Perf: edit pipeline node for zero-remap picks (Step 2, issue #34)

CLAUDE.md

@@ -131,7 +131,7 @@ If the current branch was cut from a feature branch (not from `main`), open the
 
 When opening or merging a PR, add a summary entry to `MIGRATION_LOG.md` under `## Done` (newest first).
 
-Before opening a PR, check [TODO.md](TODO.md) for related open items to update or remove. Also check and update [docs/logic_flows.md](docs/logic_flows.md) if any execution paths changed.
+Before opening a PR, check [TODO.md](TODO.md) for related open items to update or remove. Also check and update [docs/logic_flows/](docs/logic_flows/) if any execution paths changed (edit_session.md for edit flows, picking.md for pick flows, etc.).
 
 ## Current Architecture
 
@@ -148,7 +148,7 @@ Before opening a PR, check [TODO.md](TODO.md) for related open items to update o
 
 ### ParaView Backend Path
 
-- `paraview_backend.py`: owns ParaView sources/displays, pipeline nodes, filters, coloring, display controls, picking, selection overlays, saving/export. `set_edit_target_dataset()` builds four remap caches at session begin (`_edit_source_dataset`, `_edit_source_point_indexes`, `_edit_target_cell_map`, `_edit_point_id_map`) so picks avoid per-pick Fetch/rebuild — see `docs/logic_flows.md §5b`.
+- `paraview_backend.py`: owns ParaView sources/displays, pipeline nodes, filters, coloring, display controls, picking, selection overlays, saving/export. At edit session begin, `export_active_dataset_for_editing()` creates a real pipeline node (kind="edit") that reads the same backing data as the original node, so pick cell IDs directly index `working_dataset` with no remapping — see [docs/logic_flows/edit_session.md](docs/logic_flows/edit_session.md).
 - `paraview_runtime.py`: synchronizes backend state into Trame state, handles render pushes, edit-session overlay sync. Forwards event normalization calls to `paraview_event_utils.py`.
 - `paraview_event_utils.py`: pure helpers for normalizing ParaView picking event payloads (`normalize_edit_selection_ids`, `summarize_edit_event`, coordinate mapping). No Trame dependency.
 - `paraview_controllers.py`: user actions from the UI: pipeline actions, filters, edit sessions, selection, field creation, display/color controls.
@@ -263,7 +263,7 @@ conda run -n coral-paraview pytest -q tests/ --ignore-glob=tests/test_e2e*.py
 
 ## Debugging Notes
 
-- `docs/logic_flows.md` has detailed flow diagrams.
+- [docs/logic_flows/](docs/logic_flows/) has detailed flow diagrams split by domain (`edit_session.md`, `picking.md`, `color.md`, etc.).
 - If `pytest` or `python` resolves to the wrong environment (ParaView unavailable): a stale virtualenv may be active and its `PATH` entry wins. Run `deactivate` if a venv is active (`deactivate` is only defined while a venv is sourced — if not found, no venv is active), then `conda activate coral-paraview`.
 - E2E test meshes live in `test_data/`; use them instead of writing into `data/` unless needed.
 - `tests/test_e2e_edit_selection_playwright.py` has helpers for normalized box drags and switch state checks.

MIGRATION_LOG.md

@@ -1,11 +1,38 @@
-# CODEX_LOG
+# MIGRATION_LOG
 
 ## Goal
 
 Evolve the current VTK/trame viewer toward a ParaView-backed application while keeping the existing project aesthetic. The VTK→ParaView migration is complete. For current architecture see [CLAUDE.md](CLAUDE.md); for active work see [TODO.md](TODO.md).
 
 ## Done
 
+- Edit pipeline node for zero-remap picks (Step 2, issue #34, branch `perf/step2-edit-node`):
+  - replaced the per-pick remap-cache approach (Step 1, commit `298a3df`) with a real
+    pipeline node (kind="edit") that reads the same backing data as the original node,
+    so pick cell IDs directly index `working_dataset` with no coordinate-based remapping
+  - a dedicated reader proxy approach (originally planned for Step 2) was explored but
+    dismissed: it caused regressions in coloring, LUT, and representation handling because
+    those code paths all expect a regular pipeline node; a real node fits the existing
+    architecture with zero property overrides and no special-casing
+  - removed all four remap-cache fields (`_edit_source_dataset`, `_edit_source_point_indexes`,
+    `_edit_target_cell_map`, `_edit_point_id_map`) and five remap methods
+  - edit node label "✏ Editing: {original}", icon `mdi-pencil-outline`, visible in pipeline panel
+  - root readers reuse the same backing file (no write); filter nodes write a temp VTU
+  - `can_edit_active_node()` replaces the expensive export probe with a cheap `GetDataInformation` call
+  - surface picks skip a redundant `Fetch(source)` during an active session (use `working_dataset` directly)
+  - cached surface-pick geometry at session begin (B5, B5b): `set_edit_target_dataset` builds
+    `_edit_source_point_indexes` (O(n_points) coordinate index) and `_edit_boundary_elements`
+    (O(n_cells×faces) boundary map) once;
+  - the pipeline node model is also a better foundation for future improvements: real-time
+    display updates when cell values change (update the edit node in place), and server-side
+    field-value mutations via ParaView proxies to preserve MPI distribution without collapsing
+    all ranks through `Fetch()`
+  - `docs/logic_flows/` (split from monolithic `docs/logic_flows.md`), `CLAUDE.md`, and `TODO.md` updated
+
+- Eliminate per-pick Fetch and map rebuild with edit session remap cache (Step 1, issue #34, commit `298a3df`):
+  - pre-built four coordinate-based maps at session begin to avoid O(n) rebuild on every pick
+  - subsequently superseded by Step 2 above (edit node approach makes remapping unnecessary)
+
 - UI restructure, centralized notifications, save dialog, and categorical LUT fixes (PR #32):
   - moved edit-panel controls into the right inspector Edit tab; added `docs/ui_structure.md`
   - replaced 9 scattered alert state pairs with a single `VSnackbar` driven by `notifications.py`

TODO.md

@@ -2,11 +2,6 @@
 
 ## Priorita alta
 
-- **Performance selezione su dataset grandi** — bottleneck documentati in
-  `docs/logic_flows.md` §5a e §6. Piano a step progressivi in [issue #34](https://github.com/2listic/coral-visualizer/issues/34); ogni step
-  è indipendente e lascia i test esistenti verdi. Step 1 può essere fatto direttamente
-  su `paraview_backend.py` senza refactor preventivo.
-
 - Separare `paraview_backend.py` in servizi di dominio più piccoli (3900 righe).
   Iniziare da moduli con dipendenze unidirezionali (`backend/display.py`,
   `backend/coloring.py`, `backend/pipeline.py`) — più self-contained e a minor rischio