gri3-wm/docs/ipc

IPC interface (interprocess communication)
==========================================
Michael Stapelberg <michael@i3wm.org>
October 2014

This document describes how to interface with i3 from a separate process. This
is useful for example to remote-control i3 (to write test cases for example) or
to get various information like the current workspaces to implement an external
workspace bar.

The method of choice for IPC in our case is a unix socket because it has very
little overhead on both sides and is usually available without headaches in
most languages. In the default configuration file, the ipc-socket gets created
in +/tmp/i3-%u.XXXXXX/ipc-socket.%p+ where +%u+ is your UNIX username, +%p+ is
the PID of i3 and XXXXXX is a string of random characters from the portable
filename character set (see mkdtemp(3)). You can get the socketpath from i3 by
calling +i3 --get-socketpath+.

All i3 utilities, like +i3-msg+ and +i3-input+ will read the +I3_SOCKET_PATH+
X11 property, stored on the X11 root window.

[WARNING]
.Use an existing library!
There are existing libraries for many languages. You can have a look at
<<libraries>> or search the web if your language of choice is not mentioned.
Usually, it is not necessary to implement low-level communication with i3
directly.

== Establishing a connection

To establish a connection, simply open the IPC socket. The following code
snippet illustrates this in Perl:

-------------------------------------------------------------
use IO::Socket::UNIX;
chomp(my $path = qx(i3 --get-socketpath));
my $sock = IO::Socket::UNIX->new(Peer => $path);
-------------------------------------------------------------

== Sending messages to i3

To send a message to i3, you have to format in the binary message format which
i3 expects. This format specifies a magic string in the beginning to ensure
the integrity of messages (to prevent follow-up errors). Following the magic
string comes the length of the payload of the message as 32-bit integer, and
the type of the message as 32-bit integer (the integers are not converted, so
they are in native byte order).

The magic string currently is "i3-ipc" and will only be changed when a change
in the IPC API is done which breaks compatibility (we hope that we don’t need
to do that).

Currently implemented message types are the following:

COMMAND (0)::
	The payload of the message is a command for i3 (like the commands you
	can bind to keys in the configuration file) and will be executed
	directly after receiving it.
GET_WORKSPACES (1)::
	Gets the current workspaces. The reply will be a JSON-encoded list of
	workspaces (see the reply section).
SUBSCRIBE (2)::
	Subscribes your connection to certain events. See <<events>> for a
	description of this message and the concept of events.
GET_OUTPUTS (3)::
	Gets the current outputs. The reply will be a JSON-encoded list of outputs
	(see the reply section).
GET_TREE (4)::
	Gets the layout tree. i3 uses a tree as data structure which includes
	every container. The reply will be the JSON-encoded tree (see the reply
	section).
GET_MARKS (5)::
	Gets a list of marks (identifiers for containers to easily jump to them
	later). The reply will be a JSON-encoded list of window marks (see
	reply section).
GET_BAR_CONFIG (6)::
	Gets the configuration (as JSON map) of the workspace bar with the
	given ID. If no ID is provided, an array with all configured bar IDs is
	returned instead.
GET_VERSION (7)::
	Gets the version of i3. The reply will be a JSON-encoded dictionary
	with the major, minor, patch and human-readable version.

So, a typical message could look like this:
--------------------------------------------------
"i3-ipc" <message length> <message type> <payload>
--------------------------------------------------

Or, as a hexdump:
------------------------------------------------------------------------------
00000000  69 33 2d 69 70 63 04 00  00 00 00 00 00 00 65 78  |i3-ipc........ex|
00000010  69 74                                             |it|
------------------------------------------------------------------------------

To generate and send such a message, you could use the following code in Perl:
------------------------------------------------------------
sub format_ipc_command {
    my ($msg) = @_;
    my $len;
    # Get the real byte count (vs. amount of characters)
    { use bytes; $len = length($msg); }
    return "i3-ipc" . pack("LL", $len, 0) . $msg;
}

$sock->write(format_ipc_command("exit"));
------------------------------------------------------------------------------

== Receiving replies from i3

Replies from i3 usually consist of a simple string (the length of the string
is the message_length, so you can consider them length-prefixed) which in turn
contain the JSON serialization of a data structure. For example, the
GET_WORKSPACES message returns an array of workspaces (each workspace is a map
with certain attributes).

