We already claim _NET_WM_DESKTOP support in _NET_SUPPORTED since around 2009,
but haven't actually done anything with it. However, especially pagers like
gnome-panel rely on this property to be updated and many tools, like GTK, want
to use the corresponding client messages to make a window sticky, move it
around etc.
This patch implements full support according to the EWMH spec. This means:
* We set the property on all windows when managing it.
* We keep the property updated on all windows at all times.
* We read and respect the property upon managing a window if it was set before
mapping the window.
* We react to client messages for it.
* We remove the property on withdrawn windows.
Note that the special value 0xFFFFFFFF, according to the spec, means that the
window shall be shown on all workspaces. We do this by making it sticky and
float it. This shows it on all workspaces at least on the output it is on.
Furthermore, the spec gives us the freedom to ignore _NET_WM_DESKTOP when
managing a window if we have good reason to. In our case, we give window
swallowing a higher priority since the user would likely expect that and we
want to keep placeholder windows only around for as long as we have to.
However, we do prioritize this property over, for example, startup
notifications.
fixes#2153fixes#1507fixes#938
pledges for i3:
"stdio rpath unix" for talking to the i3 socket usually in /tmp
"proc exec" for executing programs
"wpath cpath" are needed for the restart-in-place functionality
To make this work, @semarie pointed out that it is sufficient to ensure
that we get physical_mem_bytes only once, namely in init_logging().
pledges for i3-msg:
"stdio rpath unix" are needed for talking to the i3-socket
pledges for i3-nagbar
"rpath getpw" to find the home directory
"wpath cpath" to write the script
"proc exec" to execute the script
This disables the default leak-check-on-exit behavior which reports a
bunch of leaks that are only leaks while exiting, at which point they
don’t matter, because the operating system will clean up the memory our
process used.
The newly introduced decoration_border color incorrectly uses the default value
for "background" as a fallback when not specified. Instead, it should use the
user-specified "background" as a fallback.
fixes#2149
This patch moves the title_format information from windows to containers.
Furthermore, it allows correctly setting it on window-less containers and
displays the title accordingly for split containers.
We now also dump and read title_format in GET_TREE / during restarts.
fixes#2120
Corrects the cases where the colorpixel is not 0xRRGGBB : we have to
use the full color_t struct to describe font colors, as Pango expects
RGB values and not an XCB colorpixel value.
We now execute the validations when "kill" is executed even if match
criteria are used. This prevents users from killing workspace containers,
which instead kills all clients (as before when not using criteria).
fixes#1761
It should be clear for callers of this function that this is an internal
function that skips certain validations which might be important. Therefore
we make it clear that this is an internal function by renaming it.
relates to #1761
We need to move the urgent flag when moving a container across
workspaces before calling on_remove_child on the parent in order
to avoid a crash.
fixes#2128
This fixes race conditions, for example when i3bar gets reconfigured
after the available outputs change. In that specific case, i3bar sends a
ConfigureWindow request (see
b5693d6fb3/i3bar/src/xcb.c (L376))
which normally is turned into a ConfigureRequest that i3 largely
ignores, only the dock client’s height is considered (see
b5693d6fb3/src/handlers.c (L390)).
Turning ConfigureWindow into ConfigureRequest is only done when the
SubstructureRedirect event mask is set, and because we temporarily
removed _all_ events from our mask, the ConfigureWindow request went
through unmodified.
This in turn lead to the i3bar client window (not its decoration frame)
being positioned at e.g. y=1304, whereas dock client windows should
always end up at x=0 y=0 within their decoration frame. The result of
the i3bar client window being out of the visible space was either a
black i3bar or graphics corruption.
This also fixes issue #1904, I think. I couldn’t reproduce issue #1904
specifically, but when i3bar is in the misconfigured state, it will
receive a VisibilityNotify event, telling i3bar that it is obscured.
This would explain why i3bar sent a SIGSTOP in issue #1904.
fixes#1904
Previously, if a match specification contained the con_id or con_mark criterion,
all other criteria were ignored. However, a user may want to specify one of
those two unique identifiers and still specify others as well, for example to
match the currently focused window, but only if it has a certain WM_CLASS:
[con_id=__focused__ class=special] kill
We now check all specified criteria.
fixes#2111
This commit fixes#1969 by adding support for matching a window's type
against _NET_WM_WINDOW_TYPE_NOTIFICATION. The userguide and tests were
updated to reflect this change.
