[cairo] [patch] enable projective transformations

[Arjen Nienhuis]

On Tue, Aug 17, 2010 at 1:40 PM, Andrea Canciani <ranma42 at gmail.com> wrote:
> On Tue, Aug 17, 2010 at 1:24 PM, Arjen Nienhuis <a.g.nienhuis at gmail.com> wrote:
>> On Tue, Aug 17, 2010 at 12:56 PM, Andrea Canciani <ranma42 at gmail.com> wrote:
>>> On Tue, Aug 17, 2010 at 12:52 PM, Arjen Nienhuis <a.g.nienhuis at gmail.com> wrote:
>>>>> To use it for 3D transformations you would need 4x3 (and 4x4 if you
>>>>> want projections, too).
>>>>> The question was: what would 3D transformations be used for?
>>>>> (Remember: cairo uses 2D surfaces, thus the third dimension input
>>>>> would be constant and its output would get discarded "soon")
>>>>>
>>>>
>>>> Use case: 2 windows with buttons that 'stick out'.
>>>>
>>>> def draw_window(ctx, w):
>>>>  draw_background(ctx, w.background)
>>>>  ctx.save()
>>>>  ctx.translate_z(-10)
>>>>  draw_buttons(ctx, w.buttons)
>>>>  ctx.restore()
>>>>
>>>> def main():
>>>>  draw_window(ctx, w1)
>>>>  ctx.rotate_y(0.5)
>>>>  ctx.translate(300, 0)
>>>>  draw_window(ctx, w2)
>>>>
>>>> I think this needs a 4x3 matrix. You need the value of the rotation
>>>> around the y axis to make translate_z have the right effect.
>>> Are you saying that if you want to use 4x3 transforms (rotate_y), you
>>> need 4x3 transforms?
>>>
>>
>> I'm saying:
>>
>> I have this use case. I think it needs rotate_y and translate_z. I
>> think it needs 4x3 transforms.
> Oh, if this is the case, then I can correct you easily.
> Your use case doesn't need 3D transforms, you can obtain the same
> effect with a 2D transforms:
> let
> A = [ [ a b c d ] [ e f g h ] [ i j k l ] [ m n o p ] ]
> be your 4x4 transform (if you just want 4x3, you will have d==h==l==0 && p==1)
> when you transform a point P [x y z w] you get:
> P' = P * A = [ (ax+by+cz+dw), (ex+fy+gz+hw), (ix+jy+kz+lw), (mx+ny+oz+pw) ]
> but if P is constrained to be in 2D (i.e. z==0) you get:
> P' = P*A = [ (ax+by+dw), (ex+fy+hw), (ix+jy+lw), (mx+ny+pw) ]
> and (assuming you won't use the output z, you will just get the same
> as P*B, with
> B = [ [ a b d ] [ e f g ] [ m n p ]]

And how would that fit into my code? I'd need to calculate and keep
track of all these matrices myself. I like the fact that cairo keeps
track of that in 2d and i'd like cairo to also do that in 3d.

> Notice that you are able to use projective transforms (since w is not
> necessarily ==1).
> Even more interesting, you can do concatenate all your 3D
> transformations (4x4), throw away the z row and column and get a 3x3
> matrix that does exactly what you wanted.
> NB: concatenating, then throwing away the elements is *NOT* the same
> as throwing away those elements, then concatenating.

thats why Bill Spitzak said:
I believe if you want to concatenate transforms you need to keep 12 numbers
(a 4x3 matrix) around.

>>
>> Do you have better way to deal with this use case? Do you have
>> different use cases? Is my use case useful/clear/typical?
> The only use cases I can see for 3D transformation matrices involve 3D
> sources/destinations (for example a path whose points are not (x,y)
> but (x,y,z)), but I think this is beyond cairo scope.
>

So, what is cairo scope then? And how would my use case fit in (if it does).