do we really want to afford 
this additional complexity?

     Martin