=== Reply format

The reply format is identical to the normal message format. There also is
the magic string, then the message length, then the message type and the
payload.

The following reply types are implemented:

COMMAND (0)::
	Confirmation/Error code for the COMMAND message.
WORKSPACES (1)::
	Reply to the GET_WORKSPACES message.
SUBSCRIBE (2)::
	Confirmation/Error code for the SUBSCRIBE message.
OUTPUTS (3)::
	Reply to the GET_OUTPUTS message.
TREE (4)::
	Reply to the GET_TREE message.
MARKS (5)::
	Reply to the GET_MARKS message.
BAR_CONFIG (6)::
	Reply to the GET_BAR_CONFIG message.
VERSION (7)::
	Reply to the GET_VERSION message.

=== COMMAND reply

The reply consists of a list of serialized maps for each command that was
parsed. Each has the property +success (bool)+ and may also include a
human-readable error message in the property +error (string)+.

*Example:*
-------------------
[{ "success": true }]
-------------------

=== WORKSPACES reply

The reply consists of a serialized list of workspaces. Each workspace has the
following properties:

num (integer)::
	The logical number of the workspace. Corresponds to the command
	to switch to this workspace. For named workspaces, this will be -1.
name (string)::
	The name of this workspace (by default num+1), as changed by the
	user. Encoded in UTF-8.
visible (boolean)::
	Whether this workspace is currently visible on an output (multiple
	workspaces can be visible at the same time).
focused (boolean)::
	Whether this workspace currently has the focus (only one workspace
	can have the focus at the same time).
urgent (boolean)::
	Whether a window on this workspace has the "urgent" flag set.
rect (map)::
	The rectangle of this workspace (equals the rect of the output it
	is on), consists of x, y, width, height.
output (string)::
	The video output this workspace is on (LVDS1, VGA1, …).

*Example:*
-------------------
[
 {
  "num": 0,
  "name": "1",
  "visible": true,
  "focused": true,
  "urgent": false,
  "rect": {
   "x": 0,
   "y": 0,
   "width": 1280,
   "height": 800
  },
  "output": "LVDS1"
 },
 {
  "num": 1,
  "name": "2",
  "visible": false,
  "focused": false,
  "urgent": false,
  "rect": {
   "x": 0,
   "y": 0,
   "width": 1280,
   "height": 800
  },
  "output": "LVDS1"
 }
]
-------------------

=== SUBSCRIBE reply

The reply consists of a single serialized map. The only property is
+success (bool)+, indicating whether the subscription was successful (the
default) or whether a JSON parse error occurred.

*Example:*
-------------------
{ "success": true }
-------------------

=== OUTPUTS reply

The reply consists of a serialized list of outputs. Each output has the
following properties:

name (string)::
	The name of this output (as seen in +xrandr(1)+). Encoded in UTF-8.
active (boolean)::
	Whether this output is currently active (has a valid mode).
current_workspace (string)::
	The name of the current workspace that is visible on this output. +null+ if
	the output is not active.
rect (map)::
	The rectangle of this output (equals the rect of the output it
	is on), consists of x, y, width, height.