Previously, mouse bindings could only be run when a window was present,
by using --whole-window. Such bindings would not work on empty
workspaces. However, this is a valid usecase for bindings like
bindsym $mod+button4 workspace prev
bindsym $mod+button5 workspace next
Hence, we need to grab the root window as well and run bindings on it.
fixes#2097
Previously, using a command like
[con_id=foo] kill
would kill the currently focused window because while an error for
not being able to parse the con_id was logged, no further action
was taken, which caused the criterion to be ignored. In this case,
the fallback behavior of using the focused window took over.
For con_id, id and window_type we now reject incorrect values with
an error and abort the command.
fixes#2091
The values of a Rect are unsigned, but can contain signed values.
Using MAX when the value is negative causes incorrect behavior and
makes the result stay negative, which is what we wanted to avoid here
in the first place.
Fix by properly casting the value for the comparison.
fixes#2094
With this patch, we use 32-bit visuals per default whenever it is
available. Otherwise, we fall back to the actual root window's
depth, which will typically be 24-bit.
Before this patch, we already used 32-bit depth for containers with
a window that uses 32-bit. However, this means that we didn't use
32-bit for split parent containers on which decoration is drawn.
For 32-bit windows using transparency, this caused a graphical glitch
because the decoration pixmap behind it would show through. This
behavior is fixed with this change.
relates to #1278
Such containers never require a pixmap anyway. In particular for the __i3
output (used for the scratchpad workspace), this would cause an allocation
error anyway because it can be very big -- so big, in fact, that X cannot
allocate the pixmap for it.
Until now, this error was silently ignored due to the fact that we did not
create the pixmap checked (and asserted its success), but with cairo this
would cause a crash because we'd try to create a surface for a pixmap
which doesn't exist.
relates to #1278
With this patch we remove the temporary draw_util_colorpixel_to_color
function we introduced previously by parsing the colors as color_t to
begin with.
relates to #1278
We return early from drawing functions if the surface to draw to is not
initialized properly. There is no immediate need to do so, at least no
crashes have been observed, but it mirrors the previous behavior a bit
more closely. Furthermore, i3 should not crash due to not being able to
make some rendering call, so this provides some stability.
relates to #1278
This patch migrates all decoration rendering of i3 to cairo. Using the
compile switch CAIRO_SUPPORT, rendering can be switched back to the
previous XCB behavior, just like with the previous migration to cairo
in i3bar.
This patch also fixes a bug in draw_util.c where copying one surface
to another would use incorrect coordinates if the source coordinates
are not 0, 0.
Furthermore, this patch implicitly fixes some minor issues in the
decoration rendering which would be ignored previously due to the fact
that errors would only show up in the event queue, but not cause the
rendering code path to crash. One example is zero-height pixmaps which
are not allowed. Using cairo, these would cause i3 to instantly segfault,
so this patch avoids this.
Lastly, this patch annotates other issues found but not fixed in this patch
using TODO comments, e.g., the zero-height check not working correctly
and the comment that it should probably work the same way for zero-width
pixmaps.
relates to #1278
Mouse bindings that target the window that was clicked send the command
to the parser with `con_id` of the clicked window serialized base 16
for compatability with FreeBSD. See 7c2842e for explaination.
Set base to 0 for strtol to handle base 16 numbers for that reason.
This allows mouse bindings that target specific windows to work
correctly. Without this change, the focused window is always targetted
rather than the window that was actually clicked.
Regression introduced in b744c5e.
With this patch, we support the special value "0" for _NET_WM_USER_TIME
on a window upon managing it, which indicates that the window shall not
be focused.
fixes#2064
This reverts commit 9692c1498b.
That commit accidentally defaulted to group mask 1, but the default
should be to match any group mask, so that having multiple layouts
loaded at the same time works.
fixes#2062
With this patch, we only grab the scrollwheel buttons (4 and 5) when
managing a window if a whole window key binding exists for these buttons.
This allows both of these usecases:
- Bindings to scrollwheel buttons using --whole-window (see #1701).
- Scrolling in a window without focusing it if no such binding
exists (see #2049).
