@commonpub/layer 0.24.0 → 0.25.1

package/composables/useLayoutEditor.ts

@@ -0,0 +1,593 @@
+/**
+ * useLayoutEditor — draft state for the /admin/layouts/[id] editor.
+ *
+ * Phase 3a.3. Owns:
+ *   - `draft`: editable copy of the layout (mutate freely; the editor
+ *     UI binds to this)
+ *   - `original`: last-saved snapshot; used to compute dirty state +
+ *     drives the If-Match header on save (3a.6)
+ *   - `dirty`: shallow-compared marker; true when draft !== original
+ *   - `save()` / `publish()` / `refresh()`: server interactions
+ *
+ * Auto-save (3a.6) plugs in by watching `dirty` + calling `save()`
+ * after a debounce. The composable itself is sync — auto-save is a
+ * separate composable to keep concerns clean.
+ *
+ * The composable expects to be created in an admin page context
+ * (`/admin/layouts/[id]`) — it doesn't ship feature-flag guards
+ * because the route + middleware already filter callers.
+ */
+import type { LayoutRecord } from '@commonpub/server';
+import { computed, ref, watch, type ComputedRef, type Ref } from 'vue';
+
+export interface LayoutEditorState {
+  /** The editable layout — mutate freely. */
+  draft: Ref<LayoutRecord | null>;
+  /** Last-saved snapshot, kept in sync with the server. */
+  original: Ref<LayoutRecord | null>;
+  /** True when draft has diverged from original. */
+  dirty: ComputedRef<boolean>;
+  /** True when a save is in flight. */
+  saving: Ref<boolean>;
+  /** Last save status — drives the toolbar indicator. */
+  status: Ref<'idle' | 'saving' | 'saved' | 'error' | 'conflict'>;
+  /** Last save error message (when status==='error'). */
+  errorMessage: Ref<string | null>;
+  /** Persist the draft. Throws on 409 (conflict — caller handles).
+   *  Pass `{ force: true }` to omit the If-Match header (overwrites
+   *  whatever's on the server). */
+  save: (opts?: { force?: boolean }) => Promise<void>;
+  /** Publish the current saved version → live. */
+  publish: () => Promise<void>;
+  /** Pull latest from server, replacing draft + original. */
+  refresh: () => Promise<void>;
+  /** Discard local changes (revert draft to original). */
+  discard: () => void;
+  /** Abort any in-flight save fetch — call from `onBeforeUnmount` to
+   *  prevent orphan PUTs from landing after the editor unmounts (which
+   *  could cause stale 409s when the user opens the editor again).
+   *  R4 P2 fix (session 161). */
+  abort: () => void;
+  /** Fire a best-effort PUT via `fetch({keepalive:true})` — survives
+   *  page teardown so unsaved edits don't vanish when the tab is
+   *  closed inside the 1.5s auto-save debounce window. Returns true
+   *  if a request was queued, false if the no-op preconditions held
+   *  (no draft/original, not dirty, or fetch unavailable). Fire-and-
+   *  forget; does NOT await. Bypasses the abort controller (the whole
+   *  point is for this request to outlive it).
+   *
+   *  Call from a `pagehide` listener (the only event that fires
+   *  reliably on mobile tab-close + bfcache eviction; `beforeunload`
+   *  doesn't run on iOS Safari at all). R2 P2 deferred (session 162). */
+  flushBeacon: () => boolean;
+  /** True when the recent conflict rate has exceeded the threshold
+   *  (3 conflicts within 60s). Wire this into useLayoutAutoSave's
+   *  `paused` prop so auto-save stops banging the server while the
+   *  user reconciles with the other editor. Cleared by
+   *  `clearConflictHistory()`. R2 P2 deferred (session 162). */
+  conflictThrashing: ComputedRef<boolean>;
+  /** Reset the conflict-window counter back to zero. Call after the
+   *  user has explicitly refreshed/force-saved + wants auto-save to
+   *  resume. */
+  clearConflictHistory: () => void;
+  /**
+   * Current selection — drives the right-hand inspector dispatcher.
+   *
+   *   `null`                    → render the page-meta form (default).
+   *   `{kind:'section', id:X}`  → render the section-config form (Phase 3f placeholder this session).
+   *   `{kind:'row', id:Y}`      → render the row-config form (Phase 3f placeholder this session).