*Example:*
-------------------
[
 {
  "name": "LVDS1",
  "active": true,
  "current_workspace": "4",
  "rect": {
   "x": 0,
   "y": 0,
   "width": 1280,
   "height": 800
  }
 },
 {
  "name": "VGA1",
  "active": true,
  "current_workspace": "1",
  "rect": {
   "x": 1280,
   "y": 0,
   "width": 1280,
   "height": 1024
  },
 }
]
-------------------

=== TREE reply

The reply consists of a serialized tree. Each node in the tree (representing
one container) has at least the properties listed below. While the nodes might
have more properties, please do not use any properties which are not documented
here. They are not yet finalized and will probably change!

id (integer)::
	The internal ID (actually a C pointer value) of this container. Do not
	make any assumptions about it. You can use it to (re-)identify and
	address containers when talking to i3.
name (string)::
	The internal name of this container. For all containers which are part
	of the tree structure down to the workspace contents, this is set to a
	nice human-readable name of the container.
	For containers that have an X11 window, the content is the title
	(_NET_WM_NAME property) of that window.
	For all other containers, the content is not defined (yet).
type (string)::
	Type of this container. Can be one of "root", "output", "con",
	"floating_con", "workspace" or "dockarea".
border (string)::
	Can be either "normal", "none" or "1pixel", dependending on the
	container’s border style.
current_border_width (integer)::
	Number of pixels of the border width.
layout (string)::
	Can be either "splith", "splitv", "stacked", "tabbed", "dockarea" or
	"output".
	Other values might be possible in the future, should we add new
	layouts.
orientation (string)::
	Can be either "none" (for non-split containers), "horizontal" or
	"vertical".
	THIS FIELD IS OBSOLETE. It is still present, but your code should not
	use it. Instead, rely on the layout field.
percent (float)::
	The percentage which this container takes in its parent. A value of
	+null+ means that the percent property does not make sense for this
	container, for example for the root container.
rect (map)::
	The absolute display coordinates for this container. Display
	coordinates means that when you have two 1600x1200 monitors on a single
	X11 Display (the standard way), the coordinates of the first window on
	the second monitor are +{ "x": 1600, "y": 0, "width": 1600, "height":
	1200 }+.
window_rect (map)::
	The coordinates of the *actual client window* inside its container.
	These coordinates are relative to the container and do not include the
	window decoration (which is actually rendered on the parent container).
	So, when using the +default+ layout, you will have a 2 pixel border on
	each side, making the window_rect +{ "x": 2, "y": 0, "width": 632,
	"height": 366 }+ (for example).
deco_rect (map)::
	The coordinates of the *window decoration* inside its container. These
	coordinates are relative to the container and do not include the actual
	client window.
geometry (map)::
	The original geometry the window specified when i3 mapped it. Used when
	switching a window to floating mode, for example.
window (integer)::
	The X11 window ID of the *actual client window* inside this container.
	This field is set to null for split containers or otherwise empty
	containers. This ID corresponds to what xwininfo(1) and other
	X11-related tools display (usually in hex).
urgent (bool)::
	Whether this container (window or workspace) has the urgency hint set.
focused (bool)::
	Whether this container is currently focused.

Please note that in the following example, I have left out some keys/values
which are not relevant for the type of the node. Otherwise, the example would
be by far too long (it already is quite long, despite showing only 1 window and
one dock window).

It is useful to have an overview of the structure before taking a look at the
JSON dump:

* root
** LVDS1
*** topdock
*** content
**** workspace 1
***** window 1
*** bottomdock
**** dock window 1
** VGA1

*Example:*
-----------------------
{
 "id": 6875648,
 "name": "root",
 "rect": {
   "x": 0,
   "y": 0,
   "width": 1280,
   "height": 800
 },
 "nodes": [

   {
    "id": 6878320,
    "name": "LVDS1",
    "layout": "output",
    "rect": {
      "x": 0,
      "y": 0,
      "width": 1280,
      "height": 800
    },
    "nodes": [

      {
       "id": 6878784,
       "name": "topdock",
       "layout": "dockarea",
       "orientation": "vertical",
       "rect": {
         "x": 0,
	 "y": 0,
	 "width": 1280,
	 "height": 0
       },
      },

