strans/README.md

# strans

strans is a small, single-user input method for Korean, Japanese, English,
emoji, and symbols. It provides one engine for Wayland compositors that
speak `zwp_input_method_v2` — the wlroots and smithay families, so sway,
river, labwc, Wayfire, dwl, niri and COSMIC — and for IBus (which GTK 4
and Qt use), XIM, and GTK 3. Vietnamese Telex is also available as a
compatibility mode.

## Input modes

| Key | Mode |
| --- | --- |
| `Ctrl+N` | Japanese Hiragana and Kanji conversion |
| `Ctrl+K` | Japanese Katakana |
| `Ctrl+S` | Korean Hangul (2-beolsik) |
| `Ctrl+T` | English |
| `Ctrl+V` | Vietnamese Telex |
| `Ctrl+E` | Emoji and symbol search |
| `Ctrl+H` | One-shot Hanja search |

Switching mode shows the mode switched to — `A`, `한`, `あ`, `ア`, `ă` — in
the popup until the next key.

A Korean keyboard's 한/영 key toggles Korean and English, and its 한자 key
is `Ctrl+H`; a Japanese keyboard's 半角/全角 toggles Japanese and English,
ひらがな/カタカナ switches between the two kana modes, 変換 turns Japanese
on and then converts like `Space`, and 無変換 turns it off. Only a plain
`Ctrl` chord is a strans key: with `Shift`, `Alt`, or `Super` held it
commits what is pending and goes to the application, so `Ctrl+Shift+V`
still pastes.

Hiragana mode composes a complete reading, then shows its Kanji candidates
with none chosen: the exact readings of the dictionary, then the verbs and
adjectives SKK keys by stem and okurigana (かく offers 確 and 書く), and
under `Space` the reading in Katakana last, so a word the dictionary lacks
converts to Katakana with one `Space`. `Space` or `Tab` chooses the first
candidate and then steps on, wrapping (with `Shift`, back); `Up`/`Down` and
`PageUp`/`PageDown` move without wrapping. Once a candidate is chosen, an unmodified `1`-`9` commits that row of the
current page; before that the digits type. `Enter` commits the chosen candidate, or the
reading when none is chosen, as `0` always does; so does typing on, or any
key that ends the composition. `Backspace` deletes the last kana shown;
romaji that made no kana stays in the reading as typed until it is mended.
The dictionary keys a verb or adjective by its stem, so `kaku` offers the
readings of かく and then 書く and its kin; typing the okurigana with
`Shift`, as SKK does — `kaKu` — asks for that split first.  Caps Lock is
not `Shift` and marks nothing.
`Backspace` and `Esc` on a chosen candidate go back to the reading, and
`Esc` on the reading cancels it. Katakana mode does not perform Kanji
conversion.

Korean composes one syllable at a time; `Enter`, `Tab`, `Esc`, and any key
that is not a jamo commit the syllable and go on to the application, so
`Esc` still leaves insert mode. Two lone consonants join into the compound
final they make (rt → ㄳ) and a vowel typed before its consonant is
reordered under it (kr → 가), as in libhangul.

Emoji and Hanja searches take over the keys until `Enter`, `Space`, or
`1`-`9` picks a result or `Esc` cancels; they return to the previous mode
after use, and switching language during a search commits the shown
query. `Tab`/`Shift-Tab` wrap through the results, and a search with
nothing typed yet shows ☺ or 漢 in the popup. The Emoji search matches the
typed keys, and what they spell in the current language, against a prefix
of every emoji's CLDR name and keywords in English, Korean, and Japanese,
and against ASCII symbol aliases such as `->` and `<=`. `Ctrl+H` takes the
syllable being composed as its query and composes on from it; `Esc` gives
the syllable back. The Hanja dictionary converts a word as well as a
syllable, so `Ctrl+H` and then 한자 gives 漢字, and 대한민국 gives 大韓民國;
a syllable already committed is the application's text, not the engine's,
and cannot be converted. Leaving the field or
clicking elsewhere commits what is pending.

## Preedit and candidates

| Frontend | Preedit | Candidates |
| --- | --- | --- |
| Wayland input method | inline in the client | strans popup |
| GTK 3 and IBus | inline in the client | strans popup |
| XIM PreeditCallbacks | inline through XIM callbacks | strans popup |
| XIM PreeditPosition | strans popup | strans popup |
| XIM PreeditNothing | strans popup | strans popup |

Each XIM style is offered with StatusNothing and with StatusNone.

Dead keys and Compose sequences are composed by strans itself for the XIM
and IBus frontends, from the table `XCOMPOSEFILE` or the locale names; the
GTK 3 module leaves them to GtkIMContextSimple. Their text follows
whatever was pending.

XIM text travels as `COMPOUND_TEXT`, which carries any UTF-8. A client that
sends `XNSpotLocation` gets the popup under that spot whatever its preedit
style, and above the line when there is no room below; otherwise the popup
sits under the focus window, or under the client window when no focus
window is set.

On a compositor with `zwp_input_method_v2` the popup is a Wayland surface
that the compositor itself puts at the text cursor, flipping it above the
line when there is no room below; a client that reports no cursor location
gets it under its window instead. Everywhere else the popup is an X11 window,
placed from the XIM spot or from the GTK caret converted to root
coordinates, and XIM is X11-only. Either way `GDK_SCALE` sizes it for
HiDPI displays.

