Docker build cache: .dockerignore, cache mount, layer reorder, restart script

- .dockerignore: exclude node_modules/.git/tmp/data/logs
- Dockerfile: BuildKit cache mount for npm, --omit=dev at install, drop
  npm prune step
- Reorder layers: shared/src/frontend copy before build, server copy +
  better-sqlite3 rebuild after. Server/shared changes no longer invalidate
  frontend build layer unless shared (FE dep) changes
- Remove duplicate shared COPY in runtime stage (bundled in FE build)
- docker/restart.sh: stop + rebuild + start in one script

.dockerignore

@@ -0,0 +1,11 @@
+node_modules
+**/node_modules
+.git
+tmp
+data
+*.log
+npm-debug.log*
+.DS_Store
+.env*
+coverage
+.nyc_output

TODO.md

@@ -65,25 +65,6 @@ not sure good way to do this
 - Ambiguous label. May expand work based on what NPC means here (ally? monster?
   display-only? skip in turn order?). Clarify intent before UX changes.
 
-### FEAT: first-class undo/redo UI buttons (B --- do now)
-- Toolbar buttons ↶/↷ in AdminView header, not buried in /logs.
-- Undo = revert latest non-undone log. Redo = re-apply latest undone.
-- Uses current 2-write undo (non-tx). Race safety = log refactor later.
-- Disabled when stack empty. Keyboard shortcuts (cmd+z / cmd+shift+z).
-
-### FEAT-LOG: unified log refactor (was FEAT-2 + M6, batched)
-- Single event schema, one source of truth:
-  `{ ts, type, payload, undo_payload, undone, encounterId,
-     snapshot:{ round, currentTurnParticipantId, turnOrderIds, activeIds } }`
-- Common format consumed by: UI log view, download/copy export,
-  replay-combat, analyze-turns. One shape, four consumers.
-- Transactional undo: server endpoint `POST /api/undo/:eventId`. Single
-  SQLite tx applies undo_payload + flips `undone`. Replaces fragile 2-write
-  (log update + encounter update as separate calls).
-- Download/copy: exports event stream as JSON for offline analysis.
-- replay-combat + analyze-turns rewritten to emit/consume same event shape.
-- Migration: keep old log entries readable; new format for new writes.
-
 
 ### quality of life fix: 2. UI says "Campaign Characters", field is players --- naming mismatch (separate concern, flag for later)
 
@@ -95,133 +76,3 @@ not sure good way to do this
 ## FEAT - clarify what end encounter does and what initiatives reset means
 
