summaryrefslogtreecommitdiff
Commit message (Collapse)AuthorAgeFilesLines
* compositor: Detach window and actor both ways on actor disposewip/carlosg/shutdown-warningCarlos Garnacho2022-04-301-1/+5
| | | | | | | | | | | | | This is currently deferred to the window being unmanaged, however the dependency of events is too indirect to consider these things always happen one after the other. Fixes possible warnings like: (mutter:872101): GLib-GObject-WARNING **: 15:29:27.943: invalid unclassed pointer in cast to 'MetaWindowActor' Seen on shutdown, a reliable reproducer is gnome-settings-daemon test suite.
* clutter/stage: Remove unused nullable annotationCorentin Noël2022-04-291-1/+1
| | | | | | | This annotation wasn't taken into account because of the colon, but shouldn't be applied as there is no way a NULL ClutterGrab can be returned. Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/2382>
* clutter/actor-meta: Notify actor property on changesJonas Dreßler2022-04-241-1/+4
| | | | | | | Seems like this was forgotten when writing the class. While at it, add EXPLICIT_NOTIFY to the property flags. Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/2377>
* clutter: Remove device/sequence entry when TOUCH_END get filtered outJonas Dreßler2022-04-231-25/+31
| | | | | | | | | | | | | | | | | | | | Mutters event filter can prevent events from getting processed by Clutter, this can also happen for TOUCH_END/CANCEL events. Processing these events in Clutter is crucial for proper tracking of touch sequences though, that's because Clutter adds a PointerDeviceEntry to the stage on a TOUCH_BEGIN *before* going through the event filter, but removes that entry on a TOUCH_END *after* going through the filter. So Clutter really needs to see those TOUCH_END events, or else there will be a stale PointerDeviceEntry on the ClutterStage. Make sure those TOUCH_END/CANCEL events always get seen by Clutter by removing the device entry immediately when those get filtered out. Because there might still be events belonging to this sequence in the event queue of the stage, we need to flush the queue before removing the entry, too. Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/2350>
* clutter: Shuffle handling of IM reset on button pressesCarlos Garnacho2022-04-233-9/+24
| | | | | | | | | | | | | | | | | | | Unfortunately we cannot do this generically since the target of the button/touch press does matter, e.g. tapping on the OSK, or clicking the IBus candidates window. These situations should not trigger a reset. So be more selective about the situations where button/touch presses trigger an IM reset, in the case of ClutterText these are still clicks inside the actor, for Wayland's text-input it is when clicking the surface that has text_input focus. For all other situations where clicking anywhere else might make sense to trigger an IM reset are covered by the focus changing paths, that also ensure a reset before changing focus between surfaces/actors. Fixes: https://gitlab.gnome.org/GNOME/mutter/-/issues/1961 Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/2384>
* wayland: Reset ClutterInputFocus on focus changesCarlos Garnacho2022-04-231-0/+2
| | | | | | | | | | | | | | Focus changes should trigger an IM reset, as some engines do want to maybe commit the preedit string before changing focus. In addition, we do not want the preedit string to be able to move between windows/applications. Ensure that the commit string is committed when the IM deems so, and ensure we send a .done event disntinct to the .leave event, so that the client doesn't miss the commit. Fixes: https://gitlab.gnome.org/GNOME/mutter/-/issues/2030 Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/2384>
* clutter: Reset ClutterInputFocus on focus_outCarlos Garnacho2022-04-231-1/+1
| | | | | | | | | Focus changes should trigger an IM reset, as some engines do want to maybe commit the preedit buffer before changing focus. Since the preedit string is also cleared on reset(), we can do without that explicit call. Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/2384>
* clutter: Actively update ClutterText surrounding text on changesCarlos Garnacho2022-04-231-0/+4
| | | | | | | | | | | | Right now we have a bit of a mixed bag between an active model where input foci set the surrounding text without being asked for (e.g. wayland's text_input), and a passive model where the IM engines ask for content. Make ClutterText take the same side than text_input, so that dealing with those is at least consistent. Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/2384>
* clutter: Fix ClutterText ::delete-surrounding IM implementationCarlos Garnacho2022-04-231-1/+7
