@cleocode/core 2026.9.20 → 2026.9.22

package/dist/store/twin-collapse.js

@@ -0,0 +1,931 @@
+/**
+ * Twin collapses (T12535): keep a prefixed twin in step with its BARE legacy
+ * table, which stays authoritative while an older CLEO build still writes it.
+ *
+ * ## Why
+ *
+ * The project `cleo.db` carries legacy bare tables beside their prefixed (E6)
+ * twins. For `schema_meta` and `sticky_tags` the runtime wrote the bare table
+ * while the twin held a frozen exodus copy. This build reads and writes only
+ * the twin. The release before it (2026.9.20) is still installed on machines
+ * that share project stores and keeps writing the bare tables, so the collapse
+ * runs at every open:
+ *
+ * - **Initial collapse** (no marker yet): check the free space, snapshot the
+ *   store, then fold the bare table into the twin, bare-authoritative.
+ * - **Incremental re-merge** (marker present): carry over only the bare keys
+ *   (sticky ids) whose hash changed since the last merge. No snapshot. An
+ *   unchanged bare table costs one read of it and no write.
+ *
+ * A later release removes the incremental re-merge once no supported build
+ * writes the bare tables; slice 3 then drops them after a backup.
+ *
+ * ## State and provenance
+ *
+ * The marker row `twin_collapse:<bare table>` in the twin's key/value table
+ * holds, per key (`schema_meta`) or per sticky id (`sticky_tags`), a sha256
+ * of the BARE value and of the TWIN value as of the last merge. A bare hash
+ * that moved means the older build wrote it; a twin hash that moved means this
+ * build wrote it.
+ *
+ * ## Rules
+ *
+ * The bare row is authoritative for every key. A monotonic counter FIELD is
+ * merged on its own: the result is the bare value with that field set to
+ * max(bare, twin). A whole twin value never replaces a bare one
+ * ({@link mergeSchemaMetaValue}).
+ *
+ * | `schema_meta` key | Counter field |
+ * |---|---|
+ * | `task_id_sequence` (`{counter,lastId,checksum}` from `sequence/index.ts`, or a plain number) | `counter` (a plain number takes the max of the two numbers) |
+ * | `sqlite_snapshot_gate` (`{generation,prefixes}`, `snapshot-gate.ts`) | `generation` |
+ * | `file_meta` (`FileMeta`; `session/engine-ops.ts` bumps `generation` on session start/end/resume) | `generation` |
+ * | `backfill:*` (the two `t877` guard keys) | not carried: the lineage re-inserts them into every bare table and nothing reads them |
+ * | `twin_collapse*` | not carried: collapse state and failures live only in the twin |
+ * | every other key (`schemaVersion`, `version`, `focus_state`, `focus_state:<session>`, `project_meta`, `project`, `parallel_state`, `activeSession`, `reconcile.<task>.release`, any unknown key) | none: the bare value |
+ *
+ * - **Initial collapse**, when the bare table holds at least one carried key:
+ *   every bare key is carried under the rules, and a key only the twin holds
+ *   is a frozen copy and is DROPPED (listed in the receipt, marker and log).
+ *   The exceptions are `task_id_sequence` and `sqlite_snapshot_gate` (a
+ *   counter must not move back) and the collapse keys. `file_meta` is dropped
+ *   only when the bare table lacks it; when both have it the field rule
+ *   applies. When the bare table holds no carried key (a fresh lineage, e.g.
+ *   after exodus landed legacy rows in the twin), nothing is dropped.
+ * - **Incremental re-merge**, per bare key whose hash changed since the last
+ *   merge:
+ *   - twin unchanged since the last merge → the bare value (counter fields
+ *     maxed);
+ *   - twin ALSO changed (this build wrote it too) → CONFLICT: the TWIN value
+ *     wins, recorded in the marker and reported by `cleo doctor` as a warning.
+ *     A counter key is never a conflict: the twin value is kept with its
+ *     counter field maxed against the bare one.
+ *   A key the older build deleted is deleted from the twin unless the twin
+ *   changed it too (conflict) or it is `task_id_sequence` /
+ *   `sqlite_snapshot_gate`. A key the bare table never changed is never
+ *   touched.
+ *
+ * `sticky_tags` → `brain_sticky_tags` (rows are `(sticky_id, tag)`, both the
+ * key, so a collision is two identical rows and one is kept):
+ *
+ * - Initial collapse, when the bare table has rows: the twin is made equal to
+ *   the bare set; frozen twin-only rows are dropped and listed. A bare tag whose
+ *   note no longer exists is not carried (`skipped`; it stays in the bare table).