+   *
+   * Session 3b/A landing: click / Enter on a section sets selection;
+   * click outside the canvas (or Esc) clears. Selection is intentionally
+   * NOT persisted across refresh/discard — those operations replace the
+   * draft wholesale, so the previously-selected id may no longer exist.
+   *
+   * See `feedback-visual-editor-ux-patterns` (inspector dispatch pattern)
+   * + docs/plans/layout-and-pages.md §7.9.
+   */
+  selectedId: Ref<EditorSelection>;
+  /** Set selection; pass `null` to clear. */
+  select: (selection: EditorSelection) => void;
+  /** Alias for `select(null)` — semantic sugar for click-outside / Esc. */
+  clearSelection: () => void;
+}
+
+/**
+ * Discriminated selection target. Encoded as an object (not a tuple) so
+ * inspector callers can branch on `kind` cleanly + so future kinds
+ * (e.g. `'zone'`) extend the union without breaking call sites.
+ *
+ * `null` = no selection (default; inspector renders page-meta form).
+ */
+export type EditorSelection =
+  | { kind: 'section'; id: string }
+  | { kind: 'row'; id: string }
+  | null;
+
+/**
+ * Discriminated failure for the multi-step publish() flow.
+ *
+ * publish() chains three independent server calls: save (only when
+ * dirty), POST /publish, then refresh. The R4 audit found that a
+ * generic "Publish failed" toast hides which step actually failed —
+ * the most important case being a publish-step failure AFTER a
+ * successful save, where the user's changes ARE durably saved as a
+ * draft but they don't know it.
+ *
+ * The consumer (editor page) catches `PublishStepError`, branches on
+ * `step`, and renders a step-specific toast. Session 162 P2.7.
+ */
+export type PublishStep = 'save' | 'publish' | 'refresh';
+
+export class PublishStepError extends Error {
+  readonly step: PublishStep;
+  override readonly cause: unknown;
+  constructor(step: PublishStep, cause: unknown) {
+    super(`Publish step "${step}" failed`);
+    this.name = 'PublishStepError';
+    this.step = step;
+    this.cause = cause;
+  }
+}
+
+/** Deep-clone via JSON. The LayoutRecord shape is JSON-safe (no Date/Map/Set). */
+function clone<T>(value: T): T {
+  return JSON.parse(JSON.stringify(value)) as T;
+}
+
+/**
+ * Stable-JSON stringify — key-order insensitive. Used only at SEED
+ * time (the first non-null assignment to draft) to decide whether to
+ * auto-mark pristine. The R2-audit perf concern was that this used to
+ * run per-keystroke via the old `dirty` computed; here it runs once
+ * per editor instance lifetime, so the O(N) cost is amortised.
+ */
+function stableString(value: unknown): string {
+  return JSON.stringify(value, (_key, val) => {
+    if (val && typeof val === 'object' && !Array.isArray(val)) {
+      const sorted: Record<string, unknown> = {};
+      for (const k of Object.keys(val as Record<string, unknown>).sort()) {
+        sorted[k] = (val as Record<string, unknown>)[k];
+      }
+      return sorted;
+    }
+    return val;
+  });
+}
+
+export function useLayoutEditor(id: string): LayoutEditorState {
+  const draft = ref<LayoutRecord | null>(null);
+  const original = ref<LayoutRecord | null>(null);
+  const saving = ref(false);
+  const status = ref<'idle' | 'saving' | 'saved' | 'error' | 'conflict'>('idle');
+  const errorMessage = ref<string | null>(null);
+
+  /**
+   * Dirty tracking — session 162 P2.6 (R2 audit).
+   *
+   * Was: `dirty = computed(stableString(draft) !== stableString(original))`.
+   * Walked the entire layout JSON on every keystroke. At N=50 sections
+   * the audit measured 5-10ms per keystroke; at N=200 it ate a frame
+   * (>16ms). Today's homepage at N=5 hid it.
+   *
+   * Now: O(1) version counters. `dirtyVersion` increments on every
+   * draft mutation via a deep watcher; `savedVersion` snapshots at
+   * save-success (or seed/refresh/discard); `dirty` is a single
+   * `dirtyVersion !== savedVersion` compare.
+   *
+   * The deep watcher subscribes to nested properties once at setup
+   * (using Vue's reactive Proxy infrastructure); subsequent mutations
+   * notify it in O(1) without re-walking. Outer ref reassignment
+   * (`draft.value = ...`) re-walks the new value — rare path (initial
+   * seed, refresh, discard).