 ## Done (history)
-
-### FEAT: first-class undo/redo UI buttons (DONE)
-- ↶/↷ pills in InitiativeControls, always visible when encounter open.
-- Undo = latest non-undone log (per encounter). Redo = latest undone.
-- encounterPath added to all 14 log contexts (filter key).
-- redo:patch (forward) added to undoData. Real redo replays forward state.
-- Disabled when stack empty. Tooltip shows target action.
-- Uses current 2-write undo (non-tx). Race safety = FEAT-LOG refactor.
-
-### Architecture: 1-list turn order model (slot, never sort)
-- Single source: turnOrderIds === participants.map(id). No re-sort after
-  startEncounter. nextTurn skips inactive (predicate), inactive stay in slot.
-- Drag (reorder) = same-init tie-break only. Cross-init blocked.
-- startEncounter sorts ALL participants by init once, then frozen.
-- addParticipant/updateParticipant slot by init (slotIndexForInit), preserve
-  drag order. Display renders participants[] directly (no sort).
-- Static guard test errs if `.sort(` reintroduced outside allowlist.
-- Design doc: docs/INITIATIVE_ORDERING.md.
-
-### Single source of truth: combat logic
-- All 15 App.js handlers delegate to @ttrpg/shared. ~498 lines inline dupes deleted.
-- shared/turn.js = only place turn logic lives.
-
-### Storage parity (firebase + server adapters)
-- Neutral queryConstraints ({__type:'orderBy'|'limit'}) honored by both adapters.
-- Shared contract test runs both identically. Memory adapter deleted; factory
-  throws on unknown mode.
-- ws storage mode renamed to server (env var + adapter name).
-
-### Logging contract
-- Every mutating op logs message + undo payload. No-op = null log.
-- Structural enforcement: per-op contract test (turn.logging.test.js) +
-  static source-scan guard (static.no-unlogged.test.js).
-
-### Custom conditions per campaign (DONE)
-- Freeform per-campaign conditions. Add applies to participant + persists to
-  campaign palette in one step. Badge render uses merged allConditions
-  (built-ins + custom). toggleCondition accepts any string. Dedup
-  case-insensitive. maxLength 40. Combat + replay tests prove arbitrary
-  string ids survive round-trip.
-
-### Death saves: D&D 5e status model (DONE)
-- `status` is source of truth: conscious, dying, stable, dead.
-- Characters and NPCs use death saves; monsters skip death saves and become dead/inactive at 0 HP.
-- Death-save actions: Success, Fail, Nat1, Nat20, Stabilize.
-- Revive: dead → 0 HP, stable/unconscious, active.
-- Dead characters/NPCs stay in encounter/initiative until DM removes or marks inactive.
-- Player display hides inactive participants; DM display keeps them visible.
-
-### FEAT-3: initiative first-class entry (DONE)
-- Initiative field at add-char, add-monster, edit participant.
-- Inline edit wired. Tie-break = drag order.
-
-### UI feedback: toast + info modal (DONE)
-- All 23 native alert() replaced. ToastStack (6s auto-dismiss + manual X)
-  for transient failures. InfoModal (persistent OK) for validations.
-  React context provider wraps all 3 App branches.
-- Fixed: native alert vanished instantly on browser focus loss.
-
-### Filter dup chars from add-participant dropdown (DONE)
-- Character dropdown excludes chars already in encounter. Prevents
-  dup-add at source. No more dup alert path needed.
-
-### Test timeouts (DONE)
-- jest.setTimeout(10000) in setupTests.js (CRA blocks config-level timeout).
-
-### Warning = failure in tests (DONE)
-- console.error/warn throw in test env.
-
-### BUG-1: addParticipant + pause/resume corrupts rotation
-- RESOLVED as side effect of BUG-2 fix.
-
-### BUG-2: addParticipant allows duplicate id
-- FIXED (addParticipant throws on dup id).
-
-### BUG-4: hide-player-HP breaks display view
-- FIXED --- mock honors setDoc{merge}, all 5 activeDisplay sites use merge.
-
-### BUG-5: mid-round addParticipant/revive corrupts rotation
-- FIXED --- slot-array + DRY advance core nextActiveAfter.
-
-### BUG-6: reorderParticipants doesn't update turnOrderIds
-- FIXED structurally by 1-list model.
-
-### BUG-7: reorderParticipants not logged
-- FIXED --- returns log:{message, undo}. Handler calls logAction. deathSave,
-  addParticipants, updateParticipant logging gaps also closed.
-
-### BUG-8: server adapter has no reconnect
-- FIXED --- onclose reconnects + re-subscribes existing paths.
-
-### BUG-10: deact+reactivate same round double-acts participant
-- FIXED --- 1-list model keeps slot position on toggle. Reactivate does not
-  grant second turn. Test: turn.bug10.test.js.
-
-### BUG-11: FE Combat.scenario test crashes
-- FIXED --- moved to shared/turn.combat.test.js, pure functions, 100 rounds.
-
-### BUG-12: campaign selection follows activeDisplay
-- FIXED.
-
-### BUG-13: reorderParticipants crossing current pointer = ambiguous
-- FIXED --- block cross-pointer reorder during active encounter (both dirs).
-  Full fix needs actedThisRound tracking. Pragmatic block prevents skip/double.
-  Pre-combat: free reorder. Test: turn.bug13.test.js.
-
-### BUG-14: addParticipant init-insertion breaks after drag-reorder
-- FIXED --- slotIndexForInit scans current list (post-drag aware).
-
-### BUG-15: DisplayView re-sorts (drag order not preserved)
-- FIXED --- display renders participants[] directly.
-
-### BUG-16: subscribeCollection hook drops queryConstraints
-- FIXED --- neutral builders, both adapters honor orderBy/limit.
-
-### BUG-17: dead SDK imports in App.js
-- FIXED --- trimmed (auth + getFirestore + getStorage remain).
-
-### BUG-18: stale comments reference deleted memory adapter
-- FIXED.
-
-### FEAT-1: Dead characters/NPCs stay in turn order
-- DONE --- character/NPC death does not auto-remove. DM controls inactive/remove.
-- Non-NPC monsters still become inactive automatically at death.
-
-### combat.scenario 100 rounds not turns
-- DONE --- loops by actual round-wrap count.
-
-### feat: add all characters to participants list
-- DONE --- addParticipants bulk add wired.

