diacritic marks get deleted

[pandadiestro]

As seen in this video, the dead key gets displayed but after any other key is pressed, the mark just disappears. At first I thought this might be a locale issue but the same test input works fine in st.

2025-01-25.19-14-31.1.mp4

---

ghostty version info:

Ghostty 1.0.2-main+71e62f96

Version
  - version: 1.0.2-main+71e62f96
  - channel: tip
Build Config
  - Zig version: 0.13.0
  - build mode : builtin.OptimizeMode.ReleaseFast
  - app runtime: apprt.Runtime.gtk
  - font engine: font.main.Backend.fontconfig_freetype
  - renderer   : renderer.OpenGL
  - libxev     : main.Backend.io_uring
  - desktop env: other
  - GTK version:
    build      : 4.16.12
    runtime    : 4.16.12
  - libadwaita : disabled
  - libX11     : enabled
  - libwayland : enabled

relevant system info:

OS: Artix Linux x86_64
Kernel: Linux 6.12.9-artix1-1
Shell: zsh 5.9
Terminal Font: Cousine Nerd Font Propo (10pt)
Locale: en_US.UTF-8

[mitchellh]

Please provide the following:

• Keyboard layout
• Windowing protocol (X11 vs Wayland)
• Example keys to press on a physical US layout keyboard to reproduce

[pandadiestro]

Keyboard Layout:

$ setxkbmap -query
rules:      evdev
model:      pc105
layout:     latam
options:    terminate:ctrl_alt_bksp

Windowing protocol: X11

As for the example keys, taking this wikipedia representation of my layout as example, the keys i'm pressing are the one to the right of 'P' (single press for the accent dead key and shift + press for the diaeresis) and the backtick from the closing bracket (altgr + the key), I don't think these are available in the US layout tho.

If it helps, in this video i repeated the test input with the key inspector on

2025-01-26.11-54-58.1.mp4

[McKael]

Hello,

FWIW, I'm able to reproduce this (or a very similar) issue on a Debian 12 (bookworm) and on a Ubuntu 24.04 (Noble Numbat).

I'm using Sway (so Wayland), with an azerty layout, and like the OP I'm unable to input special characters requiring a dead key (the compose key doesn't work anymore, nor combinations like in @pandadiestro 's video to type characters like ô).

I'm not using any specific Input Method; GTK_IM_MODULE is not set.

If I install ibus or fcitx5 and set this variable accordingly (e.g. GTK_IM_MODULE=fcitx), it works as expected (but it didn't use to be required).
I'm not sure what input method GTK/Ghostty uses when the variable is not set.

I've been able to check that the behavior has changed with the commit 52936b9b681e92f7bab37c3ec341be0013f591da. IOW, using the parent (5cb2fa6f7594202b12a2603bf32094aa75b1bf0e) fixes the issue for me.

HTH!

[pandadiestro]

This might be off-topic but I couldn't reproduce this solution. Under the AUR, fcitx5-gtk wont build and when I tried using both fcitx5 and fcitx as im module (after installing fcitx5) it didn't find the module and failed with:

(ghostty:1106): Gtk-WARNING **: 17:24:01.119: No IM module matching GTK_IM_MODULE=fcitx found

GTK_IM_MODULE=ibus after installing ibus won't work either, the module apparently is found (since no errors are displayed) but in this case the dead key is not even displayed.

[McKael]

GTK_IM_MODULE=ibus after installing ibus won't work either, the module apparently is found (since no errors are displayed) but in this case the dead key is not even displayed.

Is ibus daemon running? I had to start it manually after installing it (IIRC I used a command like ibus start --type direct for the test). With fcitx5 on Debian I didn't have to launch the daemon manually.

Hope that helps,
Regards

[pandadiestro]

Thanks! Starting the daemon through ibus start worked. It's a weird bug nonetheless.

[rajayonin]

Running CachyOS (Arch Linux) through Gnome and Wayland, installed the package through pacman and the issue appeared when upgrading to v1.1.0 (you can see my post at #5471). My layout is es.

Installing ibus and running ibus start worked even after a restart.

Very weird indeed...

[miyl]

Same issue but in Wayland rather than Xorg and my keyboard layout is dk (danish).
It also breaks things like ~ (tilde).

Have neither ibus nor fcitx, and every other program work just fine without them.

Tried downgrading to 1.0.3-3 from the arch repos and it works there.

Edit: I wonder if it's related to the changes in #3567

[McKael]

See my previous answer, it is related to the refacto that improved IBUS and FCITX support, but probably broke the builtin GTK ("Simple") input method.
I'm not sure why, but AFAIU IMContextSimple sends the final (composed) code but it doesn't get committed to the Surface.

Not sure about the ~ though, it still works here on Azerty layout, does it require a dead key with your layout?

[miyl]

Right, okay.

For tilde, on my layout it requires pressing AltGr + the same key key twice (ie. that's a dead key as I understand the term?). Specifically this key:

Incidentally neither that nor backticks have to be pressed twice in something like Firefox or LibreOffice - only in terminals, or at least kitty and ghostty.

[McKael]

Yes that seems to be a dead key (if I understand correctly, you would press AltGr+~ then n to compose the character ñ).

That's probably the same issue then.

[miyl]

Yes it can be composed with e.g. n like that: ñ.

But in fx. Firefox if I just want a tilde by itself I can just type it once and then a character, tilde doesn't compose with, such as l, and both characters have now been inputted: ~l.
If I do that in ghostty only the l appears. (no tilde)
If I do that in kitty none of the characters appear.
...so in both of those I have to press tilde twice to actually get a tilde.

That's not a part of this bug of course - that's how it works normally on ghostty - but I wonder why it works like this, now we're on the topic of diacritics/dead keys and possibly rewriting how they behave.

[McKael]

Not sure why you have a different behavior, but today I learned I had a dead ~ key as well on my fr keyboard layout, almost like yours (I can type AltGr plus the dead ^ key twice to get a ~ and AltGr+^ and o to get a õ). Works the same in Ghostty and Firefox here.
I can reproduce your tilde issue in my environment, BTW.

I have added a patch to the issue #5494 (probably buggy but it fixes the problem for me, until it gets properly addressed).

[McKael]

I think the issue #5494 might be addressing this problem (although the title mentions ibus 1.5.31, the root cause is probably the same).

[pandadiestro]

this was fixed by #5550 on my end

[pluiedev]

Please do not revive old threads. Create a new one.