| | | | | | | | | The clutter_text_delete_text() function used underneath expects character offsets for both start/end position. Fix the end position passed an offset instead of that, and compesnate for the cursor position being always -1 when the caret is at the end of the string. Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/2384>
* clutter/stage: Repick when pointer actor goes unmappedJonas Dreßler2022-04-223-57/+69
| | | | | | | | | | | | | | | | | I've overseen quite an important case in commit 98a5cb37d9159737f8f1af4196420db90bfcf879: Repicking only when actors get destroyed is not enough, we actually need to repick when actors go hidden/unmapped. While we could also listen to notify::mapped just like we listen to notify::reactive, it seems better to avoid using property notifications here due to the usage of g_object_freeze/thaw_notify() in ClutterActor. It can lead to the stage receiving a notify::mapped with mapped = true for a pointer actor, which really shouldn't happen (just like notify::reactive with reactive = true shouldn't happen). Fixes: https://gitlab.gnome.org/GNOME/gnome-shell/-/issues/5124 Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/2333>
* screen-cast/src: Always allow MemFd buffer types to be selectedJonas Ådahl2022-04-221-4/+3
| | | | | | | Just because the PipeWire client can handle DMA buffers, doesnt mean we should not advertise support for MemFd. Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/2383>
* screen-cast: Test allocate DMA buffer with implicit APIJonas Ådahl2022-04-222-0/+19
| | | | | | | | | | DMA buffers might be allocatable, but it doesn't mean the driver doesn't fail when we try to allocate a buffer with an implicit modifier. Using the proprietary NVIDIA driver for example, it will fail. Lets catch this up front and avoid advertising DMA buffer support when we know it won't work. Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/2383>
* wayland: Always acknowledge state changing .commit() events with .done()Carlos Garnacho2022-04-221-0/+2
| | | | | | | | | | As of currently, we only emit .done() on actual changes coming from the ClutterInputMethod/ClutterInputFocus. With the recent changes in the interpretation of serials, it becomes more important now that the compositor acknowledges every .commit done by the client, in order to keep them feeding future IM state updates. Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/2365>
* wayland: Always update preedit with text_input.done()Carlos Garnacho2022-04-221-5/+21
| | | | | | | | Compensate the protocol statelessness with our ClutterInputFocus statefulness. This becomes more necessary now, since sending consecutive .done() events is now considered acceptable behavior. Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/2365>
* backends/native: Update cursor also when scaled or transformedSebastian Wick2022-04-211-1/+7
| | | | | | | | | | MetaCursorRendererNative only updates the cursor state when the underlying texture changes. The cursor scale and transform do not trigger updates. This results in wrong cursor orientations on rotated displays. Use both texture changes and scale and transformation changes to figure out when to update the cursor state. Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/2363>
* clutter: Pass events to pointer a11y before going through filtersJonas Dreßler2022-04-213-28/+42
| | | | | | | | | | | | | | We want all pointer events to be passed through the pointer a11y processing before going through event filters: Once we go through event filters, events might be dispatched to Wayland and get filtered out. With the changes to immediately dispatch events to wayland, this changed and the pointer a11y is now no longer seeing any events going to wayland clients. Fix it by shuffling things around a bit and letting pointer a11y take a peek at events earlier. Fixes: https://gitlab.gnome.org/GNOME/gnome-shell/-/issues/5192 Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/2332>
* clutter: Bail out and warn on reentry into mapping/unmapping cycleJonas Dreßler2022-04-212-0/+9
| | | | | | | | | | | | | | | | | | | | | There's a bunch of crashes right now where the assertions in clutter_actor_set_mapped() after calling the map/unmap() vfuncs are failing. The only way this can happen is by re-entering clutter_actor_set_mapped() during the map/unmap recursion. The reason for those crashes is that the shell hides/shows some actors in response to crossing events and key-focus changes. These in turn get triggered by the newly introduced ungrabbing of ClutterGrabs when an actor gets unmapped, which triggers GRAB_NOTIFY crossing events and key-focus changes. Since these situations are hardly avoidable (it's a valid use-case to hide/show something in response to a crossing/key-focus event), catch the set_mapped() call early while we reenter the mapping machinery and log a warning instead of crashing. Fixes: https://gitlab.gnome.org/GNOME/gnome-shell/-/issues/3165 Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/2299>
