Commit Graph

5813 Commits

Author SHA1 Message Date
Giovanni Campagna
7360aece74 wayland: don't require Clutter backend variables to be set from outside
When running under mutter-launch, we can assume we're running on
bare metal, and set the clutter backend appropriately.

https://bugzilla.gnome.org/show_bug.cgi?id=706421
2013-09-11 16:01:09 +02:00
Giovanni Campagna
b5f3238f6f Stop messing with process groups
We can be launched by gnome-session now, which implies gdb must be
attached from outside, and the Ctrl-C problem is gone.

https://bugzilla.gnome.org/show_bug.cgi?id=706421
2013-09-11 16:01:09 +02:00
Giovanni Campagna
227fb56103 Set DISPLAY and WAYLAND_DISPLAY for dbus activated services and for autostarted apps
Call the appropriate method on gnome-session so that autostarted
and bus activated apps see the X server and wayland socket.

https://bugzilla.gnome.org/show_bug.cgi?id=706421
2013-09-11 16:01:09 +02:00
Giovanni Campagna
5ae52473c9 Don't translate debug logs and warnings
Warnings that are going to the journal should be not translated:
they're not user visible, and translating them would just make
bug reporting harder (as now the developers need to understand
what the warning is saying)

https://bugzilla.gnome.org/show_bug.cgi?id=707897
2013-09-11 16:01:09 +02:00
Giovanni Campagna
7b537d6b8f Fix regression in mutter-wayland translations
Switching meta/util.h to gi18n.h was wrong, mutter is a library
and needs gi18n-lib.h, but that cannot be included from a public
header (since it depends on config.h or command line options),
so split util.h into a public and a private part.

https://bugzilla.gnome.org/show_bug.cgi?id=707897
2013-09-11 16:01:07 +02:00
Giovanni Campagna
4494888b82 Add an option to run as a wayland compositor
Move the option from the mutter binary to libmutter, and
rename from --nested to --wayland.
This allows gnome-shell to drop the set_is_wayland_compositor()
call.

https://bugzilla.gnome.org/show_bug.cgi?id=707897
2013-09-11 16:00:47 +02:00
Giovanni Campagna
004ee2d3b5 MetaPluginManager: don't require plugins to pass events to clutter
We don't want the shell to know if it's running on ClutterX11 or
not, so we should forward the event ourselves.

https://bugzilla.gnome.org/show_bug.cgi?id=707482
2013-09-10 17:26:15 +02:00
Giovanni Campagna
168ea64a45 wayland: use symbolic constants for interface version
Replace magic numbers scattered around the code with proper
macros collected in one header file.

https://bugzilla.gnome.org/show_bug.cgi?id=707851
2013-09-10 15:55:11 +02:00
Giovanni Campagna
73ee491281 wayland: fix interface versioning
Add MIN(...) with the interface version actually implemented
to all resource constructor, so that we never risk seeing requests
we don't implement (and consequently segfault)

https://bugzilla.gnome.org/show_bug.cgi?id=707851
2013-09-10 15:54:42 +02:00
Giovanni Campagna
95a9655412 wayland: fix deprecation warnings
Replace the deprecated APIs with the modern variants (which
also give us full control on the versioning).
Also, note that we only support version 2 of wl_seat. Version
3 is for multiple pointer/keyboards for a client, but
we don't implement that yet. If a client requests version 3
of wl_seat or make a version 3 request, it will get a protocol error.

https://bugzilla.gnome.org/show_bug.cgi?id=707851
2013-09-10 15:54:37 +02:00
Giovanni Campagna
72b6699efb util: fix warnings when compiling gnome-shell-wayland
Replace our version of the gettext macros with those in glib,
so that they don't conflict with gnome-shell.

https://bugzilla.gnome.org/show_bug.cgi?id=707851
2013-09-10 13:47:30 +02:00
Giovanni Campagna
3e341e83d9 mutter-launch: remove PAM usage
Left-overs from the weston-launch import

https://bugzilla.gnome.org/show_bug.cgi?id=707584
2013-09-09 18:09:08 +02:00
Giovanni Campagna
5b3340e585 MetaCursorTracker: add methods for setting the cursor visibility
clutter_stage_show_cursor()/hide_cursor() only works in the X11
backend (where someone else is in charge of showing the cursor),
and even then, it has confusing effects when running nested wayland,
so an abstraction layer is needed.