      {
       "id": 6879344,
       "name": "content",
       "rect": {
         "x": 0,
	 "y": 0,
	 "width": 1280,
	 "height": 782
       },
       "nodes": [

         {
          "id": 6880464,
	  "name": "1",
	  "orientation": "horizontal",
	  "rect": {
            "x": 0,
	    "y": 0,
	    "width": 1280,
	    "height": 782
	  },
	  "floating_nodes": [],
	  "nodes": [

            {
             "id": 6929968,
	     "name": "#aa0000",
	     "border": "normal",
	     "percent": 1,
	     "rect": {
               "x": 0,
	       "y": 18,
	       "width": 1280,
	       "height": 782
	     }
	    }

	  ]
	 }

       ]
      },

      {
       "id": 6880208,
       "name": "bottomdock",
       "layout": "dockarea",
       "orientation": "vertical",
       "rect": {
         "x": 0,
	 "y": 782,
	 "width": 1280,
	 "height": 18
       },
       "nodes": [

         {
          "id": 6931312,
	  "name": "#00aa00",
	  "percent": 1,
	  "rect": {
            "x": 0,
	    "y": 782,
	    "width": 1280,
	    "height": 18
	  }
	 }

       ]
      }
    ]
   }
 ]
}
------------------------

=== MARKS reply

The reply consists of a single array of strings for each container that has a
mark. A mark can only be set on one container, so the array is unique.
The order of that array is undefined.

If no window has a mark the response will be the empty array [].

=== BAR_CONFIG reply

This can be used by third-party workspace bars (especially i3bar, but others
are free to implement compatible alternatives) to get the +bar+ block
configuration from i3.

Depending on the input, the reply is either:

empty input::
	An array of configured bar IDs
Bar ID::
	A JSON map containing the configuration for the specified bar.

Each bar configuration has the following properties:

id (string)::
	The ID for this bar. Included in case you request multiple
	configurations and want to differentiate the different replies.
mode (string)::
	Either +dock+ (the bar sets the dock window type) or +hide+ (the bar
	does not show unless a specific key is pressed).
position (string)::
	Either +bottom+ or +top+ at the moment.
status_command (string)::
	Command which will be run to generate a statusline. Each line on stdout
	of this command will be displayed in the bar. At the moment, no
	formatting is supported.
font (string)::
	The font to use for text on the bar.
workspace_buttons (boolean)::
	Display workspace buttons or not? Defaults to true.
binding_mode_indicator (boolean)::
	Display the mode indicator or not? Defaults to true.
verbose (boolean)::
	Should the bar enable verbose output for debugging? Defaults to false.
colors (map)::
	Contains key/value pairs of colors. Each value is a color code in hex,
	formatted #rrggbb (like in HTML).

The following colors can be configured at the moment:

background::
	Background color of the bar.
statusline::
	Text color to be used for the statusline.
separator::
	Text color to be used for the separator.
focused_workspace_text/focused_workspace_bg::
	Text color/background color for a workspace button when the workspace
	has focus.
active_workspace_text/active_workspace_bg::
	Text color/background color for a workspace button when the workspace
	is active (visible) on some output, but the focus is on another one.
	You can only tell this apart from the focused workspace when you are
	using multiple monitors.
inactive_workspace_text/inactive_workspace_bg::
	Text color/background color for a workspace button when the workspace
	does not have focus and is not active (visible) on any output. This
	will be the case for most workspaces.
urgent_workspace_text/urgent_workspace_bar::
	Text color/background color for workspaces which contain at least one
	window with the urgency hint set.