Furthermore, we drop all button grabs and regrab them after a config
reload in order to reevaluate the new bindings correctly.
fixes#2049
We refactor the button grabbing into a function to allow the next patch
both to
- conditionally grab different sets of buttons
- grab the buttons again when reloading the config.
relates to #2049
As there is no need to keep autostart commands in memory, we can safely
remove them as soon as they have been executed. As we previously didn't
clear them at all during config reloads, this also fixes a memory leak.
Note that neither autostarts nor autostarts_always is executed during
config reloads, so removing them from memory is fine as an i3 restart
will cause them to be parsed again.
fixes#2044
1. i3 config validation failed when the new line is missing from the end of file.
The error was: "ERROR: Your line continuation is too long, it exceeds 4096 bytes".
It is wrong to assume that there is always a '\n' at the end of each line in the config file.
(Not for the last line.)
Fix it via adding a end-of-file check.
2. See the issue #2051. (https://github.com/i3/i3/issues/2051)
This patch introduces the possibility to specify the tray_output directive
multiple times. All values will be used by i3bar, in the order they are
given.
This way, a single bar configuration can be used for several machines with
internal output names "eDP1" and "LVDS-0" by specifying tray_output for both.
Any external output (e.g., "DP-0") will still not receive the tray. The same
effect can be achieved by using "primary", but forces the user to couple the
tray display to the primary output which may not be desirable behavior.
relates to #555
This patch extends the previously introduced flag --no-auto-back-and-forth
to also apply to
move window to workspace <name>
move window to workspace number <number>
relates to #2028
Since splitting a docking container was allowed and successful, the check
to prevent floating it fails to work. This causes a crash because the
workspace of the container cannot be determined as the dockarea is higher
up in the tree than the workspace it belongs to.
This patch extends to sanity check to nested dock containers when trying to
float a container and also disallows manually splitting a docked container
or changing its layout.
fixes#2034
We need to store the information that an assignment was run for a window
before actually executing the command. Otherwise, if the command causes
a change that causes assignments to be run again, the window might be
matched again, causing an infinite loop and hence i3 to freeze or crash.
Assignments don't necessarily represent workspace assignments, but could
also be used, e.g., for no_focus. Hence, there's no point in logging
dest.workspace for all assignments.
Suppose a horizontally oriented workspace has two windows open. Issuing "split h" on
the right window creates a nested horizontal container. If we now resize in "width"
direction, the resize doesn't work because it will only operate on the current parent,
which is a container with only one child, so the resize command exits.
This is unexpected behavior from a user point of view. Hence, with this patch, we
ensure that we don't just go up the tree until we find a parent with the correct
orientation, but also keep going if that parent has only a single child.
fixes#2015
This patch introduces the --no-auto-back-and-forth flag to both of
workspace --no-auto-back-and-forth <name>
workspace --no-auto-back-and-forth number <number>
This flag will only have an effect if the back_and_forth feature is
enabled. If passed, the feature will be ignored for this particular
call only.
fixes#2028
This allows matching with
[con_id=__focused__] unmark
for commands that do not default to operating on the focused window
if no criteria have been specified (such as unmark).
relates to #2014
This patch allows using 'unmark' on matched windows. The old behavior
of applying it to all windows if no criteria were specified is kept.
relates to #2014
In order to keep compatibility to before allowing multiple marks on a window,
we introduce a flag "--add" that must be set to put more than one mark on a
window. The default, which is also available as "--replace", keeps the old
behavior of overwriting a mark when setting a new one.
fixes#2014
This patch allows multiple marks to be set on a single window. The restriction that a mark may
only be on one window at a time is still upheld as this is necessary for commands like
"move window to mark" to make sense.
relates to #2014
When a window is moved to a mark and the marked container is a workspace,
we can skip any other logic and just call con_move_to_workspace directly.
fixes#2003
See the issue #1798 (http://github.com/i3/i3/issues/1798).
+workspace_next+ as-is cycles through either numbered or named workspaces,
but when it reaches the last numbered/named workspace, it only looks for
named workspaces. This commit changes it: look for named workspaces after
exhausting numbered ones, but also for numbered ones after exhausting
named ones.
Also add a test case 528-workspace-next-prev.t (numbered workspaces and named
workspaces on 2 outputs) for testing this.
This introduces the flag "--pango" on the mode config directive to
explicitly enable pango markup for mode names. Not setting this will
cause the mode name to be rendered as is.
This fixes a regression in 4.11 where mode names containing characters
such as '<' would break user's configs as they didn't escape these
characters.
fixes#1992
With this patch, the no_focus directive will be ignored if the
to-be-opened window is the first on its workspace as there's no
reason the user would not want to focus it in this case.