docker/Dockerfile

@@ -1,4 +1,5 @@
-# docker/Dockerfile — single container: caddy (front) + node (back).
+# syntax=docker/dockerfile:1
+# docker/Dockerfile --- single container: caddy (front) + node (back).
 # Build context = repo root.
 # ---- build stage: frontend + install backend deps ----
 FROM node:18-alpine AS build
@@ -7,25 +8,26 @@ WORKDIR /app
 COPY package*.json ./
 COPY shared/package.json ./shared/
 COPY server/package.json ./server/
-RUN npm install --include-workspace-root
+RUN --mount=type=cache,target=/root/.npm \
+    npm install --include-workspace-root --omit=dev --prefer-offline
 
+# frontend code (changes often)
 COPY shared/ ./shared/
-COPY server/ ./server/
 COPY src/ ./src/
 COPY public/ ./public/
 COPY tailwind.config.js postcss.config.js ./
 
-# better-sqlite3 native build (alpine musl)
-RUN cd server && npm rebuild better-sqlite3
-
 # build frontend (server storage, same-origin /api + /ws via caddy)
 ARG REACT_APP_TRACKER_APP_ID=ttrpg-initiative-tracker-default
 ENV REACT_APP_STORAGE=server
 ENV REACT_APP_TRACKER_APP_ID=$REACT_APP_TRACKER_APP_ID
 RUN NODE_OPTIONS=--openssl-legacy-provider npm run build
 
-# prune backend dev deps for runtime
+# server after backend - it's fast. so changing server only doesnt invalidate slow frontend.
-RUN npm prune --omit=dev
+COPY server/ ./server/
+# better-sqlite3 native build (alpine musl). Cached if server unchanged.
+RUN cd server && npm rebuild better-sqlite3
+
 
 # ---- runtime stage: caddy + node ----
 FROM node:18-alpine
@@ -38,7 +40,6 @@ COPY --from=build /app/server/node_modules ./server/node_modules
 COPY --from=build /app/package*.json ./
 COPY --from=build /app/shared/package.json ./shared/
 COPY --from=build /app/server/package.json ./server/
-COPY shared/ ./shared/
 COPY server/ ./server/
 # built frontend served by caddy
 COPY --from=build /app/build /srv

docker/restart.sh

@@ -0,0 +1,21 @@
+#!/usr/bin/env bash
+# scripts/docker-restart.sh — rebuild + restart docker container.
+# Usage: ./scripts/docker-restart.sh
+set -euo pipefail
+cd "$(dirname "$0")/.."
+
+echo "=== stopping existing container ==="
+docker compose -f docker/docker-compose.yml down 2>/dev/null || true
+
+echo "=== rebuilding image ==="
+docker compose -f docker/docker-compose.yml build
+
+echo "=== starting container ==="
+docker compose -f docker/docker-compose.yml up -d
+
+echo "=== status ==="
+docker compose -f docker/docker-compose.yml ps
+
+echo ""
+echo "app: http://localhost:${PORT:-8080}"
+echo "logs: docker compose -f docker/docker-compose.yml logs -f"

server/index.js

@@ -1,6 +1,7 @@
 // server/index.js — generic KV document store over HTTP + WebSocket.
 // firebase mirror: doc-tree model. Thin REST, path-based WS push.
 // Adapter (src/storage/server.js) = passthrough, no shape translation.
+// TEST: cache layer rebuild check.
 
 'use strict';