Parameters passed to wp_viewport_set_destination was scaled instead of
the actual window size. In fact I noticed this problem when I was
testing the initial change (c29ba7985) but was not sure about how to fix
this and moreover, the tiling manager which I am using (sway) is
immediatelly tiling the window and all seems fine in day-to-day use
with tiled window.
The problem was only visible when the window was changed to floating:
the window content was scaled properly but the viewport was twice as
big (thus a whole window). What is worse it seems that the scaling was
done multiple times making a huge window size in some circumstances.
Now in new sway version 1.9 authors added an assert which prevents from
having a window which overlaps different renderers and this also trigger
this problem in FreeRDP, because when starting the window size was too
big.
This was leading to:
wp_viewport@31: error 2: source rectangle out of buffer bounds
or even sway crash
This commit is fixing this problem by using a proper window size values
(initially passed to a function and saved to additional variables before
scaling calculation).
Window size is now correct even when changed to a floating window.
Fix for a hybrid multimonitor configurations:
Previous code was working for me because the display with scale=2
was last on the wayland display list. After restarting it was earlier.
To fix this - just take the highest scale provided by a display.
Similar as eg. here the SDL is doing:
b5bc64aa55/src/video/wayland/SDL_waylandwindow.c (L1170)
The commit is fixing the problem when eg. a user is pressing some
modifier key to switch from active FreeRDP window and when get back to
that window, it "virtually" has that modifier key pressed, thus leading
to pressing eg. `Win+r` instead of just `r`.
The commit is fixing the following problem:
When the freerdp window gets keyboard focus and is notified,
it wrongly process keys which are *held*, not a freshly pressed.
The comment in the code explains it more.
Wayland compositors can support the wp_viewporter protocol,
which allows for the mapping of arbitrarily sized buffer
regions to output surfaces.
This result in correct scaling on HiDPI outputs with scaling
enabled.
Passing a non-zero offset to cursor buffer and then calling
wl_pointer.set_cursor doesn't make much sense, as any offset will
immediately be reset. The protocol specifies the cursor set by
wl_pointer.set_cursor to be
> The parameters hotspot_x and hotspot_y define the position of
> the pointer surface relative to the pointer location. Its
> top-left corner is always at (x, y) - (hotspot_x, hotspot_y),
> where (x, y) are the coordinates of the pointer location, in
> surface-local coordinates.
This leaves no room available for any previously set offsets to be kept
in response to wl_pointer.set_cursor.
The serial in wl_pointer.set_cursor must exactly match the one from
wl_pointer.enter, it should not use whatever serial for any input class
is the newest.