https://bugzilla.gnome.org/show_bug.cgi?id=707474
2013-09-09 18:04:07 +02:00
Giovanni Campagna
6f4f611ba8 MetaCursorTracker: extend with query pointer abilities
We need an abstraction in gnome-shell for XQueryPointer, and
MetaCursorTracker seems a good place for it.

https://bugzilla.gnome.org/show_bug.cgi?id=707474
2013-09-09 18:04:07 +02:00
Giovanni Campagna
450afbaf51 wayland: reimplement keyboard state handling properly
We can't rely on clutter's xkb_state, because that's updated
when events are pulled from the kernel, not when we see them.
Instead, use the new clutter API to get the full modifier state
from the event (which, as a side effect, also works when clutter
is using the X11 backend for running nested).

https://bugzilla.gnome.org/show_bug.cgi?id=706963
2013-09-09 17:59:45 +02:00
Giovanni Campagna
28708e4317 MetaWaylandSeat: don't use use events to count pressed buttons
Use the modifier mask instead, as events can get lost if there
is a clutter grab or if some other actor is capturing events.

https://bugzilla.gnome.org/show_bug.cgi?id=706124
2013-09-09 17:59:45 +02:00
Giovanni Campagna
9f64f2cf4f Rename META_POWER_SAVE_UNKNOWN to UNSUPPORTED
It conveys more clearly what this value means.

https://bugzilla.gnome.org/show_bug.cgi?id=707649
2013-09-09 15:33:53 +02:00
Giovanni Campagna
2b1eb206d3 Monitor: don't shortcut set_power_save if the value is the same as the old one
We don't get notifications from X11 when the mode is reset, so
our cached value can get stale. To work around that, always forward
requests to the backend (and let it deal with ignoring the change
if wanted)

https://bugzilla.gnome.org/show_bug.cgi?id=707649
2013-09-09 15:33:53 +02:00
Giovanni Campagna
a6dc454c49 wayland: constraint the pointer onto visible monitors when running on evdev
Use the new Clutter hook to make sure the pointer never enters
the dead area caused by the different monitor sizes.

You don't realize how much X is doing for you until you lose it...

https://bugzilla.gnome.org/show_bug.cgi?id=706655
2013-09-09 12:08:47 +02:00
Giovanni Campagna
c2bf44fa19 Bump version to 3.9.91
Update NEWS
2013-09-05 11:01:53 +02:00
Adel Gadllah
2d67b01c3a window-actor: Use more conservative paint volume
We know which part of the window is not obscured, so take advantaged of that
to limit the size of the paint volume that we report.
2013-09-04 20:08:24 +02:00
Neil Roberts
6014d31dea Don't create a dummy texture for the texture pipeline template
The meta_create_texture_pipeline function used to create a dummy 1x1
texture so that it could make sure that the all of the state that
affects the shader generation would be set on the template pipeline so
that Cogl could share the pipeline's shader with any other pipelines
that are just rendering a texture. This is no longer necessary because
the only thing that affects the shader generation is the texture type,
not the actual texture data and Cogl now has a function to explicitly
set the texture type which we can use instead. Additionally even if
the template mechanism is not used at all Cogl will still end up
reusing the same shader because it now has a shader cache which is
indexed by the pipeline state so pipeline's don't strictly need to
share ancestry in order to take advantage of it. However we still
might as well use the function because if there is a common ancestry
it is faster to look up the shader because Cogl doesn't need to hash
the pipeline state.

https://bugzilla.gnome.org/show_bug.cgi?id=707458
(cherry picked from commit c5bf60eab4)
2013-09-04 15:29:04 +01:00
Jasper St. Pierre
bf78f067a2 Make sure to always call va_end 2013-09-03 16:10:10 -04:00
Jasper St. Pierre
ddb9929dad compositor: Remove some uninitialized variables 2013-09-03 16:10:10 -04:00
Jasper St. Pierre
2098ec2d16 theme-parser: Remove a duplicate paste
How did this sneak in there?
2013-09-03 16:10:10 -04:00
Jasper St. Pierre
97eece6607 display: Add some missing breaks
If we somehow get an event with a wrong device ID, we should not
be comparing bad event IDs.
2013-09-03 16:10:10 -04:00
Ray Strode
7ac9a6e241 compositor: initial has_window to FALSE in sync_actor_stacking
Spotted by Jasper, discussed on IRC.
2013-09-03 16:09:05 -04:00
Giovanni Campagna
b53bf0e8c2 Fix distcheck
Missing translation files, missing dists and some -Werrors.
2013-09-03 17:00:18 +02:00
Giovanni Campagna
a26ded47d9 Add a private gtk-mutter protocol
Add a new interface, gtk_shell, than can be used by gtk to
retrieve a surface extension called gtk_surface, which will be
used to communicate with mutter all the GTK extensions to EWMH

https://bugzilla.gnome.org/show_bug.cgi?id=707128

Add support for GTK application menus

To do so, we need to be able to set surface state before creating
the MetaWindow, so we introduce MetaWaylandSurfaceInitialState as
a staging area.
The gtk-shell-surface implementation would either write to the
initial state, or directly to the window.