+ * - Incremental re-merge, per sticky id whose bare tag-set hash changed: if
+ *   the twin's set for that id is unchanged since the last merge, the twin
+ *   set is made equal to the bare set (additions AND removals propagate);
+ *   if the twin set changed too, the twin wins and a conflict is recorded.
+ * - After every merge the junction is recomputed from each note's
+ *   `tags_json` ({@link syncStickyJunction}). Both builds write `tags_json`
+ *   (on the shared `brain_sticky_notes` row) together with their own junction,
+ *   so it is the one value both see; the junction is derived from it, and the
+ *   tag filter can never disagree with the displayed tags. A conflict is still
+ *   recorded for `cleo doctor`.
+ *
+ * ## Contract
+ *
+ * 1. **Snapshot first** (initial collapse only, only when the merge changes
+ *    the twin). The free space is checked before the `VACUUM INTO`. One
+ *    snapshot covers every pair collapsing in this open, registered as a
+ *    `migration` backup (`cleo backup list`, rotation).
+ * 2. **Atomic.** Each pair merges, verifies and writes its marker in one
+ *    `BEGIN IMMEDIATE` transaction, the marker re-read under the write lock.
+ *    Any failure rolls back: both tables byte-identical.
+ * 3. **Verified.** Every row the merge decided is re-read and compared before
+ *    the marker is written; a mismatch rolls back.
+ * 4. **The bare table is never written**, with one sanctioned exception:
+ *    {@link mirrorCounterToBare}. When this build writes a counter key
+ *    (`task_id_sequence`, `sqlite_snapshot_gate`, `file_meta`) it raises the
+ *    bare row's counter field to the twin's in the same transaction, and the
+ *    allocation floors on the bare counter ({@link bareCounterOf}). The
+ *    2026.9.20 build still allocates ids from the bare counter; without the
+ *    mirror, an id one build reserved but had not stored yet could be issued
+ *    by the other.
+ * 5. **Never locked out** (inside a domain bind, `onFailure: 'degrade'`). A
+ *    failed or impossible collapse does not fail the bind. The failure is
+ *    recorded (`twin_collapse_failed:<table>`), and the connection is put in a
+ *    READ-ONLY-FOR-USERS mode: the merged view the collapse would have
+ *    committed is computed in memory and served from TEMP shadow tables of the
+ *    same names (unqualified SQL resolves `temp` before `main`), so reads see
+ *    bare-authoritative data while `main` stays untouched. Mutating operations
+ *    are refused with `E_TWIN_COLLAPSE_FAILED` by the dispatch write guard
+ *    ({@link twinCollapseFailureOf}), and the shadows carry TEMP triggers that
+ *    abort any write into them ({@link SHADOW_WRITE_REFUSED}), so a direct SDK
+ *    write fails instead of being lost. `cleo doctor twin-collapse --retry`
+ *    runs the collapse with `onFailure: 'throw'`; on success the shadows and
+ *    their triggers are dropped.
+ *
+ * ## Gate 28
+ *
+ * This module is a sanctioned writer (`scripts/lint-no-raw-table-writes.mjs`
+ * SANCTIONED): it runs on the chokepoint handle inside the domain bind, and the
+ * accessors import the modules that call it.
+ *
+ * @module
+ * @task T12535
+ */
+import { createHash } from 'node:crypto';
+import { ExitCode } from '@cleocode/contracts';
+import { CleoError } from '../errors.js';
+import { getLogger } from '../logger.js';
+import { planMigrationSnapshot, writeMigrationSnapshot } from './pre-repair-snapshot.js';
+const log = getLogger('twin-collapse');
+/** Prefix of the marker key holding a pair's collapse state. */
+export const TWIN_COLLAPSE_MARKER_PREFIX = 'twin_collapse:';
+/** Prefix of the key recording a pair's last failed collapse. */
+export const TWIN_COLLAPSE_FAILURE_PREFIX = 'twin_collapse_failed:';
+/**
+ * The snapshot gate's key (`SNAPSHOT_GATE_META_KEY` in `snapshot-gate.ts`),
+ * spelled out: `snapshot-gate.ts` imports this module to mirror its counter,
+ * so importing it back would be a cycle.
