"unmatrix" code for 2D transforms is incorrect. It can lead to matrices that are not consistent with 3x2 matrices.
Is this just editorial though?
This is feedback from D. Baron. Mozilla still differs between 2D matrices and 3D matrices, but uses quaternions on decomposing 2D matrices as well. Not sure if we can add a informative reference to decomposing code for 2d matrices. Right now I don't have a good reference.
We spent a bit of time in the meeting looking at the behavior of the lower rows (when hovered) of
The specification text changed for decomposing 2D matrices and is now conform to what Safari does. A patch for Mozilla is underway.