ISO_LEVEL_3 Latch requires delays on linux GNOME & Wayland

[Ced-C]

Requirements

• I've searched issues to see if this has not been reported before.

Describe the bug

Hi,
Thanks for writing Kanata, great app really except for this annoying bug.
Like many French using a custom layout, I am using an ISO_LEVEL_3 latch key (let's call it il3 key) to access some diacritic letters (é è ê ù â ç ñ etc.).
for instance, typing il3 + (then) e →(gives)è, similarly; il3+u →ù
The issue is, that I am not able to take full advantage of kanata because if I tape il3 key then any key that is used as a mod-tap or a layer-tap, the il3 key is ignored.

• if u is not modified by kanata: il3+u →ù
• if u is a mod-tap or a layer-tap: il3+u →u
  a workaround is to maintain the il3 key hold wile typing the u key, this provides the expected ù, but it is really strange to type like that —especially when one is touch-typing.

Relevant kanata config

;; Quick notes about the kanata configuration syntax:
;;  - the syntax is Lisp-like but totally unrelated to Lisp (EMACS), so no worries :-)
;;  - comments are prefixed by double-semicolons
;;  - only one `defsrc` is allowed, listing all keys handled by kanata
;;  - up to 25 `deflayer` are allowed, one per layer emulation
;;  - the underscore symbol `_` in `deflayer` sections means "transparent",
;;    i.e. the previous layer behaviour is used when pressing that key

;; `defsrc` defines the keys that will be intercepted by kanata.
;; The order of the keys matches with deflayer declarations and all deflayer
;; declarations must have the same number of keys as defsrc. Any keys not listed
;; in defsrc will be passed straight to the operating system.


;; Adding ISO key definition
(deflocalkeys-linux
  <    105
)

;;;
;;; ON A ISO KEYBOARD, STD DEFINITION
;;;
(defsrc
  grv           1    2    3    4    5    6    7    8    9    0    -    =    bspc
  tab           q    w    e    r    t    y    u    i    o    p    [    ]    \
  caps          a    s    d    f    g    h    j    k    l    ;    '    ret
  lsft     <    z    x    c    v    b    n    m    ,    .    /    rsft
  lctl lmet lalt           spc            ralt rmet rctl
)


;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
;; Base layer (active by default when kanata starts up)
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
;;  - the 3 main thumb keys (Alt/tbl, Space/nav, AltGr/tbr) become mod/taps
;;  - Ctrl and Alt mods on Z X, to 
;;        - avoid activating modtap by accident
;;        - Alt is also very close to its intitial position for alt+tab shortcut
;;  - CapsLock stays the sape capword (caps until end of word) seems buggy so far…
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
(deflayer qwerty
_             _    _    _    _    _    _    _    _    _    _    _    _    _
_             _    _    _    _    _    _    _    _    _    _    _    _    _
caps          _    _    _    _    _    _    _    _    _    _    _    _
@sfl     _    @ctr @alt _    _    _    _    _    _    _    _    @sfr
_    _    @tbl          @nav           @tbr  _    _
)


;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
;; Navigation layer:
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
;;  - right: Vim-like arrows on HJKL, home/end page up/down, mouse scroll
;;  - left: numbers for easy excel (navigation+numbers) and long nuber sequence
;;  - Esc & Return on thumb keys
;; The `lrld` action stands for "live reload". This will re-parse everything
;; except for linux-dev, i.e. you cannot live reload and switch keyboard devices.
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
(deflayer vimnav
  _         M-1  M-2  M-3  M-4  M-5  M-6  M-7  M-8  M-9  M-0  C--  C-=  lrld
  _         _    4    5    6    _    home pgdn pgup end  _    _    _    _
  caps      C-a  1    2    3    0    lft  down up   rght _    _    _
  _    _    _    7    8    9    C-v  @mwl @mwd @mwu @mwr _    _
  _    _    esc            _              ret  _    _
)



;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
;; Aliases
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
;; `defalias` is used to declare a shortcut for a more complicated action to keep
;; the deflayer declarations clean and aligned. The alignment in deflayers is
;; not necessary, but is highly recommended for ease of understanding visually.
;; Aliases are referred to by `@<alias_name>`.
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
(defalias