* clutter: Set event as current_event when going through event filtersJonas Dreßler2022-04-211-1/+11
| | | | | | | | | | | | | | | | | | | | | | | | | With the introduction of untrottled event delivery to wayland clients, we moved the _clutter_event_process_filters() call outside of _clutter_process_event(). This also moved the processing of event filters outside of the timespan where the event is added to Clutters current_event stack, making Clutter.get_current_event() no longer available to anything happening inside mutters event filter. One thing that happens in mutters event filter is detecting and triggering keybindings like the alt-tab switcher. Now the alt-tab switcher has a special case where it finishes and activates a window right when the keybinding gets activated, relying on the current event time as the timestamp to activate the window. Now since the current event time is no longer available from inside mutters event filter, we'd pass 0 to meta_window_activate(), causing mutter to send a notification instead of actually activating the window. To fix this, also set a current_event for the ClutterContext when going through event filters, this makes sure Clutter.get_current_event_time() works when called inside keybinding handlers. Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/2327>
* core: Account ClutterStage grabs on Wayland key focus synchronizationCarlos Garnacho2022-04-201-0/+2
| | | | | | | | | | | | | | | When switching between the existence and not of a stage ClutterGrab, we would correctly attempt to synchronize key focus from the perspective of the Wayland clients. But this synchronization should do its own checks about existing stage grabs before determining a client window has key focus or not. Add that check, so that grabs correctly unfocus the keyboard in Wayland clients, in addition to pointers and touch. Fixes: https://gitlab.gnome.org/GNOME/mutter/-/issues/2194 Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/2366>
* renderer/native: Restore EGL state after creating secondary GPU stateŁukasz Spintzyk2022-04-201-22/+18
| | | | Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/2347>
* cursor-renderer/native: Remove unused DRM_CAP_CURSOR_{WIDTH,HEIGHT}Daniel van Vugt2022-04-201-7/+0
| | | | Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/2361>
* crtc/kms: Don't add gamma to the update if unsupported by the CRTCDaniel van Vugt2022-04-191-1/+5
| | | | | Closes: https://gitlab.gnome.org/GNOME/mutter/-/issues/2197 Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/2360>
* kms/crtc: Add function meta_kms_crtc_has_gammaDaniel van Vugt2022-04-192-0/+9
| | | | Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/2360>
* kms/crtc: Add debug logging of supported vs unsupported propertiesDaniel van Vugt2022-04-191-0/+12
| | | | | | | Unsupported properties are particularly common in an Nvidia/hybrid setup so we don't want to make it a warning that's always visible. Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/2360>
* kms/crtc: Fix typo s/proporties/properties/Daniel van Vugt2022-04-191-2/+2
| | | | Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/2360>
* Update Vietnamese translationNgọc Quân Trần2022-04-161-79/+81
|
* startup-notification: Fix some introspection annotationsJonas Ådahl2022-04-132-10/+10
| | | | | | This avoids a bunch of warnings and fixes a transfer annotation. Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/2374>
* stage/x11: Remove trailing whitespaceDaniel van Vugt2022-04-131-5/+5
| | | | | | To avoid check-code-style warnings in other MRs that touch this file. Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/2373>
* kms/device: Disable modifiers when !DRM_CAP_ADDFB2_MODIFIERSDaniel van Vugt2022-04-081-0/+3
| | | | | Fixes: https://gitlab.gnome.org/GNOME/mutter/-/issues/2210 Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/2359>
* kms/impl-device: Add addfb2_modifiers to MetaKmsDeviceCapsDaniel van Vugt2022-04-082-0/+7
| | | | Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/2359>
* Update Catalan translationJordi Mas2022-04-071-4/+4
|
* shaped-texture: Paint with the right layer snippetJonas Ådahl2022-04-071-28/+98
| | | | | | | | | | | When we get passed a "snippet" to the shaped texture, it's added as a pipeline layer snippet to change how the source texture is sampled. When we draw from a texture tower however we have allocated regular textures which doesn't need any special layer snippet, so create separate pipelines for those that doesn't use that snippet. Closes: https://gitlab.gnome.org/GNOME/mutter/-/issues/528 Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/2278>
* shaped-texture: Pass along the snippet to the texture towerJonas Ådahl2022-04-073-0/+32
