Fixup multi-monitor support

[SimonHyde-BBC]

Multiple monitors with different resolutions wouldn't work, because clicks were generated with odd values in corners where no monitor existed. This fixes the calibration so it works correctly on a single monitor on multi-monitor setups

[kreijack]

Hi, thank for the patch.

I don't have hardware to test it, but my impression is that the "post-scaling" should be done more easily inside the add_click() lambda:

    gui.set_add_click([&](int x, int y) -> bool{
        auto x1 = (x - monitor_x)/monitor_width * overall_width;
        auto y1 = (y - monitor_y)/monitor_height * overall_height;
        return calib.add_click(x1, y1);
    });

without touching the Calibration class.

[SimonHyde-BBC]

That tweak would be even simpler than that, as the x/y coming from gui_x11 are already relative to the window, and therefore you don't need to subtract monitor_x or monitor_y, but:

1. You'll lose some precision as this becomes an integer division, and Calibration currently stores all clicked_x/y co-ordinates as integers rather than the floats it actually uses. I presume this will also have an affect on the accuracy of threshold_doubleclick and threshold_misclick as they'll now be looking at the larger post-translation values.
2. You still need to touch the Calibration class, as yu/yl/xl/xr need to be relative to monitor_width/height (and have monitor_x and monitor_y added to them), and the scaling of any translation into the pixel space of 0..1 at the bottom of the Calibration::finish() needs to be done with overall_width/height.

[kreijack]

I created the multimonitor branch. I started from your code but I removed the change inside calibrator.*. I tested it and to me seems that it behaves like your code. But if you can test it too, it would be really appreciated.

In the log I give you the credit for the idea and for the original code. Let me know if it is OK for you.

[SimonHyde-BBC]

This didn't work with any monitor with a rotated or flipped touchscreen, producing an incorrect translation, but it seems swapping the matrix product the other way around resolves all that. See my new pull request: https://github.com/kreijack/xlibinput_calibrator/pull/14