+   *
+   * `flush: 'sync'` so dirty reflects the mutation in the SAME tick
+   * (lets test code read `dirty.value` immediately after a mutation
+   * without `await nextTick()`). Safe re-entrancy: the watcher only
+   * mutates dirtyVersion/savedVersion, never draft itself.
+   *
+   * Auto-sync on initial seed: the canonical consumer pattern is
+   * `editor.original.value = X; editor.draft.value = clone(X);` (the
+   * editor page does exactly this after useFetch lands). On that first
+   * draft assignment the watcher fires with oldDraft===null. We do a
+   * ONE-TIME stable-string compare against original.value — if equal,
+   * sync savedVersion so dirty starts false. If different, leave
+   * savedVersion alone so dirty starts true (covers tests that seed
+   * a divergent draft to assert dirty behavior).
+   *
+   * This single O(N) walk per editor lifetime replaces the previous
+   * O(N) walk per keystroke — the audit's actual cost concern.
+   */
+  const dirtyVersion = ref(0);
+  const savedVersion = ref(0);
+
+  watch(
+    draft,
+    (newDraft, oldDraft) => {
+      dirtyVersion.value++;
+      if (
+        oldDraft === null &&
+        newDraft !== null &&
+        original.value !== null &&
+        stableString(newDraft) === stableString(original.value)
+      ) {
+        savedVersion.value = dirtyVersion.value;
+      }
+    },
+    { deep: true, flush: 'sync' },
+  );
+
+  const dirty = computed<boolean>(() => {
+    if (!draft.value || !original.value) return false;
+    return dirtyVersion.value !== savedVersion.value;
+  });
+
+  /**
+   * Conflict-window tracking — session 162 P2.5 (R2 audit).
+   *
+   * Scenario: admin clicks "Reload" in the conflict modal; while their
+   * refresh is in flight a third admin saves; their next edit 409s
+   * immediately. Each conflict is a different third-party save (not a
+   * loop), but the UX thrashes — modal in/out/in. Worse, the auto-save
+   * watcher keeps re-triggering the save-then-409 cycle.
+   *
+   * Mitigation: after 3 conflicts within a 60s rolling window, flip
+   * `conflictThrashing` true. The editor page passes this to
+   * useLayoutAutoSave's `paused` prop, halting the auto-save cycle.
+   * A banner prompts the user to refresh / force-save / coordinate;
+   * `clearConflictHistory()` (UI: "Resume auto-save") clears the
+   * window and restarts. The 3-in-60s threshold matches what mature
+   * collab editors (Notion, Linear retros) settled on after tuning.
+   */
+  const CONFLICT_WINDOW_MS = 60_000;
+  const CONFLICT_THRESHOLD = 3;
+  const conflictHistory = ref<number[]>([]);
+  function recordConflict(): void {
+    const now = Date.now();
+    conflictHistory.value = [
+      ...conflictHistory.value.filter((t) => now - t < CONFLICT_WINDOW_MS),
+      now,
+    ];
+  }
+  function clearConflictHistory(): void {
+    conflictHistory.value = [];
+  }
+  const conflictThrashing = computed<boolean>(() => {
+    // The computed only re-evaluates when `conflictHistory.value`
+    // changes — Vue caches the last result otherwise. That means a
+    // rolling window doesn't naturally "expire" via wall-clock time;
+    // once `thrashing===true` is set, it stays true until either
+    //   (a) a new conflict event mutates conflictHistory (rare while
+    //       auto-save is paused — see the feedback loop below), or
+    //   (b) the user explicitly calls `clearConflictHistory()`.
+    //
+    // Feedback loop: thrashing=true → useLayoutAutoSave.paused=true →
+    // no new save attempts → no new conflicts → conflictHistory frozen
+    // → thrashing stays true. That matches the intended UX: the admin
+    // must explicitly acknowledge (Refresh / Force save / Resume)
+    // before auto-save resumes. A wall-clock timer would silently
+    // resume in the background, defeating the throttle's purpose.
+    const now = Date.now();
+    return conflictHistory.value.filter((t) => now - t < CONFLICT_WINDOW_MS).length
+      >= CONFLICT_THRESHOLD;
+  });
+
+  /**
+   * Selection state — drives the inspector dispatcher (Phase 3b/A).
