Johan Malm [Fri, 7 Jan 2022 20:53:48 +0000 (20:53 +0000)]
view: fix bug in view_snap_to_edge()
Use view_move() and view_move_resize() correctly.
view_move_resize() should only be used when the view actually changes
width and/or height, otherwise the serials might cause a delay in moving
xdg-shell clients.
Johan Malm [Thu, 6 Jan 2022 21:40:10 +0000 (21:40 +0000)]
seat: support WLR_{WL,X11}_OUTPUTS >= 2
When running nested in X11 or a wlroots compositor with the respective
environment variables WLR_X11_OUTPUTS or WLR_WL_OUTPUTS set to value >= 2,
cursors need to be mapped to the respective outputs.
Johan Malm [Mon, 3 Jan 2022 23:32:02 +0000 (23:32 +0000)]
docs/autostart: remove wlr-randr {--off,--on}
swayidle previously used wlr-randr to enable/disable outputs, but since
commit a837fefc, this re-arranges views so is not to be used in the
context of idle system power management.
When using the keybind above (in rc.xml), on the first execution of W-d
all is okay, but the second time, a "d" pressed event is sent to dmenu
resulting in a continuous "ddddddd...") which has to be stopped pressing a
key.
This behaviour started in commit 7e57b7f because release events associated
with keybinds are no longer sent to clients (before that commit, the
release event for the ādā would have been passed to dmenu, thus cancelling
the repeat).
John Lindgren [Fri, 31 Dec 2021 22:30:55 +0000 (17:30 -0500)]
Adjust views to account for output layout changes
labwc currently doesn't handle output layout changes very well:
- Windows can end up "lost" completely offscreen
- Maximized/fullscreen windows can end up spanning multiple outputs
Currently, new_output_notify() and output_destroy_notify() contain logic
to update the cursor and force a repaint when outputs are added or
removed. This logic in fact needs to run on any output layout change,
so consolidate it into a new function, output_update_for_layout_change().
Then add a second new function, view_adjust_for_layout_change(), which
adjusts window placement to account for the new layout.
The behavior is roughly as follows:
- Normal windows that end up offscreen are centered on the closest
output (making use of the existing view_center() logic)
- Maximized windows are re-maximized on the closest output. Logic is
also added to the unmaximize step to check that the original
unmaximized position is still on-screen.
- Fullscreen windows are re-fullscreened on the same output if
possible; otherwise they are un-fullscreened.
Minimized windows don't require any special handling. Their placement
is adjusted just the same, but invisible to the user until they are
later unminimized.
There is some positioning glitch still with un-fullscreening a window
whose output has been disconnected/disabled; it can end up in an
unexpected position (but at least has the correct size and decoration).
I don't think this is due to a bug in my change per se, but perhaps the
change has exposed a bug elsewhere.
Johan Malm [Tue, 28 Dec 2021 15:06:33 +0000 (15:06 +0000)]
Fix build on Alpine Linux
Issue #144
Add `#define _POSIX_C_SOURCE 200809L` to avoid the error below:
../src/cursor.c: In function 'cursor_update_focus':
../src/cursor.c:271:2: warning: implicit declaration of function
'clock_gettime' [-Wimplicit-function-declaration]
271 | clock_gettime(CLOCK_MONOTONIC, &now);
| ^~~~~~~~~~~~~
../src/cursor.c:271:16: error: 'CLOCK_MONOTONIC' undeclared (first use in
this function)
271 | clock_gettime(CLOCK_MONOTONIC, &now);
| ^~~~~~~~~~~~~~~
../src/cursor.c:271:16: note: each undeclared identifier is reported only
once for each function it appears in
../src/cursor.c: In function 'is_double_click':
../src/cursor.c:486:16: error: 'CLOCK_MONOTONIC' undeclared (first use in
this function)
486 | clock_gettime(CLOCK_MONOTONIC, &now);
| ^~~~~~~~~~~~~~~
Consolatis [Mon, 27 Dec 2021 22:41:33 +0000 (23:41 +0100)]
view.c: Remove 'activated' flag when minimizing a view
Before this change a window that had been minimized showed up with states Minimized and Activated.
