[PCMan]

DRAFT. Rework needed:

• init now blocks until the cin table loads (defeats the loader thread; stalls the whole backend; can trip the 15 s launcher watchdog). Use a threading.Event or keep loading async.
• Lock must be created in the class body, not via hasattr() (racy). Move imports to module level.
• Only 4 of 8 cin IMEs were changed (checj, chearray, chedayi, chephonetic untouched). Move the logic into cinbase once.
• CinBaseConfig.class() is a workaround for config.py rebinding the class name to an instance; export the class properly. Remove now-unused import copy.
• getKeyState() now returns bool; audit callers in cinbase/init.py.

Created by Antigravity + Claude + Gemini

[asdf80730]

Triaged this WIP branch (fix/cinbase-backends) against today's master — status per commit:

• Busy-wait removal (6741a66): superseded by feat(cinbase): asynchronous table loading with Future and CinTableBase #1004, which does the same job on a Future-backed CinTableBase, covers all 8 cin IMEs, and ships tests — which also resolves the "only 4 of 8 IMEs" and "keep loading async" items in this PR's own description.
• Key-state masks (from d0fa7e4): already on master via 78d2ca8 (KEY_STATE_TOGGLED / KEY_PRESSED_MASK).
• Per-IME config isolation (from d0fa7e4): still unfixed on master — all 8 backends still call CinBaseConfig.__init__() on the shared singleton and copy.deepcopy() it, which resets the object configtool reads and races between the reset and the copy. Salvaged into fix(cinbase): give each input method its own config instance #1014, which exports the class properly (no __class__() workaround), covers all 8 IMEs, and drops the deepcopy.

Once #1004 and #1014 are in, this PR has nothing left it uniquely carries — consider closing it and deleting fix/cinbase-backends so the branch stops drifting.

(Comment generated with assistance from omp (Qwen3.8-Flash-Next))