+ */
+const SNAPSHOT_GATE_KEY = 'sqlite_snapshot_gate';
+/** `schema_meta` keys whose named field is a monotonic counter (merged as max). */
+export const SCHEMA_META_COUNTER_FIELDS = {
+    task_id_sequence: 'counter',
+    [SNAPSHOT_GATE_KEY]: 'generation',
+    file_meta: 'generation',
+};
+/** `schema_meta` keys never deleted from the twin (a counter never moves back). */
+export const SCHEMA_META_NEVER_DELETED = new Set([
+    'task_id_sequence',
+    SNAPSHOT_GATE_KEY,
+]);
+/** Most conflicts kept in the marker (the doctor warning lists them). */
+const MAX_CONFLICTS = 50;
+const sha = (value) => createHash('sha256').update(value).digest('hex');
+/** Hash of an absent value (distinct from any real value's hash). */
+const ABSENT = sha('\u0000absent');
+function hasMainTable(db, name) {
+    return (db.prepare("SELECT 1 FROM main.sqlite_master WHERE type = 'table' AND name = ?").get(name) !==
+        undefined);
+}
+function readKv(db, kvTable, key) {
+    return db.prepare(`SELECT value FROM main.${kvTable} WHERE key = ?`).get(key)?.value;
+}
+function writeKv(db, kvTable, key, value) {
+    db.prepare(`INSERT INTO main.${kvTable} (key, value) VALUES (?, ?) ` +
+        'ON CONFLICT(key) DO UPDATE SET value = excluded.value').run(key, value);
+}
+function sameHashes(a, b) {
+    const ka = Object.keys(a);
+    return ka.length === Object.keys(b).length && ka.every((k) => a[k] === b[k]);
+}
+/** Parse a marker row; a pre-release marker (no hashes) yields `hashes: null`. */
+function readState(db, pair) {
+    const raw = readKv(db, pair.kvTable, `${TWIN_COLLAPSE_MARKER_PREFIX}${pair.table}`);
+    if (raw === undefined)
+        return undefined;
+    let parsed = {};
+    try {
+        parsed = JSON.parse(raw);
+    }
+    catch {
+        parsed = {};
+    }
+    const str = (v) => (typeof v === 'string' ? v : '');
+    return {
+        version: 3,
+        task: 'T12535',
+        collapsedAt: str(parsed.collapsedAt),
+        lastMergedAt: str(parsed.lastMergedAt),
+        snapshot: typeof parsed.snapshot === 'string' ? parsed.snapshot : null,
+        hashes: parsed.version === 3 && parsed.hashes ? parsed.hashes : null,
+        dropped: Array.isArray(parsed.dropped) ? parsed.dropped : [],
+        conflicts: Array.isArray(parsed.conflicts) ? parsed.conflicts : [],
+        conflictsAt: typeof parsed.conflictsAt === 'string' ? parsed.conflictsAt : null,
+    };
+}
+/** Whether the bare side moved since the stored hashes (or no usable state exists). */
+function bareChanged(db, pair, state) {
+    return !state?.hashes || !sameHashes(state.hashes.bare, pair.bareHashes(db));
+}
+/** Hashes of the last merge; a pre-release marker treats the current twin as unchanged. */
+function lastHashes(db, pair, state) {
+    return state.hashes ?? { bare: {}, twin: pair.twinHashes(db) };
+}
+// ── schema_meta ──────────────────────────────────────────────────────────────
+function parseJson(value) {
+    try {
+        return JSON.parse(value);
+    }
+    catch {
+        return undefined;
+    }
+}
+/** A counter field's value, whether the whole value is a number or an object holding it. */
+function counterOf(value, field) {
+    const parsed = parseJson(value);
+    if (typeof parsed === 'number' && Number.isFinite(parsed))
+        return parsed;
+    if (parsed === null || typeof parsed !== 'object')
+        return undefined;
+    const n = parsed[field];
+    return typeof n === 'number' && Number.isFinite(n) ? n : undefined;
+}
+/**
+ * `base` with its counter field raised to the other value's counter when that
+ * is larger. `base` is returned verbatim otherwise, and when it has no
+ * counter to raise (not a number, not an object holding the field).
+ */
+function withMaxCounter(base, other, field) {
+    if (other === undefined)
+        return base;
+    const b = counterOf(base, field);
+    const o = counterOf(other, field);
+    if (b === undefined || o === undefined || o <= b)
+        return base;
+    const parsed = parseJson(base);
+    if (typeof parsed === 'number')
+        return String(o);
+    return JSON.stringify({ ...parsed, [field]: o });
+}
+/** Whether a `schema_meta` key is carried at all. */
+function carried(key) {
+    return !key.startsWith('backfill:') && !key.startsWith('twin_collapse');
+}
+/**
+ * The value a `schema_meta` key gets when the bare side is authoritative (the
+ * rule table is in the module comment).