For foreign-toplevel clients like taskbars that could mean to handle multiple windows at once
having a 'activated' state even though they were clearly minimized.
John Lindgren [Mon, 27 Dec 2021 14:44:03 +0000 (09:44 -0500)]
Fix use-after-free errors detected by valgrind
Handlers for the "destroy" signal need to unregister themselves from the
signal (using wl_list_remove()) before invoking free().
Invalid write of size 8
at 0x487DF27: wl_list_remove (in /usr/lib/libwayland-server.so.0.20.0)
by 0x4906FF0: wlr_signal_emit_safe (signal.c:32)
by 0x48EA304: reset_xdg_surface (wlr_xdg_surface.c:430)
by 0x48EA3D6: UnknownInlinedFun (wlr_xdg_surface.c:464)
by 0x48EA3D6: xdg_surface_handle_resource_destroy (wlr_xdg_surface.c:282)
by 0x4877899: ??? (in /usr/lib/libwayland-server.so.0.20.0)
by 0x487DEBF: ??? (in /usr/lib/libwayland-server.so.0.20.0)
by 0x487E3DF: ??? (in /usr/lib/libwayland-server.so.0.20.0)
by 0x487830D: wl_client_destroy (in /usr/lib/libwayland-server.so.0.20.0)
by 0x48783F6: ??? (in /usr/lib/libwayland-server.so.0.20.0)
by 0x487B1C9: wl_event_loop_dispatch (in /usr/lib/libwayland-server.so.0.20.0)
by 0x4878D36: wl_display_run (in /usr/lib/libwayland-server.so.0.20.0)
by 0x110DAC: main (main.c:81)
Address 0xc3213d8 is 424 bytes inside a block of size 704 free'd
at 0x484118B: free (vg_replace_malloc.c:755)
by 0x4906FDD: wlr_signal_emit_safe (signal.c:29)
by 0x48EA304: reset_xdg_surface (wlr_xdg_surface.c:430)
by 0x48EA3D6: UnknownInlinedFun (wlr_xdg_surface.c:464)
by 0x48EA3D6: xdg_surface_handle_resource_destroy (wlr_xdg_surface.c:282)
by 0x4877899: ??? (in /usr/lib/libwayland-server.so.0.20.0)
by 0x487DEBF: ??? (in /usr/lib/libwayland-server.so.0.20.0)
by 0x487E3DF: ??? (in /usr/lib/libwayland-server.so.0.20.0)
by 0x487830D: wl_client_destroy (in /usr/lib/libwayland-server.so.0.20.0)
by 0x48783F6: ??? (in /usr/lib/libwayland-server.so.0.20.0)
by 0x487B1C9: wl_event_loop_dispatch (in /usr/lib/libwayland-server.so.0.20.0)
by 0x4878D36: wl_display_run (in /usr/lib/libwayland-server.so.0.20.0)
by 0x110DAC: main (main.c:81)
Block was alloc'd at
at 0x48435FF: calloc (vg_replace_malloc.c:1117)
by 0x113E22: xdg_surface_new (xdg.c:401)
by 0x4906FDD: wlr_signal_emit_safe (signal.c:29)
by 0x48EABE7: handle_xdg_surface_commit (wlr_xdg_surface.c:331)
by 0x4908FE7: surface_commit_state (wlr_surface.c:457)
by 0x524DD49: ??? (in /usr/lib/libffi.so.8.1.0)
by 0x524D266: ??? (in /usr/lib/libffi.so.8.1.0)
by 0x487D322: ??? (in /usr/lib/libwayland-server.so.0.20.0)
by 0x48785CB: ??? (in /usr/lib/libwayland-server.so.0.20.0)
by 0x487B1C9: wl_event_loop_dispatch (in /usr/lib/libwayland-server.so.0.20.0)
by 0x4878D36: wl_display_run (in /usr/lib/libwayland-server.so.0.20.0)
by 0x110DAC: main (main.c:81)
Invalid write of size 8
at 0x487DF27: wl_list_remove (in /usr/lib/libwayland-server.so.0.20.0)
by 0x4906FF0: wlr_signal_emit_safe (signal.c:32)
by 0x48F5529: UnknownInlinedFun (wlr_input_device.c:34)
by 0x48F5529: wlr_input_device_destroy (wlr_input_device.c:29)
by 0x48CE4B5: backend_destroy.part.0.lto_priv.0 (backend.c:148)
by 0x48D2D55: multi_backend_destroy (backend.c:59)
by 0x48788FE: wl_display_destroy (in /usr/lib/libwayland-server.so.0.20.0)
