XRandR transform (keystone correction) not working as I would expect

[AJ Heller]

Oh good! That works well. Thank you very much, Glynn. I had it backwards.

Hopefully my experience here can be of some help. The man page section for
--transform is a bit misleading. For example:

> The final device coordinates of the pixel are then calculated with the
so-called homogenic division by the transformed w coordinate.

>From what I understand now, the transformation matrix doesn't directly
calculate final device coordinates, but instead calculates graphic buffer
coordinates *from* your desired final device coordinates. Maybe someone who
knows the terminology better can rewrite this section of the man page to
make its behavior clearer?


On Wed, Oct 2, 2013 at 3:19 AM, Glynn Clements <glynn at gclements.plus.com>wrote:

>
> AJ Heller wrote:
>
> > I implemented a transformation matrix to pinch the top of my screen 100px
> > on each side. My hope is for the resulting image to be an isosceles
> > trapezoid (trapezium) with the smaller base on top, larger base on bottom
> > filling the width of the screen, and with no part of the image missing or
> > drawn off the screen.
> >
> > After transforming, the top remains fixed at the width of the screen, and
> > the bottom of the trapezoidal image is drawn off the screen on either
> side.
> > I could be confused about what the transform should be doing, but
> otherwise
> > I think this is a bug.
> >
> > Have I setup my transformation incorrectly? Should I file a bug report?
>
> > I'm working with 1440x900 resolution. I mapped the four corners to my
> > desired output coordinates:
> >
> > (0,0) => (0,0)
> > (1440,0) => (1440,0)
> > (0,900) => (100,900)
> > (1440,900) => (1340,900)
>
> > My transformation matrix is:
> > 1, 0.30769, 0
> > 0, 1.38462, 0
> > 0, 0.00043, 1
>
> Based upon the xrandr manual page and your pictures, that's not what
> you want, and will produce the results you describe.
>
> First, the transformation converts from output coordinates to input
> coordinates. So the above will result in the bottom corners of the
> screen being mapped 100 pixels in from the edge (i.e. you'll lose 100
> pixels at each side). Except, it's actually 200 pixels for the matrix
> which you give.
>
> Second, 0,0 is the top-left corner, not the bottom-left (this isn't
> stated in the manual page, but it explains why the the oversized edge
> is the bottom edge rather than the top edge).
>
> If I'm understanding you (and xrandr) correctly, the matrix you want
> is:
>
> 1.16129, 0.12903, -116.12903
> 0.     , 1.16129,    0.
> 0.     , 0.00018,    1.
>
> This will map:
>
>  output        input
>
>  100,   0       0,   0
> 1340,   0    1440,   0
>    0, 900       0, 900
> 1440, 900    1440, 900
>
> --
> Glynn Clements <glynn at gclements.plus.com>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.x.org/archives/xorg/attachments/20131002/63c9f8be/attachment.html>