  ;; layer-switch changes the base layer.
  std (layer-switch qwerty)
  nav (tap-hold 200 200 spc (layer-toggle vimnav))

  ;; Thumb keys :
  tbl (tap-hold-press 200 200 bspc lsft) 
  tbr (tap-hold-press 200 200 del ralt)
  ;; previous config with dead shift (tap-hold-press 200 200 (one-shot-press 2000 rsft) ralt)

  ;; Tap-mods definition
  ;; Must be hold long enough (200ms) to become a modifier.
  ctr (tap-hold 200 200 z lctl)
  alt (tap-hold 200 200 x lalt)
  sfl (tap-hold 200 200 del lsft)
  sfr (tap-hold 200 200 ret  rsft)
  
  ;; Unused for now
  cw (caps-word 2000)

  ;; Mouse wheel emulation
  mwu (mwheel-up    50 120)
  mwd (mwheel-down  50 120)
  mwl (mwheel-left  50 120)
  mwr (mwheel-right 50 120)
)

;; vim: set ft=lisp

To Reproduce

1. Use any layout with iso_level_3_latch - mine is here
2. I have for instance il3 on space (␣) to give the curved apostrophe il3+␣→’
3. With the above configuration of kanata active, space on hold is a num-nav layer thus

il3+␣→␣

Expected behavior

il3+␣→’

Kanata version

1.4

Debug logs

Activating the debug mod and doing il3+␣:

2024-01-02T12:32:11.511255505+01:00 [DEBUG] (2) kanata::kanata: process recv ev KeyEvent { code: KEY_N, value: Press }
2024-01-02T12:32:11.511294759+01:00 [DEBUG] (2) kanata::kanata: key press     N
2024-01-02T12:32:11.511303265+01:00 [DEBUG] (2) kanata::oskbd::linux: send to uinput: InputEvent { time: SystemTime { tv_sec: 0, tv_nsec: 0 }, kind: Key(KEY_N), value: 1 }
2024-01-02T12:32:11.568231766+01:00 [DEBUG] (2) kanata::kanata: process recv ev KeyEvent { code: KEY_N, value: Release }
2024-01-02T12:32:11.568266+01:00 [DEBUG] (2) kanata::kanata: key release   N
2024-01-02T12:32:11.568272963+01:00 [DEBUG] (2) kanata::oskbd::linux: send to uinput: InputEvent { time: SystemTime { tv_sec: 0, tv_nsec: 0 }, kind: Key(KEY_N), value: 0 }
2024-01-02T12:32:13.707200639+01:00 [DEBUG] (2) kanata::kanata: process recv ev KeyEvent { code: KEY_SPACE, value: Press }
2024-01-02T12:32:13.775085966+01:00 [DEBUG] (2) kanata::kanata: process recv ev KeyEvent { code: KEY_SPACE, value: Release }
2024-01-02T12:32:13.775111754+01:00 [DEBUG] (2) kanata::kanata: key press     Space
2024-01-02T12:32:13.775119939+01:00 [DEBUG] (2) kanata::oskbd::linux: send to uinput: InputEvent { time: SystemTime { tv_sec: 0, tv_nsec: 0 }, kind: Key(KEY_SPACE), value: 1 }
2024-01-02T12:32:13.775155276+01:00 [DEBUG] (2) kanata::kanata: key release   Space
2024-01-02T12:32:13.775162359+01:00 [DEBUG] (2) kanata::oskbd::linux: send to uinput: InputEvent { time: SystemTime { tv_sec: 0, tv_nsec: 0 }, kind: Key(KEY_SPACE), value: 0 }

Operating system

Linux, Ubuntu 22.04, Gnome-42 Wayland

Additional context

I can make additional test if required. I am not sure how trivial this is…
Maybe something to do with action on released key rather than pressed ?

[rszyma]

I don't see any issues with your setup, it's just that you have (tap-hold-press 200 200 del ralt) on your alt key, so you actually have to hold alt (or on your layout N) for 200 ms before alt (ISO_LEVEL_3_Latch) is actually sent.

