As it currently stands notification_init uses xctx.color to initialize
the notification colors which means that it depends on x11 having been
previously initialized and crashes otherwise. This bug was introduced in
04b327f where draw_setup() was after the startup notification was
created.
As a temporary fix the notification creation was moved under draw_setup
but for a long term solution we should look into removing these xctx
dependencies.
According to the glib docs we should use g_source_add_unix_fd rather
than g_source_add_poll. Additionally, this helps us with memory
management since we don't have to handle the allocation of dpy_pollfd.
Move the code to register the x11 event source under x_win_create so
multiple windows can be created and be properly registered as the event
source.
One more missing piece is that the callback functions are still using
the global win variable and not getting the window from the source.
Move the double-buffering implementation and the x_win_call move into
x.c so we only need to call x_display_surface for each draw.
This isolates all uses of the cairo context and root surface outside of
x.c which will help to add support for other display servers.
Move the main window reference from xctx to draw.c and at the same time
make each X11 function that modifies the window require the pointer as a
parameter. This makes it easier to implement features that require
multiple windows.
According to glib docs the dispatch function should be G_SOURCE_REMOVE
or G_SOURCE_CONTINUE. Since this was worked properly until now I assume
we were lucky enough that true == G_SOURCE_CONTINUE but it may not
always be the case.
Subscribe to receive window creation events on the root window and raise
the window when that happens to ensure dunst is on top.
We do not raise the window when another override redirect window is
created to avoid being on top of things like screen lockers.
Fixes#160
Rather than calculating the width of the subsurface we are drawing on
everywhere simply add a return pointer to render_background.
This commit also fixes a bug where the text and icon where drawn more to
the right than usual (according to frame_width) which was caused because
we added the frame_width to the draw position while we are drawing in a
subsurface that already accounts for it.
Cleans up the clutter in draw and since only a single icon function is
called externally from icon.c it'll make things easier if we ever decide
to switch icon libraries.
Remove prefixes used to distinguish between X and draw functions in x.c,
this is obviously no longer necessary. Additionally, add appropriate
layout_ and color_ prefixes to the relevant functions.
Rename both functions to make clear, what the difference between the
previous functions _from_file and _from_path is. Also remove the
superfluous `== NULL` checks as these don't match dunst's current style.
The `dimension_t` struct was used in many different places and even held
on the geometry info at some point which is counter-intuitive as it's
best to have each struct serve a single purpose.
Change the value of geometry in settings from an unparsed string to a
struct containing the properly parsed geometry info. Since we depend on
X11 for geometry parsing this (unfortunately) introduces an X11
dependency on the settings module, this can only be resolved if we
implement our own parsing.
Copy-paste all the drawing functions from x.c to draw.c. In the spirit
of making changes easily traceable the minimal amount of changes
required to make the functions work was made. Further improvement (like
removing the cairo-xlib dependency) will be done in later commits.
Setup the initial template for draw.c to be between X11 calls and the
rest of the codebase. This prepares the file to move all the drawing
related function here in the next commit.
On Linux, CLOCK_MONOTONIC has got a bug. It does not count
onwards during sleep, albeit required by POSIX. This behavior
is reasoned with the requirement for poll().
Also the GLib people are sticking to this behavior in their
g_get_monotonic_time() function.
So we have to use a drop in replacement, which respects CLOCK_BOOTTIME
on Linux, as this is the clock, what would be CLOCK_MONOTONIC on POSIX
systems.
The PropertyNotify handling has been changed in the dropgtk branch to
ignore XEvents, when redrawing was unnecessary. But when the fullscreen
state of a window changes, we can't ignore the event, because it didn't
change the screen.
Additionally, there had been a mistake in the handling of the cur_screen:
The xctx.cur_screen field will only get updated, when the application is
visible and gets redrawn. Therefore, when a PropertyNotify event arrived
while the application had been hidden and the screens do not match
anymore, wake_up() will be called albeit being unnecessary.
Calling x_win_draw() when the screens change is also the preferable
solution over wake_up(), as there is nothing subject to change in the
queues when the displays change.
