Migemo Search Feature

Overview

Migemo lets incremental search find Japanese names by typing romaji — no IME needed. Typing kensaku also matches 検索, けんさく, ケンサク, ｹﾝｻｸ, and ｋｅｎｓａｋｕ; typing mudai finds 無題.png, and daunro-do finds ダウンロード (typed hiragana finds katakana too). It is on by default and needs no mode or switch: plain matching always still applies, Migemo only ever adds matches.

Where it works

Migemo applies to every incremental-search surface:

• File pane incremental search (F) — each space-separated token matches by substring or by Migemo, so tenki yohou finds 天気予報.csv.
• Text viewer search (F inside the viewer) — matching lines are highlighted, including the Japanese text a romaji pattern matched.
• Diff viewer search — both sides of every row.
• Rich view modes (Markdown / JSON / CSV) — the embedded renderer’s search uses the same matcher.
• Filter-list dialogs — favorites, history, drives, external programs: the filter field takes the same query as the pane (space-separated tokens, wildcards), so typing romaji finds Japanese labels.

How patterns behave

• Plain patterns (kensaku, TenkiYohou, KENSAKU) match by substring and by Migemo. Capitalization doesn’t matter; camel case splits words, so TenkiYohou finds 天気予報.
• Glob patterns — anything containing *, ?, or [ — keep their exact wildcard behavior and skip Migemo entirely. *.py matches only .py files.
• Short patterns (fewer than 3 characters by default) match by substring only. Two romaji characters are barely one kana, and expanding them is slow, so Migemo waits for the third character.
• Filenames macOS stores in decomposed form (NFD) match correctly.

AZIK

If you type Japanese with AZIK — the extended romaji scheme where a second stroke below the vowel spells the kana plus “n” — set:

MIGEMO_ROMAJI_TABLE = 'azik'

Search then also understands the spellings AZIK trained your fingers on:

So kdsaku finds 検索, kzxa finds 感謝, and se;kw finds 設計.

AZIK is added alongside plain romaji, never in place of it: the handful of keys the two spell differently (ca is か in plain romaji and ちゃ in AZIK) match either reading, so turning this on cannot make a search you already use stop working.

Configuration

# In ~/.xefm/config.py
MIGEMO_SEARCH = True   # Add Migemo (romaji -> Japanese) matches to incremental search
MIGEMO_MIN_LENGTH = 3  # Shortest pattern handed to Migemo
MIGEMO_ROMAJI_TABLE = 'default'  # 'default' or 'azik'

Set MIGEMO_SEARCH = False to turn the feature off — searches then behave exactly as before. Raise or lower MIGEMO_MIN_LENGTH to move the point where Migemo kicks in (1 sends even single characters through Migemo; expect a noticeable pause on the first keystroke).

The gate also decides how soon incremental search refuses a character that matches nothing. A romaji reading finds Japanese only once it reaches MIGEMO_MIN_LENGTH, so a shorter pattern is never treated as a dead end — unless the directory holds no Japanese name at all, where there is nothing for Migemo to find and the search stops on the first miss like any other.

Requirements

Migemo uses the pymigemo package, installed automatically with XeFM (its dictionary is bundled — about 1.4 MB, loaded lazily on first use). If the package is missing, searches quietly fall back to plain matching; nothing breaks.