This improves usability when, for example, using a tabbed
workspace_layout.
fixes#1987
This patch creates all necessary windows for i3bar with 32-bit visuals if available.
It also introduces the possibility to define RGBA colors (next to RGB colors), which
allows the user to set the opacity of any color. This requires running a compositor.
With this patch we also start supporting _NET_SYSTEM_TRAY_VISUAL, which is necessary
for the tray icons so they create the tray window with the correct depth and visual.
1). See the issue #1926. For example, the second keybinding is not detected as a duplicate:
bindcode Mod4+24 sticky toggle
bindsym Mod4+q focus parent
2). To fix it, check duplicated bindings when translating the keysym to keycodes.
`vlog()` can not handle log messages longer than 4096 bytes. However, the
message generated in `store_restart_layout()` is likely to exceed this
as it contains a long JSON string.
This has caused a few SEGFAULTS during restarts for me when running with
`-d all`.
Fix this by truncating the message to 4096 bytes and punching in a newline at
the end.
This fixes a bug I introduced in #1921. When restarting i3 in place a
stray workspace was created on the root_output during restart. On first
start, this workspace would have been moved to the first real and empty
output.
However, this does not produce the desired result during restarts when
workspaces are alread present on all real outputs. The stray workspace would
still be added to the first real output which already contains some
workspaces. Thus, adding a new empty workspace to it.
Fix this by delaying creation of the root output's workspace until it is
known whether the output is active or not.
Fixes#1940
This is necessary during a restart of i3 when restoring floating windows. In this situation, we restore the layout before setting up the RandR outputs which would set the window's size to 0, making it invisible.
Thanks to hwangcc23 and spudowiar for reporting.
fixes#1910fixes#1934
1. Reference: issue #1929
2. When restarting, add the argument "-d all" if debuglog is on.
3. Add add_argument() for adding/replacing the given argument.
This patch introduces a root output covering the root window. It is used
in two cases:
1. RandR is not available. In this case, the previous behaviour of
creating a single output covering the root window is preserved.
2. RandR is available, but there is no active output. In this case,
the root output is enabled and will be the only active output.
If any RandR output becomes available, the root output will be
disabled again. Existing mechanisms for migrating workspaces will
just work without modification.
I've carefully slipped in a global variable `Output root_output` representing
that output.
Fixes#926 and #1489
This commit also reworks the way focusing sticky windows is prevented by not focusing them temporarily at all, but preventing the focus in the first place.
If no other window is available on the active workspace, we now select the EWMH support window (used to indicate that an EWMH-compliant window manager is preent) as the focus window rather than the root window. The NET_WM_ACTIVE window will still be set to XCB_WINDOW_NONE to pretend that no window is actually focused.
This fixes the issue that when using the root window, a fallback mechanism in X11 takes effect which routes keyboard input to the window under the cursor, independent of whether that window has the input focus. Using the EWMH window instead, we can avoid this behavior. We cannot simply set it to XCB_WINDOW_NONE as this would discard all keyboard events, breaking keybindings.
fixes#1378
Reordering once (as we did it before this commit) would only sort the
bindings by the _first_ bit of their event_state_mask, but we need to
sort them by _all_ bits of their event_state_mask.
fixes#1870
fixes#1835
This commit improves the translation of keysyms to keycodes by loading
keymaps using libxkbcommon-x11 and using libxkbcommon for figuring out
the keymap, depending on each keybinding’s modifiers. This way, the
upper layers of complex layouts are now usable with i3’s bindsym
directive, such as de_neo’s layer 3 and higher.
Furthermore, the commit generalizes the handling of different XKB
groups. We formerly had support only for two separate groups, the
default group 1, and group 2. While Mode_switch is only one way to
switch to group 2, we called the binding option Mode_switch. With this
commit, the new names Group1, Group2 (an alias for Mode_switch), Group3
and Group4 are introduced for configuring bindings. This is only useful
for advanced keyboard layouts, such as people loading two keyboard
layouts and switching between them (us, ru seems to be a popular
combination).
When grabbing keys, one can only specify the modifier mask, but not an
XKB state mask (or value), so we still dynamically unbind and re-bind
keys whenever the XKB group changes.