At the same, implement set_title and set_class too, because it's
easy enough.

https://bugzilla.gnome.org/show_bug.cgi?id=707128
2013-09-03 10:57:28 -04:00
Giovanni Campagna
cad9e14463 Move surface state tracking and surface interface to a separate file
Move everything surface related from meta-wayland.c to meta-wayland-surface.c

https://bugzilla.gnome.org/show_bug.cgi?id=707128
2013-09-03 10:57:28 -04:00
Giovanni Campagna
806d5939e3 wayland: split headers and distribute structure definitions
Instead of having all structures in one huge headers, move them
in the appropriate place, and create one header for surface state.

https://bugzilla.gnome.org/show_bug.cgi?id=707128
2013-09-03 10:57:28 -04:00
Giovanni Campagna
0f0c23fbab wayland: generalize ShellSurface to SurfaceExtension
We want to implement new shells and new extension interfaces
for wl_surface, so generalize the book-keeping structures.

https://bugzilla.gnome.org/show_bug.cgi?id=707128
2013-09-03 10:57:28 -04:00
Jasper St. Pierre
5d1fcc26c0 Update .gitignore 2013-09-03 10:57:28 -04:00
Giovanni Campagna
cbe5b6b3bc meta-wayland: intersect the damage region with the window size before applying
According to the wayland documentation, damage outside the
window size is ignored.
This happened with xwayland+wlshm (causing a GL error when calling
TexSubImage2D), probably due to not resizing the buffer
until we receive the corresponding X event.
Might also be an off-by-one in xwayland, as the window size did
not actually change.

Note: we might want to take the configure_notify path instead,
and keep the GL/clutter size consistent with wayland rather than
X, because in the end that's what matters for events and composition.

https://bugzilla.gnome.org/show_bug.cgi?id=706289
2013-09-03 10:16:47 -04:00
Giovanni Campagna
590cf4e832 MonitorXrandr: check the event timestamps before reconfiguring
If, checking the event timestamps, we see that a new configuration
was explicitly requested by an another XRandR client, don't proceed to
apply the intended configuration again, even if looking at the
EDIDs it appears that the outputs changed.
This works around some buggy Xorg drivers (qxl, vbox) that generate
a new serial number everytime the user resizes the host window.

https://bugzilla.gnome.org/show_bug.cgi?id=706735
2013-09-02 17:36:03 +02:00
Cosimo Cecchi
22ca820c44 idle-monitor: fix event propagation to devices
device_id_max is set to the device_id in ensure_device_monitor(), but we
will loop only to (device_id_max - 1) when propagating the sync XEvent
down, missing the device correspondng to device_id_max.

https://bugzilla.gnome.org/show_bug.cgi?id=707250
2013-09-01 17:51:10 -07:00
Jasper St. Pierre
b1206ceb66 wayland: Ensure that opaque / input regions are set at commit time
The protocol specification says that opaque / input regions should be
considered pending state and should only be actually swapped out when
the surface is committed, so it can be set atomically.
2013-08-31 13:29:02 -04:00
Giovanni Campagna
ad4053ab84 wayland: fix pointer focus for destroyed surfaces
We had an assertion in meta_wayland_surface_free() that after
a repick() we would not choose the freed surface, but that didn't
consider surfaces destroyed while holding the implicit pointer
grab (ie, because the user clicked on the X button). In that case,
we need to bypass the grab infrastructure and explicitly unfocus
the dead surface.

https://bugzilla.gnome.org/show_bug.cgi?id=706982
2013-08-30 16:07:00 +02:00
Colin Walters
7bd4e6ecb0 weston-launch: Don't segfault if passed no arguments
Probably not exploitable since I think Linux always puts NULL there,
but let's actually check.
2013-08-30 08:26:31 -04:00
Giovanni Campagna
deeb1db1ac wayland: don't free surfaces that have a window associated
After a MetaWaylandSurface is associated with a MetaWindow, it
should be freed only when the MetaWindow is unmanaged. For wayland
clients, the window is unmanaged when the resource is destroyed,
but for X11 clients we want to wait for the unmap event.