+   *
+   * Held as a single ref containing a discriminated union. Cleared on
+   * refresh/discard because those operations replace draft wholesale +
+   * the selected id may no longer exist in the new draft. NOT cleared
+   * on save() — the server returns the same ids in the snapshot.
+   */
+  const selectedId = ref<EditorSelection>(null);
+  function select(selection: EditorSelection): void {
+    selectedId.value = selection;
+  }
+  function clearSelection(): void {
+    selectedId.value = null;
+  }
+
+  async function refresh(): Promise<void> {
+    const fresh = await $fetch<LayoutRecord>(`/api/admin/layouts/${id}`);
+    original.value = fresh;
+    // Replaces a non-null draft → the auto-sync-on-null path doesn't
+    // fire. Bump-and-sync explicitly so dirty starts false post-refresh.
+    draft.value = clone(fresh);
+    savedVersion.value = dirtyVersion.value;
+    status.value = 'idle';
+    errorMessage.value = null;
+    // The id that was selected may not exist in the fresh snapshot
+    // (another admin deleted it; we can't keep a phantom selection).
+    // Clear; the inspector falls back to the page-meta form.
+    selectedId.value = null;
+  }
+
+  /**
+   * In-flight save promise (single-flight guard — session 160 audit P1).
+   * Two distinct triggers can fire save() in parallel: manual Save click
+   * + auto-save timer, OR auto-save + visibility-flush. Without this
+   * guard, both calls pass the dirty check, both set saving=true, both
+   * send PUTs with the SAME (stale) If-Match. Server accepts both
+   * (last-write-wins). Client's `original.updatedAt` ends up referring
+   * to whichever response landed LAST in the await queue, which is not
+   * guaranteed to be the same as the actual DB row's updatedAt under
+   * network jitter — leading to spurious 409s on the next save.
+   *
+   * Pattern: store the in-flight promise; concurrent callers return
+   * the SAME promise (await it). Cleared in finally so the next save
+   * starts fresh.
+   */
+  let inFlightSave: Promise<void> | null = null;
+
+  /**
+   * AbortController for the editor's lifetime — aborted by abort() on
+   * unmount (called from the page's onBeforeUnmount). The single-flight
+   * guard above prevents parallel saves, so one controller suffices for
+   * the whole composable instance: after abort, no further saves should
+   * fire (the editor is unmounting). A new mount creates a new composable
+   * instance → new controller.
+   *
+   * The `typeof` guard makes this resilient to environments without
+   * the AbortController global (older Node test harnesses, etc).
+   * R4 P2 fix (session 161).
+   */
+  const abortController = typeof AbortController !== 'undefined'
+    ? new AbortController()
+    : null;
+
+  function abort(): void {
+    abortController?.abort();
+  }
+
+  /**
+   * Beacon save — session 162 P2.3.
+   *
+   * The auto-save composable's `visibilitychange` flush handles the
+   * common Cmd+Tab / minimize case via regular `$fetch`. That works
+   * because the page is still alive when the request goes out and the
+   * browser drains the connection normally. But for tab close (and
+   * especially iOS Safari, which doesn't fire `beforeunload` at all),
+   * the page is torn down before the request finishes — the abort
+   * controller cancels it, or the browser kills the network stack.
+   *
+   * `pagehide` fires reliably on every tab-close + bfcache event on
+   * every browser; combined with `fetch(..., { keepalive: true })` the
+   * browser commits to delivering the request even after the page is
+   * gone (subject to the 64KB body cap, which our LayoutPayload sits
+   * well under).
+   *
+   * sendBeacon would work too but is POST-only; we want PUT to share
+   * the existing endpoint + If-Match contract. The keepalive flag is
+   * the modern way to get the same lifecycle guarantee with arbitrary
+   * methods. Native `fetch` is used directly (not `$fetch`/ofetch)
+   * because we don't want response parsing, retries, or error wrapping
+   * — the page is gone, nobody can read the result.
