Direct input of Unicode hiragana from XKB does not trigger IME suggestions

[aaannestad]

Description

I've been experimenting with alternative Japanese keyboard layouts, and have created an XKB layout that simply outputs Unicode hiragana. Unfortunately, this does not seem to actually trigger the IME to provide suggestions. Within fcitx I have my custom keyboard selected as mozc's keyboard layout, and switching to mozc does allow me to type kana with my layout, but the actual IME functions simply do not happen.

I can switch mozc to 'kana' input mode, but that doesn't solve the issue; that seems to be a simple conversion of a Roman letter's QWERTY location to the kana in that location on a tateisukan layout, and thus it still expects Roman-letter input.

Steps to reproduce

Steps to reproduce the behavior:

1. Create a custom xkb symbols file that will allow you to type at least one hiragana codepoint, and add it to your evdev.xml so that fcitx can see it (then reboot)
2. In fcitx's configuration tool, set mozc's underlying keyboard layout to this custom keyboard layout
3. Switch to mozc, and type the key that has that hiragana codepoint

Expected behavior

mozc treats that hiragana exactly as a hiragana converted from romaji, and supplies kanji etc suggestions

Actual behavior

mozc directly outputs the kana as a letter it seems to know nothing about, bypassing the conversion process entirely

Version or commit-id　

Mozc-3.34.6239.2+24.11.oss

Environment

• OS: Arch
• IMF (for Linux): fcitx

Investigations

• Whether this issue happens on both Ibus and other IMF (e.g. Fcitx, uim).
  • have not installed Ibus etc
• Whether this issue happens on other IMEs (e.g. Anthy, SKK).
  • have not installed Anthy etc
• What applications this issue happens on (e.g. Chromium, gedit).
  • all
• What applications this issue does not happen on (e.g. Chromium, gedit).
  • none

[yukawa]

I've been experimenting with alternative Japanese keyboard layouts

For that use case, I think what people have used is to create a custom Romaji-table instead. Can you look into the following discussion threads to see if it works for you?

• Incorrect keyboard layout #427
• US Layout, Kana sublegends don't match up anymore. #757
• Is it technically difficult to let users customise Kana key mapping? #1235

If it works, then it should work not only for fcitx-mozc but also for other clients such as ibus-mozc and even Windows/macOS version of Mozc.

Another approach might be directly editing the following table in src/unix/fcitx5/fcitx_key_translator.cc.
https://github.com/fcitx/mozc/blob/a8ddb92b3617f869b88fb6185b6c95cb0b8787ae/src/unix/fcitx5/fcitx_key_translator.cc#L158-L297

[aaannestad]

I don't know if those are options for this particular keyboard layout, as it has different modifier keys than my usual Roman-letter keyboard - I'm trying a thumb-shift style layout, and my right thumb button isn't a modifier otherwise. It also mildly bothers me to have to go through some stage of Roman letter input, even if it's just a hidden layer; though that's not the end of the world if it has to be there - it just seems like dependence on ASCII character sets and Roman-letter keyboard input is a legacy of older technology and has no need to remain. Still, I'm somewhat confused as to why mozc can't just start with kana directly and convert from there.