The commit was manually tested using the following i3 config:
bindsym Group4+n nop heya from group 4
bindsym Group3+n nop heya from group 3
bindsym Group2+n nop heya from group 2
bindsym n nop heya
bindsym shift+N nop explicit shift binding
bindsym shift+r nop implicit shift binding
bindcode Group2+38 nop fallback overwritten in group 2 only
bindcode 38 nop fallback
…with the following layout:
setxkbmap -layout "us,ua,ru,de" -variant ",winkeys,,neo" \
-option "grp:shift_caps_toggle,grp_led:scroll" \
-model pc104 -rules evdev
By default (xkb group 1, us layout), pressing “n” will result in the
“heya” message appearing. Pressing “a” will result in the “fallback”
message appearing. “j” is not triggered.
By pressing Shift+CapsLock you switch to the next group (xkb group 2, ua
layout). Pressing “a” will result in the “fallback overwritten in group
2 only” message, pressing “n” will still result in “heya”. “j” is not
triggered.
In the next group (xkb group 3, ru layout), pressing “a” will result in
the “fallback” message again, pressing “n” will result in “heya”,
“j” is not triggered.
In the last group (xkb group 4, de_neo layout), pressing “a” will still
result in “fallback”, pressing “n” will result in “heya”, pressing “j”
will result in “heya from group 4”.
Pressing shift+n results in “explicit shift binding”, pressing shift+r
results in “implicit shift binding”. This ensures that keysym
translation falls back to looking at non-shift keys (“r” can be used
instead of ”R”) and that the order of keybindings doesn’t play a role
(“bindsym n” does not override “bindsym shift+n”, even though it’s
specified earlier in the config).
The fallback behavior ensures use-cases such as ticket #1775 are still
covered.
Only binding keys when the X server is in the corresponding XKB group
ensures use-cases such as ticket #585 are still covered.
This reverts commit e71c304444.
It turns out that several users have workflows in which they turn off
their monitors without using e.g. `xrandr --output DP-1 --off`. The
result is that the monitors are disconnected, but not disabled.
With commit e71c304444, i3 started to see
these two states as one and the same state, but that causes more harm
than it does good. For example, for some users with only one monitor, i3
would just exit when these users turned off their monitor.
related to #1858, #1839fixes#1845
With commit c738b2e454 we changed i3 so
that the default keybindings can be used when ISO_Next_Group is enabled,
but bindings which explicitly use Mode_switch have precedence. This
behavior required the use of bindcode instead of bindsym.
With this commit, when switching from group 1 to group 2 using
ISO_Next_Group, i3 will re-translate all keybindings (looking at column
2/3, regardless of whether the keybinding itself specifies Mode_switch)
and re-grab them.
That way, the keybinding “bindsym $mod+x nop foo” will work when
pressing $mod+x without Mode_switch and when pressing the corresponding
$mod+x (different key) with Mode_switch. A binding such as “bindsym
Mode_switch+$mod+x nop bar” will still have precedence.
The intention here is to make bindsym keybindings work well with dual
keyboard layouts (such as {dvorak, us} or {us, ru}), so that users can
switch between groups and still have their (logical) keybindings behave
the same way.
fixes#1775
Do not log an XCB_MOTION_NOTIFY event in handlers.c because this
clutters the log.
Only log in the motion notify handler if something important happens.
* common.mk: use -lsocket -liconv -lgen on Illumos/Solaris
* mkdirp: return int and accept a mode argument
* use i3's mkdirp on everything except Illumos
1. Allow to use the line continuation, which is indicated by \ before the new line character, in config files.
2. Add a new testcase "247-config-line-continuation.t" for
a). testing line continuation
b). making sure string escaping still works
c). testing line continuations within a string
b). testing line continuations with too many lines
If the match expression is a plain number (e.g., '99'), the number of a workspace will be compared strictly. Otherwise, the match expression is taken as a regular expression and compared against the workspace's name.
This allows all of the following:
for_window [workspace=5] ...
for_window [workspace="5:foo"] ...
for_window [workspace="foo"] ...
fixes#1769
Fix TODO in bindings.c.
There is no problem to use atoi here since either keycode 0 or button0 is invalid.
But strtol is more flexible and is recommanded for conversion.
The format string set with "title_format" can contain the placeholder "%title" which will be replaced with the actual window title.
By not overwriting window->name itself, we make sure that assignment matching still works as expected.
fixes#1723
Ingo Bürk