## Build and test

Initialize the bundled test library, then build in the Docker environment:

```sh
git submodule update --init
make docker-image
make docker-build
```

Build output:

```text
strans              daemon with Wayland, IBus, and XIM frontends
gtk/im-strans.so    GTK 3 frontend module
```

Tests have three tiers:

```sh
make check          # generated-map validation and focused unit tests
make check-live     # one IBus, GTK, XIM, and IPC daemon smoke
make check-stress   # randomized, capacity, collision, failure, and restart tests
```

`UNITARGS` filters only the focused C unit suite, for example
`make check UNITARGS=hangul`. It does not select or skip live and stress
tests. `make docker-check` rebuilds and runs all three tiers in the
container, and `make docker-valgrind` runs the unit suite under Valgrind
there; run `make docker-image` first after changing `Dockerfile`.

Native Linux source builds require a C toolchain, Make, and pkg-config, plus
development files for:

- Plan 9 port (`9c`, `9l`, `libthread`, and `libbio`);
- D-Bus development files;
- XCB, XCB RandR, XCB XKB, xcb-util, xcb-imdkit, xkbcommon, and xkbcommon-x11;
- the Wayland client library and `wayland-scanner`, which writes the
  input-method and virtual-keyboard bindings from the XML in `proto/`;
- Pango, PangoCairo, Cairo, and Fontconfig;
- GTK 3, plus libibus and Xlib for the live clients;
- Xvfb for the X11 live tests;
- Python 3 and `cmp` for `make check` and the map generators;
- a working `C.UTF-8` locale and fonts covering Latin, CJK, and emoji.

The package names used by the canonical Arch Linux image are listed in
[`Dockerfile`](Dockerfile).

## Run

The popup uses the normal system fonts found by Pango and Fontconfig. Run
it directly from the build tree:

```sh
./run.sh
# equivalently
./strans map
```

`run.sh` stops every same-user process whose exact name is `strans`, starts one
build-tree daemon in the background, and returns to the shell. It requires
`pkill` from procps. Startup and frontend failures remain visible on the
caller's standard error. A service supervisor should instead run
`./strans map` directly in the foreground, with the session's
`XDG_RUNTIME_DIR` and its `WAYLAND_DISPLAY` or `DISPLAY`: the GTK module
finds the daemon at `$XDG_RUNTIME_DIR/strans.sock` (else
`/tmp/strans.UID`), and IBus clients
through the address file libibus expects under `~/.config/ibus/bus/`,
which strans writes as `ibus-daemon` would, named after `WAYLAND_DISPLAY`
when there is one and `DISPLAY` otherwise.

The build does not copy or install the daemon or its data. `./strans DIR`
opens `hira.map`, `kata.map`, `telex.map`, `kanji.dict`, `emoji.dict`, and
`hanja.dict` from `DIR` at runtime; `run.sh` passes the tracked `map/`
directory. Only the GTK subdirectory has install and uninstall targets.

strans shows one popup per session, so it picks its frontend at startup: a
session whose compositor offers `zwp_input_method_v2` gets that frontend,
and neither XIM nor the X11 popup is started; every other session gets XIM
as before. The IBus endpoint and the GTK 3 module are served in both.

On Wayland, then, set none of the variables below. GTK, Qt and Firefox
speak text-input-v3 themselves and need nothing — GTK 4 binds it without
`GTK_IM_MODULE`, checked with 4.22 under sway 1.12 — and `GTK_IM_MODULE`
is what pushes them off the path that works. A client that reaches strans
over IBus or the GTK 3 module while the Wayland frontend runs still
composes inline, but shows no candidate list, because the popup belongs to
that frontend. Chromium is helped by neither: it still asks for
`text-input-v1`, which wlroots does not implement at all.

Configure clients as needed:

```sh
# XIM
export XMODIFIERS=@im=strans

# GTK 3 module
doas make -C gtk install
export GTK_IM_MODULE=strans

# IBus client example
GLFW_IM_MODULE=ibus kitty
```

Building `gtk/im-strans.so` requires GTK development files; installing one
built elsewhere, in the container for instance, does not.  The install
target uses `GTK_MODULE_DIR` when it is set, else GTK's pkg-config
variables, else the directory the installed `gtk-query-immodules-3.0`
reports.  Packagers and cross builds should set `GTK_MODULE_DIR`, and a
prebuilt module must still match the target architecture and library ABI.

`DESTDIR` staging is supported and deliberately skips the host module-cache
update. A native install needs the GTK runtime query utility so it can refresh
that cache. Remove a native installation with:

```sh
doas make -C gtk uninstall
```

strans provides its own IBus endpoint; `ibus-daemon` and fcitx are not
required. IBus clients that process keys synchronously — GTK 4, or GTK 3
with `IBUS_ENABLE_SYNC_MODE=1` — get their commits and preedits through
`PostProcessKeyEvent`, in order with the key. With neither
`WAYLAND_DISPLAY` nor `DISPLAY`, the daemon, its IBus endpoint and its
socket still work, but no frontend draws a popup.

## Dictionary data

After changing an input source, regenerate and verify the maps and
dictionaries:

```sh
python3 map/mktelex.py >map/telex.map
map/mkemoji >map/emoji.dict
map/mkhanja >map/hanja.dict
make verify-map
```

See [`map/README`](map/README) for the emoji, Hanja, and Japanese data and
where it comes from.

## Benchmark

`make bench` builds `bench/bench`. With the daemon stopped, `./bench.sh` starts
one build-tree instance, warms it up, and records the workload with Linux
`perf`. It requires `perf` and permission to profile the daemon, and writes
`bench/perf.data`.

## Licensing

Third-party license notices are in [`LICENSES`](LICENSES). The repository does
not currently declare a license for the strans source as a whole.