3 - generate pointer_focus (and drag focus) on raise/lower, move
4 windows, all kinds of changes in surface stacking.
8 - dnd, figure out large object transfer: through wayland protocol or
9 pass an fd through the compositor to the other client and let them
14 How to roboustly handle failing drag, ie the case where an
15 application gets a button event, tries to activate a drag, but when
16 the server gets the drag request, the button has already been
17 released and the grab is no longer active. What's the concern:
19 - Application may set a drag cursor that doesn't revert back,
20 since a failed drag doesn't result in a pointer_focus event to
21 give focus back to the surface. We could just do that: if the
22 pointer_focus is the same surface as we tried to start a grab
23 for, just remove and give back pointer_focus.
25 Alternatively, set drag cursors only in response to drag events,
26 like drag focus. But drag_focus and drag_motion are sent to the
27 drag target, so the source surface won't always get those. We
28 may also end up setting the cursor after the drag ends, but in
29 this case the drag started and ended and we'll get a
30 pointer_focus event, which will make the application reset the
31 pointer image. Could introduce a drag start event that
32 indicates that the drag active.
34 How to handle drop decline (accept with type=NULL)
36 - Targets must send a NULL type in accept if they don't accept a
37 drop at the drag_focus/drag_motion position. Root window will
38 send a NULL type or x-wayland/root-something type if the source
41 Races between pointer motion, ending the drag, the target sending
42 accept request and the source receiving the target event.
44 - We've sent a drag focus or motion event to the source, but
45 haven't received an accept request corresponding to that event
46 and now the button is release. The compositor could wait for
47 the source to reply to outstanding focus/motion events before
48 sending the finish event to the source. Or we could send the
49 finish event through the source so that it needs to reply to the
50 finish event too. Either way, the state of the drag blocks on
51 the client. What if we drag to a client that doesn't doo dnd?
53 - copy-n-paste, store data in server (only one mime-type available)
54 or do X style (content mime-type negotiation, but data goes away
57 - Optional pointer images.
59 - Discard buffer, as in "wayland discarded your buffer, it's no
60 longer visible, you can stop updating it now.", reattach, as in "oh
61 hey, I'm about to show your buffer that I threw away, what was it
62 again?". for wayland system compositor vt switcing, for example,
63 to be able to throw away the surfaces in the session we're
64 switching away from. for minimized windows that we don't want live
67 - Initial placement of surfaces. Guess we can do, 1)
68 surface-relative (menus), 2) pointer-relative (tooltips and
69 right-click menus) or 3) server-decides (all other top-levels).
71 - When a surface is the size of the screen and on top, we can set the
72 scanout buffer to that surface directly. Like compiz unredirect
73 top-level window feature. Except it won't have any protocol state
74 side-effects and the client that owns the surface won't know. We
75 lose control of updates. Should work well for X server root window
76 under wayland. Should be possible for yuv overlays as well.
78 - what about cursors then? maybe use hw cursors if the cursor
79 satisfies hw limitations (64x64, only one cursor), switch to
80 composited cursors if not.
82 - clients needs to allocate the surface to be suitable for
83 scanout, which they can do whenever they go fullscreen.
85 - multihead, screen geometry and crtc layout protocol, hotplug
87 - input device discovery, hotplug
89 - Advertise axes as part of the discovery, use something like
90 "org.wayland.input.x" to identify the axes.
92 - keyboard state, layout events at connect time and when it
93 changes, keyboard leds
99 - synaptics, 3-button emulation, scim
101 - Figure out if we need the batch/commit scheme and what to do
102 instead. Since dropping the "copy" request, we have a race between
103 copy from back to front and reporting damage. "copy" did this
104 atomically, but copy is a rendering operation (wayland doesn't do
105 rendering) and requires synchronization between server and client
106 before client can reuse backbuffer.
108 The race condition happens when a client copies new content into
109 its window and then, before the client reports the damage, the
110 compositor then does a partial repaint (triggered by another
111 client) that only pulls in part of the repainted area. It's only a
112 one-frame glitch, as the client will submit the damage and the
113 compositor will repaint the damaged area next frame. And ideally
114 clients should do all rendering as early in the frame as possible
117 - auth; We need to generate a random socket name and advertise that
118 on dbus along with a connection cookie. Something like a method
119 that returns the socket name and a connection cookie. The
120 connection cookie is just another random string that the client
121 must pass to the wayland server to become authenticated. The
122 Wayland server generates the cookie on demand when the dbus method
123 is called and expires it after 5s or so.
125 - or just pass the fd over dbus
127 - drm bo access control, authentication, flink_to
129 - Range protocol may not be sufficient... if a server cycles through
130 2^32 object IDs we don't have a way to handle wrapping. And since
131 we hand out a range of 256 IDs to each new clients, we're just
132 talking about 2^24 clients. That's 31 years with a new client
133 every minute... Maybe just use bigger ranges, then it's feasible
134 to track and garbage collect them when a client dies.
136 - Add protocol to let applications specify the effective/logical
137 surface rectangle, that is, the edge of the window, ignoring drop
138 shadows and other padding. The compositor needs this for snapping
139 and constraining window motion. Also, maybe communicate the opaque
140 region of the window (or just a conservative, simple estimate), to
141 let the compositor reduce overdraw.
143 - multi gpu, needs queue and seqno to wait on in requests
149 - eek, so much X legacy stuff there...
151 - draw window decorations in gtkwindow.c
153 - start from alexl's client-side-windows branch
155 - Details about pointer grabs. wayland doesn't have active grabs,
156 menus will behave subtly different. Under X, clicking a menu
157 open grabs the pointer and clicking outside the window pops down
158 the menu and swallows the click. without active grabs we can't
159 swallow the click. I'm sure there much more...
161 - Port Qt? There's already talk about this on the list.
165 - move most of the code from xf86-video-intel into a Xorg wayland
168 - don't ask KMS for available output and modes, use the info from
169 the wayland server. then stop mooching off of drmmode.c.
171 - map multiple wayland input devices to MPX in Xorg.
173 - rootless; avoid allocating and setting the front buffer, draw
174 window decorations in the X server (!), how to map input?
176 - gnome-shell as a wayland session compositor
178 - runs as a client of the wayland session compositor, uses
179 clutter+egl on wayland
181 - talks to an Xorg server as the compositing and window manager
182 for that server and renders the output to a wayland surface.
183 the Xorg server should be modified to take input from the system
184 compositor through gnome-shell, but not allocate a front buffer.
186 - make gnome-shell itself a nested wayland server and allow native
187 wayland clients to connect and can native wayland windows with
188 the windows from the X server.
190 - qemu as a wayland client; session surface as X case
192 - qemu has too simple acceleration, so a Wayland backend like the
193 SDL/VNC ones it has now is trivial.
195 - paravirt: forward wayland screen info as mmio, expose gem ioctls as mmio
197 - mapping vmem is tricky, should try to only use ioctl (pwrite+pread)
199 - not useful for Windows without a windows paravirt driver.
201 - two approaches: 1) do a toplevel qemu window, or 2) expose a
202 wayland server in the guest that forwards to the host wayland
203 server, ie a "remote" compositor, but with the gem buffers
204 shared. could do a wl_connection directly on mmio memory, with
205 head and tail pointers. use an alloc_head register to indicate
206 desired data to write, if it overwrites tail, block guest. just
207 a socket would be easier.
209 - moblin as a wayland compositor
211 - clutter as a wayland compositors