*Example of configured bars:*
--------------
["bar-bxuqzf"]
--------------

*Example of bar configuration:*
--------------
{
 "id": "bar-bxuqzf",
 "mode": "dock",
 "position": "bottom",
 "status_command": "i3status",
 "font": "-misc-fixed-medium-r-normal--13-120-75-75-C-70-iso10646-1",
 "workspace_buttons": true,
 "binding_mode_indicator": true,
 "verbose": false,
 "colors": {
   "background": "#c0c0c0",
   "statusline": "#00ff00",
   "focused_workspace_text": "#ffffff",
   "focused_workspace_bg": "#000000"
 }
}
--------------

=== VERSION reply

The reply consists of a single JSON dictionary with the following keys:

major (integer)::
	The major version of i3, such as +4+.
minor (integer)::
	The minor version of i3, such as +2+. Changes in the IPC interface (new
	features) will only occur with new minor (or major) releases. However,
	bugfixes might be introduced in patch releases, too.
patch (integer)::
	The patch version of i3, such as +1+ (when the complete version is
	+4.2.1+). For versions such as +4.2+, patch will be set to +0+.
human_readable (string)::
	A human-readable version of i3 containing the precise git version,
	build date and branch name. When you need to display the i3 version to
	your users, use the human-readable version whenever possible (since
	this is what +i3 --version+ displays, too).

*Example:*
-------------------
{
   "human_readable" : "4.2-169-gf80b877 (2012-08-05, branch \"next\")",
   "minor" : 2,
   "patch" : 0,
   "major" : 4
}
-------------------

== Events

[[events]]

To get informed when certain things happen in i3, clients can subscribe to
events. Events consist of a name (like "workspace") and an event reply type
(like I3_IPC_EVENT_WORKSPACE). The events sent by i3 are in the same format
as replies to specific commands. However, the highest bit of the message type
is set to 1 to indicate that this is an event reply instead of a normal reply.

Caveat: As soon as you subscribe to an event, it is not guaranteed any longer
that the requests to i3 are processed in order. This means, the following
situation can happen: You send a GET_WORKSPACES request but you receive a
"workspace" event before receiving the reply to GET_WORKSPACES. If your
program does not want to cope which such kinds of race conditions (an
event based library may not have a problem here), I suggest you create a
separate connection to receive events.

=== Subscribing to events

By sending a message of type SUBSCRIBE with a JSON-encoded array as payload
you can register to an event.

*Example:*
---------------------------------
type: SUBSCRIBE
payload: [ "workspace", "output" ]
---------------------------------


=== Available events

The numbers in parenthesis is the event type (keep in mind that you need to
strip the highest bit first).

workspace (0)::
	Sent when the user switches to a different workspace, when a new
	workspace is initialized or when a workspace is removed (because the
	last client vanished).
output (1)::
	Sent when RandR issues a change notification (of either screens,
	outputs, CRTCs or output properties).
mode (2)::
	Sent whenever i3 changes its binding mode.
window (3)::
	Sent when a client's window is successfully reparented (that is when i3
	has finished fitting it into a container), when a window received input
	focus or when certain properties of the window have changed.
barconfig_update (4)::
    Sent when the hidden_state or mode field in the barconfig of any bar
    instance was updated and when the config is reloaded.
binding (5)::
	Sent when a configured command binding is triggered with the keyboard or
	mouse

*Example:*
--------------------------------------------------------------------
# the appropriate 4 bytes read from the socket are stored in $input

# unpack a 32-bit unsigned integer
my $message_type = unpack("L", $input);

# check if the highest bit is 1
my $is_event = (($message_type >> 31) == 1);

# use the other bits
my $event_type = ($message_type & 0x7F);

if ($is_event) {
  say "Received event of type $event_type";
}
--------------------------------------------------------------------

=== workspace event

This event consists of a single serialized map containing a property
+change (string)+ which indicates the type of the change ("focus", "init",
"empty", "urgent"). A +current (object)+ property will be present with the
affected workspace whenever the type of event affects a workspace (otherwise,
it will be +null).

When the change is "focus", an +old (object)+ property will be present with the
previous workspace.  When the first switch occurs (when i3 focuses the
workspace visible at the beginning) there is no previous workspace, and the
+old+ property will be set to +null+.  Also note that if the previous is empty
it will get destroyed when switching, but will still be present in the "old"
property.