So kanata is working as intended.

[Ced-C]

Hi,
Sorry I did not receive a notification, so I did not see your answer.
I have no issue with the alt key:

• If I tap on the alt key, I get delete
• If I hold for 200 ms (wich is long enought to prevent accidental hold), I get access to the AltGr layer set as an ISO_LEVEL_5_Shift)

The issue is, I have another key (called il3 in my first post) located in [N] (with [N] the QWERTY physical position of keycaps with letter N) that is used to provide French diacritics. Technically, this is implemented as an ISO_LEVEL_3_Latch (different from AltGr that is an ISO_LEVEL_5_Shift). It is like a dead key (like dead ^ is used to make ê) but that can be combined with many more letters/symbols so that one has complete access of all needed characters to write properly in its language.

This key (il3) do not work properly when combined with another key that is modified by kanata (mod-tap or layer-tap).
I think kanata thinks of this key as an ISO_LEVEL_3**_Shift** instead of an ISO_LEVEL_3**_Latch**. A latch should work on the next key pressed with shift requires to be kept held.

I hope it is clearer now. I do think this should be labelled as a bug and not just a question, but I might have to further explain myself as the matter is not common for most english-speaking devs (few-to-no diacritics in english, and classical deadkeys are often enough).

[jtroo]

Thanks for the reply @Ced-C.

I'll keep it marked as a question for now since it's still unclear to me how this should be fixed within kanata.

There is no concept of il3 at the layer that kanata works with. Kanata receives inputs and sends outputs at the Linux kernel device event level, whereas according to my reading, il3 operates further along in the pipeline at the desktop/xkb level. When tapping a tap-hold key like @nav, what xkb sees should be very similar as if it was simply spc.

When I think about in what ways it looks different, the main thing I can think of is the time gap between the press and release events. Perhaps since the release of the spc event is sent only 40μs after the press, the xkb layer isn't processing it as desired?

A potential workaround might be to have the tap key be (macro spc) since that will extend the time between press and release to approximately 1ms.

  nav (tap-hold 200 200 (macro spc) (layer-toggle vimnav))

Or alternatively, use something like below to make the press and release even further apart - in this example 20ms.

(deffakekeys spc spc)
(defalias
  nav (tap-hold 200 200 
    (macro (on-press-fakekey spc press) 20 (on-press-fakekey spc release))
    (layer-toggle vimnav)
  )
)

[Ced-C]

Thanks for your swift answer. The first suggestion did not seem to work, but the 2nd one did!
So I suppose, all I need to do now, is to use the same trick on every tap-mod (like [X] ) that is used in my config (if they are used in a sequence after my il3 key.
Thanks :-)

Should I close this issue now ? or do we want to document that somewhere ? Let me know!

Édit : timing becomes critical and seed to lead to some inversions, I’ll continue to test and play with the timing to see if the behaviour can be improved. Will not close the issue and let you know how it goes if it is ok with you.

[jtroo]

Sure thing, once we have more info let's add it to the documentation.

I wonder if it might be worth opening a bug, or see if there is an existing one, against xkb - or whatever project handles this functionality, I don't have any links on hand. To me it seems undesirable to have this timing-dependent behaviour, when all that matters is the key press/event sequence.

[Ced-C]

Agreed, depending on the timing the exact same sequence at approx the same frequency can give a different result which is not what one would want to write.
I will soon test if inverting AltGr to be on the iso_level_3 and the il3 key on the iso_level_5 instead improve things. I think the iso_level_5 was introduced around 2005 and it was quite unstable for a long time, so it would not surprise me if it improves things somehow.

[Ced-C]

Hey,
I am not sure if it is a Ubuntu+GNOME+Wayland thing, but a guy on discord provided a great config to avoid all of these issues
instead of adding delays which were inconsistent in my experience, just add /etc/environment

INPUT_METHOD=ibus
GTK_IM_MODULE=ibus
QT_IM_MODULE=ibus
XMODIFIERS=@im=ibus

works with ibus 1.28-1.30-rc2 which is needed by GNOME but apparently not used in all apps.