by 0x110E71: UnknownInlinedFun (server.c:406)
by 0x110E71: main (main.c:83)
Address 0xc466568 is 24 bytes inside a block of size 56 free'd
at 0x484118B: free (vg_replace_malloc.c:755)
by 0x4906FDD: wlr_signal_emit_safe (signal.c:29)
by 0x48F5529: UnknownInlinedFun (wlr_input_device.c:34)
by 0x48F5529: wlr_input_device_destroy (wlr_input_device.c:29)
by 0x48CE4B5: backend_destroy.part.0.lto_priv.0 (backend.c:148)
by 0x48D2D55: multi_backend_destroy (backend.c:59)
by 0x48788FE: wl_display_destroy (in /usr/lib/libwayland-server.so.0.20.0)
by 0x110E71: UnknownInlinedFun (server.c:406)
by 0x110E71: main (main.c:83)
Block was alloc'd at
at 0x48435FF: calloc (vg_replace_malloc.c:1117)
by 0x11590F: new_input_notify (seat.c:137)
by 0x4906FDD: wlr_signal_emit_safe (signal.c:29)
by 0x4906FDD: wlr_signal_emit_safe (signal.c:29)
by 0x48CDCEE: UnknownInlinedFun (events.c:109)
by 0x48CDCEE: UnknownInlinedFun (events.c:242)
by 0x48CDCEE: UnknownInlinedFun (backend.c:58)
by 0x48CDCEE: handle_libinput_readable (backend.c:48)
by 0x48CE1BE: backend_start.lto_priv.0 (backend.c:114)
by 0x48D2CE5: multi_backend_start (backend.c:33)
by 0x110C45: UnknownInlinedFun (server.c:375)
by 0x110C45: main (main.c:68)
Consolatis [Thu, 23 Dec 2021 04:37:57 +0000 (05:37 +0100)]
config/rcxml: Allow multiple <action>s inside of a <mousebind>
Issue arises when using the default config from docs/rc.xml.all.
Without this patch only the last action defined inside a <mousebind>
will have an effect.
Without a config or when defining the same <mousebind> multiple times
with each containing only a single <action> the issue does not exist.
John Lindgren [Sat, 4 Dec 2021 15:06:21 +0000 (10:06 -0500)]
action: Switch to next window when pressing Alt-Tab once
In OpenBox (as in most other stacking window managers), pressing
Alt-Tab once will cycle to the next window. This is especially
convenient in workflows which require switching back and forth
frequently between the two top windows.
bi4k8 [Wed, 1 Dec 2021 02:44:55 +0000 (02:44 +0000)]
forward click events to layer surfaces
this was broken when we moved to swallowing mouse events that triggered "Frame"-context mouse bindings. layer surfaces don't run mouse binding logic, so they need mouse events unconditionally forwarded.
bi4k8 [Thu, 2 Dec 2021 14:07:17 +0000 (14:07 +0000)]
fix interaction between Move action and click bindings
in order to allow the Move action to be bound to "press" on mouse binding contexts that also have "click" bindings, we should not short-circuit event processing when the input_mode is not passthrough (the "normal" mode). doing so seems to have been intended to prevent mouse bindings from being triggered during move/resize/menu interactions, but this does not seem to occur in practice. instead, `cursor_button`'s early return in this case caused the "release" side of "click" bindings to be ignored if their "press" side began an action that changes input_mode (e.g. Move).
the cleaner way to fix this interaction would be to use "drag" rather than "press" for Move bindings, but implementing "drag" is more complexity than I want to include in this changeset.
this change is its own commit to make it easy to bisect for, in case it causes regressions.