*Example:*
---------------------
{
 "change": "focus",
 "current": {
  "id": 28489712,
  "type": "workspace",
  ...
 }
 "old": {
  "id": 28489715,
  "type": "workspace",
  ...
 }
}
---------------------

=== output event

This event consists of a single serialized map containing a property
+change (string)+ which indicates the type of the change (currently only
"unspecified").

*Example:*
---------------------------
{ "change": "unspecified" }
---------------------------

=== mode event

This event consists of a single serialized map containing a property
+change (string)+ which holds the name of current mode in use. The name
is the same as specified in config when creating a mode. The default
mode is simply named default.

*Example:*
---------------------------
{ "change": "default" }
---------------------------

=== window event

This event consists of a single serialized map containing a property
+change (string)+ which indicates the type of the change

* +new+ - the window has become managed by i3
* +close+ - the window has closed
* +focus+ - the window has received input focus
* +title+ - the window's title has changed
* +fullscreen_mode+ - the window has entered or exited fullscreen mode
* +move+ - the window has changed its position in the tree
* +floating+ - the window has transitioned to or from floating
* +urgent+ - the window has become urgent or lost its urgent status

Additionally a +container (object)+ field will be present, which consists
of the window's parent container. Be aware that for the "new" event, the
container will hold the initial name of the newly reparented window (e.g.
if you run urxvt with a shell that changes the title, you will still at
this point get the window title as "urxvt").

*Example:*
---------------------------
{
 "change": "new",
 "container": {
  "id": 35569536,
  "type": "con",
  ...
 }
}
---------------------------

=== barconfig_update event

This event consists of a single serialized map reporting on options from the
barconfig of the specified bar_id that were updated in i3. This event is the
same as a +GET_BAR_CONFIG+ reply for the bar with the given id.

=== binding event

This event consists of a single serialized map reporting on the details of a
binding that ran a command because of user input. The +change (sring)+ field
indicates what sort of binding event was triggered (right now it will always be
+"run"+ but may be expanded in the future).

The +binding (object)+ field contains details about the binding that was run:

command (string)::
	The i3 command that is configured to run for this binding.
mods (array of strings)::
	The modifier keys that were configured with this binding.
input_code (integer)::
	If the binding was configured with +bindcode+, this will be the key code
	that was given for the binding. If the binding is a mouse binding, it will be
	the number of the mouse button that was pressed. Otherwise it will be 0.
symbol (string or null)::
	If this is a keyboard binding that was configured with +bindsym+, this
	field will contain the given symbol. Otherwise it will be +null+.
input_type (string)::
	This will be +"keyboard"+ or +"mouse"+ depending on whether or not this was
	a keyboard or a mouse binding.

*Example:*
---------------------------
{
 "change": "run",
 "binding": {
  "command": "nop",
  "mods": [
    "shift",
    "ctrl"
  ],
  "input_code": 0,
  "symbol": "t",
  "input_type": "keyboard"
 }
}
---------------------------

== See also (existing libraries)

[[libraries]]

For some languages, libraries are available (so you don’t have to implement
all this on your own). This list names some (if you wrote one, please let me
know):

C::
	* i3 includes a headerfile +i3/ipc.h+ which provides you all constants.
	* https://github.com/acrisci/i3ipc-glib
Go::
	* https://github.com/proxypoke/i3ipc
JavaScript::
	* https://github.com/acrisci/i3ipc-gjs
Lua::
	* https://github.com/acrisci/i3ipc-lua
Perl::
	* https://metacpan.org/module/AnyEvent::I3
Python::
	* https://github.com/acrisci/i3ipc-python
	* https://github.com/whitelynx/i3ipc (not maintained)
	* https://github.com/ziberna/i3-py (not maintained)
Ruby::
	* http://github.com/badboy/i3-ipc