+ *
+ * @param key - The key.
+ * @param bare - Its value in the bare table.
+ * @param twin - Its value in the twin, or `undefined` when absent.
+ * @returns The value to keep, or `null` when the key is not carried.
+ * @task T12535
+ */
+export function mergeSchemaMetaValue(key, bare, twin) {
+    if (!carried(key))
+        return null;
+    const field = SCHEMA_META_COUNTER_FIELDS[key];
+    return field === undefined ? bare : withMaxCounter(bare, twin, field);
+}
+/**
+ * Raise the BARE `schema_meta` row of a counter key to the twin's counter,
+ * right after this build wrote the twin row, on the same connection (so inside
+ * the writer's transaction).
+ *
+ * The one sanctioned write to a bare table (T12535): the 2026.9.20 build still
+ * allocates task ids from the bare `task_id_sequence` and reads the bare
+ * snapshot-gate and `file_meta` generations. Without this, an id this build
+ * reserved but has not stored yet (the counter is advanced before the task
+ * row is inserted) could be issued again by the older build. Only the counter
+ * field moves (the bare row keeps its other fields, see
+ * {@link mergeSchemaMetaValue}); a missing bare row gets the twin value. A
+ * non-counter key, or a store without the bare table, is left alone. The
+ * collapse treats the resulting bare change like any other: a counter key is
+ * never a conflict.
+ *
+ * @param db - The project `cleo.db` connection that wrote the twin row.
+ * @param key - The `schema_meta` key just written.
+ * @task T12535
+ */
+export function mirrorCounterToBare(db, key) {
+    if (SCHEMA_META_COUNTER_FIELDS[key] === undefined)
+        return;
+    if (!hasMainTable(db, 'schema_meta'))
+        return;
+    const twin = readKv(db, 'tasks_schema_meta', key);
+    if (twin === undefined)
+        return;
+    const bare = readKv(db, 'schema_meta', key);
+    const next = bare === undefined ? twin : mergeSchemaMetaValue(key, bare, twin);
+    if (next === null || next === bare)
+        return;
+    writeKv(db, 'schema_meta', key, next);
+}
+/**
+ * The counter of a key's BARE `schema_meta` row (the older build's), or 0.
+ * This build's allocation floors on it, so an id the older build reserved but
+ * has not stored yet is never issued here either.
+ *
+ * @param db - The project `cleo.db` connection.
+ * @param key - A counter key.
+ * @returns The bare counter field, or 0 when absent.
+ * @task T12535
+ */
+export function bareCounterOf(db, key) {
+    const name = SCHEMA_META_COUNTER_FIELDS[key];
+    if (name === undefined || !hasMainTable(db, 'schema_meta'))
+        return 0;
+    const bare = readKv(db, 'schema_meta', key);
+    return (bare === undefined ? undefined : counterOf(bare, name)) ?? 0;
+}
+function kvRows(db, table, schema = 'main') {
+    const rows = db.prepare(`SELECT key, value FROM ${schema}.${table}`).all();
+    return new Map(rows.map((r) => [r.key, r.value]));
+}
+/** Carried rows of a `schema_meta` table. */
+function carriedRows(db, table) {
+    const all = kvRows(db, table);
+    for (const key of [...all.keys()])
+        if (!carried(key))
+            all.delete(key);
+    return all;
+}
+const hashMap = (m) => Object.fromEntries([...m].map(([k, v]) => [k, sha(v)]));
+const emptyPlan = () => ({ set: new Map(), del: [], dropped: [], conflicts: [], skipped: 0 });
+function planSchemaMeta(db, state) {
+    const plan = emptyPlan();
+    const bare = carriedRows(db, 'schema_meta');
+    const twin = carriedRows(db, 'tasks_schema_meta');
+    const want = (key, value) => {
+        if (value !== null && value !== twin.get(key))
+            plan.set.set(key, value);
+    };
+    if (state === undefined) {
+        if (bare.size === 0)
+            return plan;
+        for (const [key, value] of bare)
+            want(key, mergeSchemaMetaValue(key, value, twin.get(key)));
+        for (const key of [...twin.keys()].sort()) {
+            if (!bare.has(key) && !SCHEMA_META_NEVER_DELETED.has(key)) {
+                plan.del.push(key);
+                plan.dropped.push(key);
+            }
+        }
+        return plan;
+    }
+    const last = lastHashes(db, SCHEMA_META, state);
+    const twinMoved = (key) => {
+        const now = twin.has(key) ? sha(twin.get(key)) : ABSENT;
+        return now !== (last.twin[key] ?? ABSENT);
+    };
+    for (const key of [...bare.keys()].sort()) {
+        const value = bare.get(key);
+        if (last.bare[key] === sha(value))
+            continue;
+        const field = SCHEMA_META_COUNTER_FIELDS[key];
+        const current = twin.get(key);
+        if (field !== undefined) {
+            // A counter is never a conflict: keep whichever side is authoritative
+            // for the rest of the value, and max the counter field against the other.
