159bb9d70a
## RecordTable Performance Investigation & Optimization (WIP) Investigates what makes the RecordTable slow (14 components, ~12 hooks per cell) and starts applying fixes. ### Key Findings (2,000 cells benchmark)   - **Jotai atoms are the dominant cost**: 10 atom reads/cell = +312%. Full sim with atoms = +476%. - **Derived atoms are 3x cheaper** than individual reads (12 sources: +93% vs +294%). - **Component depth is expensive**: 4-level nesting = +82%, 14 wrappers = +109%. - **Styling engines are comparable**: Linaria vs Emotion is within noise. - **Context reads and useState are nearly free** vs baseline. ### Optimizations Applied 1. **Static focus providers** — replaced per-cell `useState(false)` with static context. Eliminates 400 useState instances. 2. **Delegated onMouseMove** — single handler on table body instead of 400 per-cell handlers. 3. **Hoisted `useObjectMetadataItems()`** — moved from per-cell to table-level context. Eliminates 400 global atom reads. ### Tooling - Perf page at `/__perf__/table`: 17 cell render + 13 state access benchmarks - Render profiler: `window.__RECORD_TABLE_PROFILE = true` - Full plan in `__perf__/PERFORMANCE_PLAN.md` ### Remaining Phases (not high priority to-be-honest) | Phase | What | Status | |-------|------|--------| | 1 | Separate display from interaction | Partial | | 2 | Flatten hierarchy (14 → ~5 components/cell) | TODO | | 3 | Reduce atom reads per cell | Partial | | 4 | CSS-only hover/focus | TODO | | 5 | Event delegation, lazy Draggable | TODO |