https://bugzilla.gnome.org/show_bug.cgi?id=705917
2013-08-30 10:37:25 +02:00
Giovanni Campagna
7eb4bfbea3 wayland: implement support for plugin modality
Calling XIGrabDevice has no effect under wayland, because the
xserver is getting events from us. Instead, we need to use our
own interfaces for grabs.
At the same time, we can simplify the public API, as plugins
should always listen for events using clutter.

https://bugzilla.gnome.org/show_bug.cgi?id=705917
2013-08-30 10:37:25 +02:00
Giovanni Campagna
59b274f12f MetaWaylandKeyboard: use the new clutter-evdev support for xkbcommon
We need to track the full xkb_state to have the necessary information
to send to the clients, otherwise they may get confused and lock
or invert the modifiers. In the evdev backend, we just retrieve the
same state object that clutter is using, while in the other backends
we fake the state using what clutter is providing (which is a subset
of what X11 provides, which would be necessary to have full state)

https://bugzilla.gnome.org/show_bug.cgi?id=705862
2013-08-30 10:37:25 +02:00
Giovanni Campagna
e311cef013 Add keybindings for switching VTs
Once mutter is started from weston-launch on its own VT, there is
no way to change VT again (for example to actually start an application),
because the keyboard is put in raw mode.
So introduce some keybindings mimicking the standard X ones (Ctrl+Alt+Fn)
that switch the VT manually when activated.

https://bugzilla.gnome.org/show_bug.cgi?id=705861
2013-08-30 10:32:55 +02:00
Giovanni Campagna
e72f81c24f wayland: add TTY and DRM master management
Now that we have a setuid launcher binary, we can make use of
using a private protocol through the socket we're passed at startup.

We also use the new hook in clutter-evdev to ask mutter-launch for
the FDs of the input devices we need, and we emulate the old X
DRM lock with a nested GMainContext without sources.

In the future, mutter-launch will be replaced with the new logind
API currently in development.

https://bugzilla.gnome.org/show_bug.cgi?id=705861
2013-08-30 10:32:55 +02:00
Giovanni Campagna
96fa518576 mutter-launch: augment with VT and TTY handling
Set the TTY mode appropriately at startup, and clean it up
when the compositor exits. Also, take control of VT switching,
including the calls to drmSetMaster and drmDropMaster as appropriate.
In the future, we the kernel implements the mute evdev ioctl,
we'll also make sure that input devices are appropriately released.

https://bugzilla.gnome.org/show_bug.cgi?id=705861
2013-08-30 10:32:55 +02:00
Giovanni Campagna
e263b3624c mutter-launch: use systemd to obtain the TTY
Using the command line or an environment variable is dangerous,
as those can be spoofed to gain access to other sessions.

https://bugzilla.gnome.org/show_bug.cgi?id=705861
2013-08-30 10:32:55 +02:00
Giovanni Campagna
1c34f0b342 mutter-launch: make sure that the spawned binaries sees the right libraries
Being a setuid binary, our LD_LIBRARY_PATH is cleared by glibc at
startup, but we need the spawned binary to see it, otherwise
jhbuild doesn't work, so hardcode it using the configured libdir.

https://bugzilla.gnome.org/show_bug.cgi?id=705861
2013-08-30 10:32:55 +02:00
Giovanni Campagna
2d27873f98 mutter-launch: simplify by removing features we don't need
Remove the ability to launch as a different user, which we don't
need because we're spawned by gdm or by the user manually on the
command line.
At the same time, require an active local session, and remove
the ability to run from anywhere by being in the right user group
(which automatically gives you root-like privileges)

https://bugzilla.gnome.org/show_bug.cgi?id=705861
2013-08-30 10:32:55 +02:00
Giovanni Campagna
fd40a12213 wayland: import weston-launch setuid launcher
To run mutter as a display server, one needs to acquire and
release the DRM master, which is only possible for root, so
we take advantage of weston-launch, a small setuid helper binary
written for the weston project. We import our own slightly
modified copy of it, because weston-launch only launches weston,
for security reasons.

https://bugzilla.gnome.org/show_bug.cgi?id=705861
2013-08-30 10:32:55 +02:00
Jasper St. Pierre
57406e0a9e shaped-texture: Actually fetch rectangles from the blended_region
We checked for the blended region but actually fetched rectangles from
the clip region.
2013-08-29 18:35:18 -04:00