| | | | | | | | | The snippet is used make sure the right source is sampled in the shader. This wasn't done in the texture tower, meaning the textures from the tower were not correct. Related: https://gitlab.gnome.org/GNOME/mutter/-/issues/528 Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/2278>
* startup-notification: Specify the actual type in the "changed" signalCorentin Noël2022-04-051-1/+1
| | | | | | Allows to have type checks and the introspection to be aware of the actual type. Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/2362>
* mutter: Add missing nullable annotation when possibleCorentin Noël2022-04-054-20/+65
| | | | | | Add missing (nullable) annotation when a possible NULL return value is expected. Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/2362>
* cursor-tracker: Add missing annotation to meta_cursor_tracker_get_pointerCorentin Noël2022-04-051-0/+8
| | | | Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/2362>
* cogl/context: Add transfer full annotation to destroy timestamp queryCorentin Noël2022-04-051-0/+7
| | | | Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/2362>
* onscreen/native: Fix typo in log message "the the"Daniel van Vugt2022-04-041-1/+1
| | | | Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/2357>
* clutter: Keep actors dirty if a redraw was queued up during paint()Carlos Garnacho2022-03-311-2/+4
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | In the right combination of circumstances, and given 2 actors (parent actor P with an offscreen effect and child actor C), we may have the following situation happening: - A redraw is queued on the actor C, actors C and P are marked as priv->is_dirty and priv->propagated_one_redraw. - During paint() handling we paint actor P, priv->propagated_one_redraw is turned off. - We recurse into child actor C, priv->propagated_one_redraw is turned off. - A new redraw is queued on actor C, actors C and P are marked as priv->is_dirty and priv->propagated_one_redraw. - The paint() method recurses back, actors C and P get priv->is_dirty disabled, priv->propagated_one_redraw remains set. - At this point queueing up more redraws on actor C will not propagate up, because actor C has priv->propagated_one_redraw set, but the parent actor P has priv->is_dirty unset, so the offscreen effect will not get CLUTTER_EFFECT_PAINT_ACTOR_DIRTY and will avoid repainting actor C. The end result is that actor C does not redraw again, despite requesting redraws. This situation eventually resolves itself through e.g. relayouts on actor P, but may take some time to happen. In order to fix this, consider actors that did get a further redraw request still dirty after paint(). Fixes: https://gitlab.gnome.org/GNOME/mutter/-/issues/2188 Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/2353>
* clutter: Refactor code marking actors dirty for paint()Carlos Garnacho2022-03-311-9/+7
| | | | | | | | | | Simplify the function arguments (the origin is just the actor that the function is originally called from), and make it also handle marking as dirty the actor that got the redraw queued up explicitly. This makes it a single place where priv->is_dirty is being enabled. Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/2353>
* wayland/shm: Add support for ABGR8888 and XBGR8888 formatsJonas Ådahl2022-03-301-0/+14
| | | | | | | This avoids shuffling bits for clients drawing in these formats. Closes: https://gitlab.gnome.org/GNOME/mutter/-/issues/2200 Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/2352>
* x11: Set WM_S0 lastOlivier Fourdan2022-03-301-19/+23
| | | | | | | | | | | | | | | | | | | | | | With Xwayland on demand, a number of maintenance X11 applications need to be run first, before Xwayland starts accepting requests from the normal clients, as soon as the WM_S0 selection is acquired by mutter. On startup, mutter also sets a number of X11 properties that can be queried by X11 clients. Unfortunately, mutter acquires the WM_S0 selection before setting those properties, so mutter and the first regular X11 client will race on startup. As a result, the X11 properties set by mutter on startup may not be available to the very first X11 client when Xwayland starts. To avoid that issue, make sure to take the WM_S0 selection last when opening the display. Closes: https://gitlab.gnome.org/GNOME/mutter/-/issues/2176 Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/2336>
* xwayland: Init DnD on every Xwayland startOlivier Fourdan2022-03-301-3/+0
