keyszer - a smart key remapper for Linux/X11

Keyszer is a smart key remapper for Linux (and X11) written in Python. It's similar to xmodmap but allows far more flexible remappings. Keyszer was forked from xkeysnail which no longer seems actively maintained.

How does it work?

Keyszer works at quite a low-level. It grabs input directly from the kernel's evdev input devices ( /dev/input/event*) and then creates an emulated uinput device to inject those inputs back into the kernel. During this process the input stream is transformed on the fly as necessary to remap keys.

Upgrading from xkeysnail

• Some small configuration changes will be needed.
• A few command line arguments have changed.
• For xkeysnail 0.4.0 see UPGRADING_FROM_XKEYSNAIL.md.
• For xkeysnail (Kinto variety) see USING_WITH_KINTO.md and Using with Kinto v1.2-13.

Key Highlights

• Low-level library usage (evdev and uinput) allows remapping to work from the console all the way into X11.
• High-level and incredibly flexible remapping mechanisms:
  • per-application keybindings - bindings that change depending on the active X11 application or window
  • multiple stroke keybindings - Ctrl+x Ctrl+c could map to Ctrl+q
  • very flexible output - Ctrl-s could type out :save, and then hit enter
  • stateful key combos - build Emacs style combos with shift/mark
  • multipurpose bindings - a regular key can become a modifier when held
  • arbitrary functions - a key combo can run custom Python function

New Features (since xkeysnail 0.4.0)

• simpler and more flexible configuration scripting APIs
• better debugging tools
  • configurable EMERGENCY EJECT hotkey
  • configurable DIAGNOSTIC hotkey
• fully supports running as semi-privileged user (using root is now deprecated)
• adds include to allow config to pull in other Python files
• adds throttle_delays to allow control of output speed of macros/combos
• adds immediately to nested keymaps
• adds Meta, Command and Cmd aliases for Super/Meta modifier
• add C combo helper (eventually to replace K)
• supports custom modifiers via add_modifier (such as Hyper)
• supports Fn as a potential modifier (on hardware where it works)
• adds bind helper to support persistent holds across multiple combos
  • most frequently used for persistent Mac OS style Cmd-tab app switcher panels
• adds --check for checking the config file for issues
• adds wm_name context for all conditionals (PR #40)
• adds device_name context for all conditionals (including keymaps)
• (fix) xmodmap cannot be used until some keys are first pressed on the emulated output
• (fix) ability to avoid unintentional Alt/Super false keypresses in many setups
• (fix) fixes multi-combo nested keymaps (vs Kinto's xkeysnail)
• (fix) properly cleans up pressed keys before termination
• individually configurable timeouts (multipurpose and suspend)
• (fix) removed problematic launch macro
• (fix) suspend extra keys during sequential sequences to create less key "noise"
• (fix) handle X Display errors without crashing or bugging out

---

Installation

Requires Python 3.

Over time we should add individual instructions for various distros here.

From source

Just download the source and install.

git clone https://github.com/joshgoebel/keyszer.git
cd keyszer
pip3 install --user --upgrade .

For testing/hacking/contributing

Using a Python venv might be the simplest way to get started:

git clone https://github.com/joshgoebel/keyszer.git
cd keyszer
python -m venv .venv
source .venv/bin/activate
pip3 install -e .
./bin/keyszer -c config_file

System Requirements

Keyszer requires read/write access to:

• /dev/input/event* - to grab input from your evdev input devices
• /dev/uinput - to provide an emulated keyboard to the kernel

Running as a semi-privileged user

It's best to create an entirely isolated user to run the keymapper. Group or ACL based permissions can be used to provide this user access to the necessary devices. You'll need only a few udev rules to ensure that the input devices are all given correct permissions.

ACL based permissions (narrow, more secure)

First, lets make a new user:

sudo useradd keymapper

...then use udev and ACL to grant our new user access:

Manually edit /etc/udev/rules.d/90-keymapper-acl.rules to include the following:

KERNEL=="event*", SUBSYSTEM=="input", RUN+="/usr/bin/setfacl -m user:keymapper:rw /dev/input/%k"
KERNEL=="uinput", SUBSYSTEM=="misc", RUN+="/usr/bin/setfacl -m user:keymapper:rw /dev/uinput"

...or do it by copypasting these lines into a shell:

cat <<EOF | sudo tee /etc/udev/rules.d/90-keymapper-acl.rules
KERNEL=="event*", SUBSYSTEM=="input", RUN+="/usr/bin/setfacl -m user:keymapper:rw /dev/input/%k"
KERNEL=="uinput", SUBSYSTEM=="misc", RUN+="/usr/bin/setfacl -m user:keymapper:rw /dev/uinput"
EOF

