Affine transforms - matrix algebra

[Martin Desruisseaux]

The array methods may also be useful for working on matrix coefficients 
using indices rather than hard-coded access to the mxx, mxy, etc. 
properties. For example: when the affine may contain scale, rotation, 
flip and translation but no shear, at least in the 2D case (I think it 
works in 3D too), we can compute the scale factors "as if there were no 
flip or rotation" by computing the square root of the sum of the square 
of all values (except translation) on a row or a column. Whatever we 
compute on rows or on columns depend on whatever we want the scale 
factors to be expressed relative to the axes of the source coordinate 
system or the target coordinate system (this is not the same if there is 
a rotation). For this kind of computation, looping over an array using 
indices - or alternatively providing a Matrix.get(row, column) method - 
is convenient.

     Martin


Le 12/07/12 08:35, Pavel Safrata a écrit :
> Hi Kirill,
> you are not right with the performance, the elements are (possibly 
> invalid) properties so we need to call all the getters anyway, plus 
> construct the array, so the performance is actually worse. But the 
> use-case seems to be valid, especially if the other libraries support 
> such conversion. I just wouldn't call it 'asArray' which suggests that 
> the array will keep updating with the matrix changes but rather 
> 'toArray'. To be consistent we need to use doubles instead of floats 
> and we should also add a similar setter. So I propose:
>
> public double[] toArray()
> public double[] toArray(double[] a)
> public Affine(double[] matrix)
> public void setTransform(double[] matrix)
>
> Where the first two will behave similarly as if it was a List of 12 
> doubles, the latter two will throw IllegalArgumentException if length 
> of the array is not 12.
>
> Regards,
> Pavel