+   */
+  function flushBeacon(): boolean {
+    if (typeof fetch === 'undefined') return false;
+    if (!draft.value || !original.value) return false;
+    if (!dirty.value) return false;
+
+    const body = JSON.stringify({
+      scope: draft.value.scope,
+      name: draft.value.name,
+      pageMeta: draft.value.pageMeta ?? undefined,
+      zones: draft.value.zones,
+      state: draft.value.state,
+    });
+    // Session 163 deep audit: the `keepalive: true` PUT has a 64KB body
+    // cap enforced by every browser. A complex layout (many sections
+    // with rich text + image URLs + visibility metadata) can exceed
+    // this. The browser silently drops the request → fire-and-forget
+    // becomes fire-and-vanish + the user's edits are lost. Surface
+    // false here so the caller (or the beforeunload prompt) is aware
+    // the beacon path can't carry this payload. Beforeunload still
+    // catches the user's intent to leave; the auto-save's pre-hide
+    // visibility-flush handles smaller payloads (no keepalive cap).
+    const BEACON_BODY_MAX_BYTES = 60 * 1024; // 60KB — 4KB headroom under the 64KB browser cap
+    if (body.length > BEACON_BODY_MAX_BYTES) {
+      return false;
+    }
+    const headers: Record<string, string> = {
+      'Content-Type': 'application/json',
+      'If-Match': original.value.updatedAt,
+      // Lets the server audit-log distinguish beacon saves from regular
+      // auto-saves (helpful when tracing "did my last edit land?").
+      'X-Cpub-Save-Source': 'beacon',
+    };
+    try {
+      // Fire-and-forget. No signal so the unmount abort() cannot cancel
+      // it. `keepalive:true` lets the browser deliver the request after
+      // page teardown — without it, the connection dies with the page.
+      void fetch(`/api/admin/layouts/${id}`, {
+        method: 'PUT',
+        headers,
+        body,
+        keepalive: true,
+      });
+      return true;
+    } catch {
+      // Swallow — the page is going away; the user isn't here to see
+      // an error, and the fetch is best-effort anyway.
+      return false;
+    }
+  }
+
+  async function save(opts: { force?: boolean } = {}): Promise<void> {
+    if (!draft.value || !original.value) return;
+    if (!dirty.value) return;
+    // Single-flight: if a save is already in flight, coalesce — return
+    // the in-flight promise instead of starting a parallel request.
+    // Subsequent saves are picked up by the auto-save watcher when the
+    // first one completes (dirty stays true if the user kept editing).
+    if (inFlightSave) return inFlightSave;
+
+    saving.value = true;
+    status.value = 'saving';
+    errorMessage.value = null;
+    // Capture the original.updatedAt BEFORE the request — used as the
+    // If-Match value. The server's response will give us a fresh
+    // updatedAt to use for the next save's optimistic-concurrency check.
+    const ifMatch = opts.force ? undefined : original.value.updatedAt;
+    // Capture dirtyVersion AT SAVE START so any edits the user makes
+    // during the await leave `dirty` true post-save (their unsaved
+    // newer edits survive). Session 162 P2.6.
+    const savingVersion = dirtyVersion.value;
+
+    inFlightSave = (async () => {
+      try {
+        const headers: Record<string, string> = {};
+        if (ifMatch) headers['If-Match'] = ifMatch;
+        // Signal a deliberate force-save so the server can audit-log
+        // it distinctly from first-creation (both share no-If-Match).
+        if (opts.force) headers['X-Cpub-Force-Save'] = '1';
+        const updated = await $fetch<LayoutRecord>(`/api/admin/layouts/${id}`, {
+          method: 'PUT',
+          headers,
+          body: {
+            scope: draft.value!.scope,
+            name: draft.value!.name,
+            pageMeta: draft.value!.pageMeta ?? undefined,
+            zones: draft.value!.zones,
+            state: draft.value!.state,
+          },
+          // Cancel the fetch if the editor unmounts mid-save. Catch block
+          // below recognises the AbortError and short-circuits without
+          // surfacing it as a user-visible error.
+          signal: abortController?.signal,
+        });
+        // Update `original` only — DON'T overwrite `draft`. The user may
+        // have made further edits while the save was in flight; those
+        // edits stay in draft + the dirty comparison correctly flips
+        // true again so the auto-save composable schedules a follow-up.
+        // The server returns the saved snapshot which becomes the new
+        // baseline for If-Match.
+        original.value = updated;
+        // Mark the version we sent as durable. If the user made edits
+        // during the await, dirtyVersion > savingVersion → dirty stays
+        // true → auto-save schedules a follow-up. If no edits during
+        // await, dirtyVersion === savingVersion → dirty becomes false.