Group based permissions (slightly wider, less secure)

Many distros already have an input group; if not, you can create one. Next, add a new user that's a member of that group:

sudo useradd keymapper -G input

...then use udev to grant our new user access (via the input group):

Manually edit /etc/udev/rules.d/90-keymapper-input.rules to include the following:

SUBSYSTEM=="input", GROUP="input"
KERNEL=="uinput", SUBSYSTEM=="misc", GROUP="input"

...or do it by copypasting these lines into a shell:

cat <<EOF | sudo tee /etc/udev/rules.d/90-keymapper-input.rules
SUBSYSTEM=="input", GROUP="input"
KERNEL=="uinput", SUBSYSTEM=="misc", GROUP="input"
EOF

systemd

For a sample systemd service file for running Keyszer as a service please see keyszer.service.

Running as the Active Logged in User

This may be appropriate in some limited development scenarios, but is not recommended. Giving the active, logged in user access to evdev and uinput potentially allows all keystrokes to be logged and could allow a malicious program to take over (or destroy) your machine by injecting input into a Terminal session or other application.

It would be better to open a terminal, su to a dedicated keymapper user and then run Keyszer inside that context, as shown earlier.

Running as root

Don't do this, it's dangerous, and unnecessary. A semi-privileged user with access to only the necessary input devices is a far better choice.

Usage

keyszer

A successful startup should resemble:

keyszer v0.5.0
(--) CONFIG: /home/jgoebel/.config/keyszer/config.py
(+K) Grabbing Apple, Inc Apple Keyboard (/dev/input/event3)
(--) Ready to process input.

Limiting Devices

Limit remapping to specific devices with --devices:

keyszer --devices /dev/input/event3 'Topre Corporation HHKB Professional'

The full path or complete device name may be used. Device name is usually better to avoid USB device numbering jumping around after a reboot, etc...

Other Options:

• -c, --config - location of the configuration file
• -w, --watch - watch for new keyboard devices to hot-plug
• -v - increase verbosity greatly (to help with debugging)
• --list-devices - list out all available input devices

Configuration

By default we look for the configuration in ~/.config/keyszer/config.py. You can override this location using the -c/--config switch. The configuration file is written in Python. For an example configuration please see example/config.py.

The configuration API:

• timeouts(multipurpose, suspend)
• throttle_delays(key_pre_delay_ms, key_post_delay_ms)
• wm_class_match(re_str)
• not_wm_class_match(re_str)
• add_modifier(name, aliases, key/keys)
• modmap(name, map, when_conditional)
• multipurpose_modmap(name, map, when_conditional)
• keymap(name, map, when_conditional)
• conditional(condition_fn, map) - used to wrap maps, applying them conditionally
• dump_diagnostics_key(key)
• emergency_eject_key(key)
• include(relative_filename)

include(relative_filename)

Include a sub-configuration file into the existing config. This file is loaded and executed at the point of inclusion and shares the same global scope as the existing config. These files should be present in the same directory as your main configuration.

include("os.py")
include("apps.py")
include("deadkeys.py")

timeouts(...)

Configures the timing behavior of various aspects of the keymapper.

• multipurpose - The number of seconds before a held multi-purpose key is assumed to be a modifier (even in the absence of other keys).
• suspend - The number of seconds modifiers are "suspended" and withheld from the output waiting to see whether if they are part of a combo or if they may be the actual intended output.

Defaults:

timeouts(
    multipurpose = 1,
    suspend = 1,
)

throttle_delays(...)

Configures the speed of virtual keyboard keystroke output to deal with issues that occur in various situations with the timing of modifier key presses and releases being misinterpreted.

• key_pre_delay_ms - The number of milliseconds to delay the press-release keystroke of the "normal" key after pressing modifier keys.
• key_post_delay_ms - The number of milliseconds to delay the next key event (modifier release) after the "normal" key press-release event.

Defaults:

throttle_delays(
    key_pre_delay_ms    = 0,    # default: 0 ms, range: 0 to 150 ms, suggested: 1-50 ms
    key_post_delay_ms   = 0,    # default: 0 ms, range: 0 to 150 ms, suggested: 1-100 ms
)

Use the throttle delays if you are having the following kinds of problems:

• Shortcut combos seeming to behave unreliably, sometimes as if the unmapped shortcut (or part of the unmapped shortcut) is being pressed at the same time.
• Macros of sets of keystrokes, or strings or Unicode sequences processed by the keymapper into keystrokes, having various kinds of failures, such as:
  • Missing characters
  • Premature termination of macro
  • Shifted or uppercase characters coming out as unshifted/lowercase
  • Unshifted or lowercase characters coming out as shifted/uppercase
  • Unicode sequences failing to complete and create the desired Unicode character

