Re: units updater for figures is broken when using fltk

[Michael Goffioul]

On Sat, Feb 4, 2012 at 9:57 PM, Ben Abbott <address@hidden> wrote:
> On Feb 4, 2012, at 4:42 PM, Michael Goffioul wrote:
>
>> On Sat, Feb 4, 2012 at 9:32 PM, Ben Abbott <address@hidden> wrote:
>>>
>>> On Feb 4, 2012, at 2:46 PM, Ben Abbott wrote:
>>>
>>>> Running the script below ...
>>>>
>>>>       close all
>>>>       graphics_toolkit fltk
>>>>       figure
>>>>       position = get (gcf, "position")
>>>>       set (gcf, "units", "normalized")
>>>>       position_pre_drawnow = get (gcf, "position")
>>>>       drawnow
>>>>       position_post_drawnow = get (gcf, "position")
>>>>
>>>> ... produces the result ...
>>>>
>>>>       position =
>>>>
>>>>          300   200   560   420
>>>>
>>>>       position_pre_drawnow =
>>>>
>>>>          0.17798   0.19358   0.33333   0.40856
>>>>
>>>>       position_post_drawnow =
>>>>
>>>>           0   44    1    0
>>>>
>>>> I have no idea what is happing here, and my debugging skills for c++ are 
>>>> embarrassingly poor. Does anyone have any insight ?
>>>>
>>>> There's a bug report. Link below.
>>>>
>>>>       https://savannah.gnu.org/bugs/index.php?35430
>>>>
>>>> Ben
>>>
>>> I gave up on myself a bit too early.
>>>
>>> I've attached a changeset that fixes the problem for me.
>>
>> That (setting units to "pixels" during redraw) should not be needed.
>> Looks to me like a workaround for the real problem. For instance, a
>> figure redraw can happen outside the drawnow function.
>>
>> Michael.
>
> Ouch ! You are obviously correct. If I eliminate the drawnow(), the figure 
> shrinks to [1x0] size in the UL of my screen.
>
> As this isn't needed for gnpulot, should this be handled in __init_flkt__ 
> (__fltk_redraw__).
>
> Or is there somewhere in graphics.cc that bottle-necks all redraws ?

No. A redraw can be triggered in the backend itself due to some user
action. Anyway, temporarily changing the units is not a solution. This
is just a hack that hides the real problem, which I think is the
following: in __init_fltk__.cc, plot_window::draw() maps figure's
position property to the FLTK window location (which is in pixels).
This is only valid when the figure's units is in pixels [1]. The
proper implementation is to use the
figure::properties::get_boundingbox() method to obtain the figure
location in pixels.

Michael.

[1]: I'm also wondering how that piece of code is handling the fact
that in Matlab world (0,0) corresponds to the bottom-left corner of
the screen, not the top-left corner; but that's another story.