+        //
+        // Session 164 audit-of-audit P3 fix — savedVersion is monotonic:
+        // discard()/refresh() also write to savedVersion (syncing it to
+        // the post-replacement dirtyVersion), which can leave it AHEAD
+        // of the savingVersion captured at save start. Without this
+        // guard, save's completion would move savedVersion BACKWARD,
+        // spuriously flipping `dirty` true post-discard. Concrete trace:
+        //
+        //   t=0   edit (dirtyVersion=1, savedVersion=0)
+        //   t=1500 save fires (savingVersion=1, in flight)
+        //   t=1800 discard (draft=original; dirtyVersion=2; discard sets
+        //          savedVersion=dirtyVersion=2 → dirty=false)
+        //   t=2000 save completes
+        //          old: savedVersion = savingVersion = 1 → dirty = (2!=1)=true (WRONG)
+        //          new: savingVersion=1 NOT > savedVersion=2; skip → dirty=false ✓
+        //
+        // Server-state divergence (server saved pre-discard edit while
+        // client discarded) still exists; abort-on-discard would close
+        // it but requires more surface area. Defer; the dirty-flag UX
+        // fix here is the high-leverage piece.
+        if (savingVersion > savedVersion.value) {
+          savedVersion.value = savingVersion;
+        }
+        status.value = 'saved';
+      } catch (err) {
+        const e = err as { statusCode?: number; statusMessage?: string; message?: string; name?: string };
+        // AbortError: the editor unmounted mid-save (the user navigated
+        // away). The component is gone — surfacing "Save failed" is
+        // wrong (the user isn't here to see it, and the fetch was
+        // CANCELLED, not failed). Reset to idle + re-throw so the
+        // outer caller knows the promise didn't complete normally,
+        // but skip the user-visible status/error mutations.
+        if (e.name === 'AbortError') {
+          status.value = 'idle';
+          errorMessage.value = null;
+          throw err;
+        }
+        if (e.statusCode === 409) {
+          status.value = 'conflict';
+          errorMessage.value = 'Another admin edited this layout while you were working.';
+          // Session 162 P2.5: feed the rolling-window throttle. A single
+          // 409 is normal collab; 3 within 60s is thrashing → pause
+          // auto-save so the user reconciles before the next round trip.
+          recordConflict();
+        } else {
+          status.value = 'error';
+          errorMessage.value = e.statusMessage ?? e.message ?? 'Save failed';
+        }
+        throw err;
+      } finally {
+        saving.value = false;
+        inFlightSave = null;
+      }
+    })();
+
+    return inFlightSave;
+  }
+
+  async function publish(): Promise<void> {
+    if (!draft.value) return;
+
+    // Save first if dirty — publish snapshots the LAST SAVED state.
+    // Session 162 P2.7: wrap each step in its own try/catch so the
+    // caller can render a step-specific failure toast. The most
+    // important case is a 'publish' step failure AFTER a successful
+    // 'save' — the user's changes are durably saved as a draft but
+    // the generic "Publish failed" toast hid that.
+    if (dirty.value) {
+      try {
+        await save();
+      } catch (err) {
+        throw new PublishStepError('save', err);
+      }
+    }
+
+    try {
+      await $fetch(`/api/admin/layouts/${id}/publish`, { method: 'POST' });
+    } catch (err) {
+      throw new PublishStepError('publish', err);
+    }
+
+    try {
+      await refresh();
+    } catch (err) {
+      // Server data IS published. Only the editor's local view is
+      // stale. The page handler can recommend a manual reload.
+      throw new PublishStepError('refresh', err);
+    }
+  }
+
+  function discard(): void {
+    if (!original.value) return;
+    draft.value = clone(original.value);
+    // Same as refresh — non-null replacement. Sync explicitly.
+    savedVersion.value = dirtyVersion.value;
+    status.value = 'idle';
+    errorMessage.value = null;
+    // Same rationale as refresh — the discarded draft's selected id may
+    // not exist in the original snapshot if the user added a section
+    // then discarded.
+    selectedId.value = null;
+  }
+
+  return {
+    draft,
+    original,
+    dirty,
+    saving,
+    status,
+    errorMessage,
+    save,
+    publish,
+    refresh,
+    discard,
+    abort,
+    flushBeacon,
+    conflictThrashing,
+    clearConflictHistory,
+    selectedId,
+    select,
+    clearSelection,
+  };
+}