+            const base = twinMoved(key) && current !== undefined ? current : value;
+            want(key, withMaxCounter(base, base === value ? current : value, field));
+        }
+        else if (twinMoved(key)) {
+            plan.conflicts.push(key);
+        }
+        else {
+            want(key, value);
+        }
+    }
+    for (const key of Object.keys(last.bare).sort()) {
+        if (bare.has(key) || SCHEMA_META_NEVER_DELETED.has(key) || !twin.has(key))
+            continue;
+        if (twinMoved(key))
+            plan.conflicts.push(key);
+        else
+            plan.del.push(key);
+    }
+    return plan;
+}
+function applyKv(db, schema, table, plan) {
+    const before = kvRows(db, table, schema);
+    const upsert = db.prepare(`INSERT INTO ${schema}.${table} (key, value) VALUES (?, ?) ` +
+        'ON CONFLICT(key) DO UPDATE SET value = excluded.value');
+    const del = db.prepare(`DELETE FROM ${schema}.${table} WHERE key = ?`);
+    let inserted = 0;
+    let replaced = 0;
+    let deleted = 0;
+    for (const [key, value] of plan.set) {
+        upsert.run(key, value);
+        if (before.has(key))
+            replaced++;
+        else
+            inserted++;
+    }
+    for (const key of plan.del)
+        deleted += Number(del.run(key).changes);
+    return { inserted, replaced, deleted };
+}
+const SCHEMA_META = {
+    table: 'schema_meta',
+    twin: 'tasks_schema_meta',
+    kvTable: 'tasks_schema_meta',
+    tables: ['schema_meta', 'tasks_schema_meta'],
+    bareHashes: (db) => hashMap(carriedRows(db, 'schema_meta')),
+    twinHashes: (db) => hashMap(carriedRows(db, 'tasks_schema_meta')),
+    plan: planSchemaMeta,
+    apply(db, plan) {
+        const counts = applyKv(db, 'main', 'tasks_schema_meta', plan);
+        for (const [key, value] of plan.set) {
+            if (readKv(db, 'tasks_schema_meta', key) !== value)
+                throw new Error(`schema_meta collapse did not verify for key ${JSON.stringify(key)}`);
+        }
+        for (const key of plan.del) {
+            if (readKv(db, 'tasks_schema_meta', key) !== undefined)
+                throw new Error(`schema_meta collapse did not verify the removal of ${JSON.stringify(key)}`);
+        }
+        return counts;
+    },
+    shadow(db, plan) {
+        db.exec('CREATE TEMP TABLE IF NOT EXISTS tasks_schema_meta (key TEXT PRIMARY KEY, value TEXT NOT NULL)');
+        db.exec('DELETE FROM temp.tasks_schema_meta');
+        db.exec('INSERT INTO temp.tasks_schema_meta (key, value) SELECT key, value FROM main.tasks_schema_meta');
+        applyKv(db, 'temp', 'tasks_schema_meta', plan);
+    },
+};
+// ── sticky_tags ──────────────────────────────────────────────────────────────
+/** Tag sets by sticky id (tags sorted). */
+function stickySets(db, table) {
+    const rows = db
+        .prepare(`SELECT sticky_id, tag FROM main.${table} ORDER BY sticky_id, tag`)
+        .all();
+    const out = new Map();
+    for (const r of rows) {
+        const tags = out.get(r.sticky_id) ?? [];
+        tags.push(r.tag);
+        out.set(r.sticky_id, tags);
+    }
+    return out;
+}
+const hashSets = (m) => Object.fromEntries([...m].map(([k, tags]) => [k, sha(JSON.stringify(tags))]));
+const EMPTY_SET = sha(JSON.stringify([]));
+function planSticky(db, state) {
+    const plan = emptyPlan();
+    const bare = stickySets(db, 'sticky_tags');
+    const twin = stickySets(db, 'brain_sticky_tags');
+    const noteExists = db.prepare('SELECT 1 FROM main.brain_sticky_notes WHERE id = ?');
+    /** Make the twin's set for `id` equal the bare set. */
+    const mirror = (id, dropping) => {