Suggested values to try if you are in a virtual machine and having major problems with even common shortcut combos:

• key_pre_delay_ms: 40
• key_post_delay_ms: 70

The post delay seems a little more effective in testing, but your situation may be different. For a bare-metal install where you are just having a few glitches in macro output, try much smaller delays:

• key_pre_delay_ms: 0.1
• key_post_delay_ms: 0.5

These are just examples that have worked fairly well in current testing on machines that have had these issues.

dump_diagnostics_key(key)

Configures a key that when hit will dump additional diagnostic information to STDOUT.

dump_diagnostics_key(Key.F15)  # default

emergency_eject_key(key)

Configures a key that when hit will immediately terminate keyszer; useful for development, recovering from bugs, or badly broken configurations.

emergency_eject_key(Key.F16)  # default

add_modifier(name, aliases, key/keys)

Allows you to add custom modifiers and then map them to actual keys.

add_modifier("HYPER", aliases = ["Hyper"], key = Key.F24)

Note: Just adding HYPER doesn't necessarily make it work with your software, you may still need to configure X11 setup to accept the key you choose as the "Hyper" key.

wm_class_match(re_str)

Helper to make matching conditionals (and caching the compiled regex) just a tiny bit simpler.

keymap("Firefox",{
    # ... keymap here
}, when = wm_class_match("^Firefox$"))

not_wm_class_match(re_str)

The negation of wm_class_match, matches only when the regex does NOT match.

modmap(name, mappings, when_conditional = None)

Maps a single physical key to a different key. A default modmap will always be overruled by any conditional modmaps that apply. when_conditional can be passed to make the modmap conditional. The first modmap found that includes the pressed key and matches the when_conditional will be used to remap the key.

modmap("default", {
    # mapping caps lock to left control
    Key.CAPSLOCK: Key.LEFT_CTRL
})

If you don't create a default (non-conditional) modmap a blank one is created for you. For modmap both sides of the pairing will be Key literals (not combos).

multipurpose_modmap(name, mappings)

Used to bestow a key with multiple-purposes, both for regular use and for use as a modifier.

multipurpose_modmap("default",
    # Enter is enter if pressed and immediately released...
    # ...but Right Control if held down and paired with other keys.
    {Key.ENTER: [Key.ENTER, Key.RIGHT_CTRL]}
)

keymap(name, mappings)

Defines a keymap of input combos mapped to output equivalents.

keymap("Firefox", {
    # when Cmd-S is input instead send Ctrl-S to the output
    C("Cmd-s"): C("Ctrl-s"),
}, when = lambda ctx: ctx.wm_class == "Firefox")

Because of the when conditional this keymap will only apply for Firefox.

The argument mappings is a dictionary in the form of { combo: command, ...} where combo and command take following forms:

• combo: Combo to map, specified by K(combo_str)
  • For the syntax of combo specifications, see Combo Specifications.
• command: one of the following
  • K(combo_str): Type a specific key combo to the output.
  • [command1, command2, ...]: Execute multiple commands sequentially.
  • { ... }: Sub-keymap. Used to define Multiple Stroke Keys.
  • escape_next_key: Escape the next key pressed.
  • ignore_key: Ignore the key that is pressed next. (often used to disable native combos)
  • bind: Bind an input and output modifier together such that the output is not lifted until the input is.
  • arbitrary function: The function is executed and the returned value (if any) is used as a command.

The argument name specifies the keymap name. Every keymap has a name - using default is suggested for a non-conditional keymap.

conditional(fn, map)

Applies a map conditionally, only when the fn function evaluates True. The below example is a modmap that is only active when the current WM_CLASS is Terminal.

conditional(
    lambda ctx: ctx.wm_class == "Terminal",
    modmap({
        # ...
    })
)

The context object passed to the fn function has several attributes:

• wm_class - the WM_CLASS of the [input] focused X11 window
• wm_name - the WM_NAME of the [input] focused X11 window
• device_name - name of the device where an input originated
• capslock_on - state of CapsLock (boolean)
• numlock_on - state of NumLock (boolean)

Note: The same conditional fn can always be passed directly to modmap using the when argument.

---

Marks

TODO: need docs (See issue #8)

Combo Specifications

The Combo specification in a keymap is written in the form of C("(<Modifier>-)*<Key>").

<Modifier> is one of the following:

• C or Ctrl -> Control key
• Alt -> Alt key
• Shift -> Shift key
• Super, Win, Command, Cmd, Meta -> Super/Windows/Command key
• Fn -> Function key (on supported keyboards)

You can specify left/right modifiers by adding the prefixes L or R.

<Key> is any key whose name is defined in key.py.