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | Currently, meta_xwayland_shutdown_dnd() is called from the handler on_x11_display_closing() triggered from the signal "x11-display-closing" hooked up from meta_xwayland_init_display(). Once the signal has been triggered, on_x11_display_closing() removes the signal handler, disconnecting from the signal. As meta_xwayland_init_display() is called from meta_display_new() which is issued only once, the signal handler is not restored again. As a result, meta_xwayland_shutdown_dnd() is not called anymore after Xwayland has been restarted, but meta_xwayland_init_dnd() will check and assert that the manager's DND object is NULL. Basically, restarting Xwayland more that once will trigger an assertion failure in mutter. That's even more of a problem with autoclose-xwayland where Xwayland is expected to terminate when there is no meaningful X11 client remaining, which can happen multiple times during the lifetime of a user session. To make sure that meta_xwayland_init_display() is called for every new instance of Xwayland, simply keep the signal hooked in place by not disconnecting it when triggered. This reverts commit 9a10b8ff94cc9506a4ecd8ade7489fa9a5af406a. Even though, originally, this issue was first introduced with commit b4fe1fdd95d52e8625c66ded1ec7162371dcf32b ("xwayland: Make setup/teardown a bit more symmetrical") which didn't actually kept 'x11-display-setup' and 'x11-display-closing' connected. Closes: https://gitlab.gnome.org/GNOME/mutter/-/issues/2168 Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/2339>
* window: Check fullscreen status when window changes monitorSebastian Wick2022-03-301-0/+2
| | | | | | | | | | | For wayland meta_window_move_to_monitor sends a configure to the client without actually moving the window, yet and the meta_display_queue_check_fullscreen call won't detect any changes. Checking for fullscreen in meta_window_update_monitor fixes the problem because it is called whenever the window actually changed the monitor it is on. Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/2325>
* onscreen/native: Fall back if COPY_MODE_SECONDARY_GPU fails to initDaniel van Vugt2022-03-301-6/+16
| | | | | | | | | | | | | | | | | | | | | | This can happen if we've got a GBM driver that is incomplete or buggy for any reason. Currently this includes the Nvidia driver when used as a secondary GPU, and as soon as they fix that it will use the fast path instead. Falling back means we can still use it as a KMS device with all rendering done on the primary GPU instead. Just like a DisplayLink. It's slow, but at least the screens all light up now and you get a usable desktop. Surprisingly it's still a better experience than using Xorg, maybe because we can flip the outputs independently with the native backend. Fixes: https://gitlab.gnome.org/GNOME/mutter/-/issues/2166, https://gitlab.gnome.org/GNOME/mutter/-/issues/2182, https://launchpad.net/bugs/1959888, https://launchpad.net/bugs/1964037 Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/2341>
* onscreen/native: Don't try rendering on a secondary GPU without EGLDaniel van Vugt2022-03-301-0/+4
| | | | | | | | | | We rather confusingly still call a secondary display card that is GPU-less (DisplayLink or other basic KMS device) a "secondary GPU", so just because secondary_gpu_state is non-NULL doesn't mean we can use it for rendering. The clearest indication of this is when there is no EGL surface. Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/2341>
* core: Port to input device capabilities in pad mapperCarlos Garnacho2022-03-301-7/+9
| | | | | | Check for tablets and pad devices using capabilities. Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/2331>
* backends: Update MetaInputSettings to apply settings based on capabilitiesCarlos Garnacho2022-03-301-116/+213
| | | | | | | | | | | | | | | | | Since devices may be multiple things now, check all capabilities in order to ensure all aspects of the device are correctly configured. This change does the following observations: - Devices that have TOUCHPAD | POINTER capabilities prefer the 'touchpad' settings path. The regular pointer settings path is left for all non-touchpads. - Devices that are both a tablet and a touchscreen prefer the tablet relocatable schema. This works for both aspects as the touchscreen schema is a subset of the tablet one. Other than that it's a rather boring, even if verbose search and replace. Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/2331>
* wayland: Check input device capabilities in tablet seatsCarlos Garnacho2022-03-302-21/+19
| | | | | | | Instead of looking for tablets and pads based on input device type, check capabilities. Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/2331>
* wayland: Set wayland seat capabilities based on input device capabilitiesCarlos Garnacho2022-03-301-39/+19
| | | | | | | Instead of looking at device types, which might be incomplete. Fixes: https://gitlab.gnome.org/GNOME/mutter/-/issues/2154 Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/2331>
View Lorry Controller Status