Some combo examples:

• C("LC-Alt-j"): left Control, Alt, j
• C("Ctrl-m"): Left or Right Control, m
• C("Win-o"): Cmd/Windows, o
• C("Alt-Shift-comma"): Alt, Left or Right Shift, comma

Multiple Stroke Keys

To use multiple stroke keys, simply define a nested keymap. For example, the following example remaps C-x C-c to C-q.

keymap("multi stroke", {
    C("C-x"): {
      C("C-c"): C("C-q"),
    }
})

If you'd like the first keystroke to also produce it's own output, immediately can be used:

keymap("multi stroke", {
  C("C-x"): {
    # immediately output "x" when Ctrl-X is pressed
    immediately: C("x"),
    C("C-c"): C("C-q"),
  }
})

Finding out the proper Key.NAME literal for a key on your keyboard

From a terminal session run evtest and select your keyboard's input device. Now hit the key in question.

Event: time 1655723568.594844, type 1 (EV_KEY), code 69 (KEY_NUMLOCK), value 1
Event: time 1655723568.594844, -------------- SYN_REPORT ------------

Above I've just pressed "clear" on my numpad and see code 69 (KEY_NUMLOCK) in the output. For Keyszer this would translate to Key.NUMLOCK. You can also browse the full list of key names in the source.

Finding an Application's WM_CLASS and WM_NAME using xprop

Use the xprop command from a terminal:

xprop WM_CLASS WM_NAME

...then click an application window. Let's try it with Google Chrome:

WM_CLASS(STRING) = "google-chrome", "Google-chrome"
WM_NAME(UTF8_STRING) = "README - Google Chrome"

Use the second WM_CLASS value (in this case Google-chrome) when matching context.wm_class.

Example of Case Insensitive Matching

terminals = ["gnome-terminal","konsole","io.elementary.terminal","sakura"]
terminals = [term.casefold() for term in terminals]
USING_TERMINAL_RE = re.compile("|".join(terminals), re.IGNORECASE)

modmap("not in terminal", {
    Key.LEFT_ALT: Key.RIGHT_CTRL,
    # ...
    }, when = lambda ctx: ctx.wm_class.casefold() not in terminals
)

modmap("terminals", {
    Key.RIGHT_ALT: Key.RIGHT_CTRL,
    # ...
    }, when = lambda ctx: USING_TERMINAL_RE.search(ctx.wm_class)
)

FAQ

Can I remap the keyboard's Fn key?

It depends. Most laptops do not allow this as the Fn keypress events are not directly exposed to the operating system. On some keyboards, it's just another key. To find out you can run evtest. Point it to your keyboard device and then hit a few keys; then try Fn. If you get output, then you can map Fn. If not, you can't.

Here is an example from a full size Apple keyboard I have:

Event: time 1654948033.572989, type 1 (EV_KEY), code 464 (KEY_FN), value 1
Event: time 1654948033.572989, -------------- SYN_REPORT ------------
Event: time 1654948033.636611, type 1 (EV_KEY), code 464 (KEY_FN), value 0
Event: time 1654948033.636611, -------------- SYN_REPORT ------------

What if my keyboard seems laggy or is not repeating keys fast enough?

You likely need to set the [virtual] keyboards repeat rate to match your actual keyboard.

Here is the command I use:

xset r rate 200 20

For best results your real keyboard and Keyszer [virtual] keyboard should have matching repeat rates. That seems to work best for me. Anytime you restart keyszer you'll need to reconfigure the repeat rate because each time a new virtual keyboard device is created... or maybe it's that there is only a single repeat rate and every time you "plug in" a new keyboard it changes?

If you could shed some light on this, please get in touch.

Does Keyszer support FreeBSD/NetBSD or other BSDs?

Not at the moment, perhaps never. If you're an expert on the BSD kernel's input layers please join the discussion. I'm at the very least open to the discussion to find out if this is possible, a good idea, etc...

Does this work with Wayland?

Not yet. This is desires but seems impossible at the moment until there is a standardized system to quickly and easily determine the app/window that has input focus on Wayland, just like we do so easily on X11.

Is keyszer compatible with Kinto.sh?

That is certainly the plan. The major reason Kinto.sh required it's own fork has been resolved. Kinto.sh should simply "just work" with keyszer (with a few tiny config changes). In fact, hopefully it works better than before since many quirks with the Kinto fork should be resolved. (such as nested combos not working, etc)

Reference:

• Kinto GitHub issue regarding the transition.
• Instructions on altering your kinto.py config slightly. See USING_WITH_KINTO.md.

How can I help or contribute?

Please open an issue to discuss how you'd like to get involved or respond on one of the existing issues. Also feel free to open new issues for feature requests. Many issues are tagged good first issue or help welcome.

License

keyszer is distributed under GPL3. See LICENSE.