+        const b = new Set(bare.get(id) ?? []);
+        const t = new Set(twin.get(id) ?? []);
+        const alive = b.size === 0 || noteExists.get(id) !== undefined;
+        for (const tag of b) {
+            if (!alive)
+                plan.skipped++;
+            else if (!t.has(tag))
+                plan.set.set(`${id}\t${tag}`, '');
+        }
+        for (const tag of t) {
+            if (b.has(tag))
+                continue;
+            plan.del.push(`${id}\t${tag}`);
+            if (dropping)
+                plan.dropped.push(`${id}\t${tag}`);
+        }
+    };
+    if (state === undefined) {
+        if (bare.size === 0)
+            return plan;
+        for (const id of [...new Set([...bare.keys(), ...twin.keys()])].sort())
+            mirror(id, true);
+        return plan;
+    }
+    const last = lastHashes(db, STICKY_TAGS, state);
+    const ids = new Set([...bare.keys(), ...Object.keys(last.bare)]);
+    for (const id of [...ids].sort()) {
+        const bareNow = bare.has(id) ? sha(JSON.stringify(bare.get(id))) : EMPTY_SET;
+        if (bareNow === (last.bare[id] ?? EMPTY_SET))
+            continue;
+        const twinNow = twin.has(id) ? sha(JSON.stringify(twin.get(id))) : EMPTY_SET;
+        if (twinNow !== (last.twin[id] ?? EMPTY_SET))
+            plan.conflicts.push(id);
+        else
+            mirror(id, false);
+    }
+    return plan;
+}
+function applyStickyRows(db, schema, plan) {
+    const insert = db.prepare(`INSERT OR IGNORE INTO ${schema}.brain_sticky_tags (sticky_id, tag) VALUES (?, ?)`);
+    const del = db.prepare(`DELETE FROM ${schema}.brain_sticky_tags WHERE sticky_id = ? AND tag = ?`);
+    let inserted = 0;
+    let deleted = 0;
+    for (const row of plan.set.keys()) {
+        const [id, tag] = row.split('\t');
+        inserted += Number(insert.run(id, tag).changes);
+    }
+    for (const row of plan.del) {
+        const [id, tag] = row.split('\t');
+        deleted += Number(del.run(id, tag).changes);
+    }
+    return { inserted, replaced: 0, deleted };
+}
+/** A note's tags as the accessor reads `tags_json` (strings, non-empty, deduplicated). */
+function tagsOf(tagsJson) {
+    if (!tagsJson)
+        return [];
+    try {
+        const parsed = JSON.parse(tagsJson);
+        if (!Array.isArray(parsed))
+            return [];
+        return [...new Set(parsed.filter((t) => typeof t === 'string' && t.length > 0))];
+    }
+    catch {
+        return [];
+    }
+}
+/**
+ * Make the junction in `schema` equal every note's `tags_json`.
+ *
+ * `tags_json` lives on the shared `brain_sticky_notes` row, which both builds
+ * write together with their own junction; the junction is derived from it.
+ * Recomputing it after every merge keeps the tag filter (the junction) and the
+ * displayed tags (`tags_json`) from disagreeing when both builds retag one
+ * note (the merge's twin-wins rule would otherwise keep a stale set).
+ *
+ * @returns Junction rows added and removed.
+ */
+function syncStickyJunction(db, schema) {
+    const notes = db.prepare('SELECT id, tags_json FROM main.brain_sticky_notes').all();
+    const have = new Map();
+    for (const r of db
+        .prepare(`SELECT sticky_id, tag FROM ${schema}.brain_sticky_tags`)
+        .all()) {
+        const set = have.get(r.sticky_id) ?? new Set();
+        set.add(r.tag);
+        have.set(r.sticky_id, set);
+    }
+    const insert = db.prepare(`INSERT OR IGNORE INTO ${schema}.brain_sticky_tags (sticky_id, tag) VALUES (?, ?)`);
+    const del = db.prepare(`DELETE FROM ${schema}.brain_sticky_tags WHERE sticky_id = ? AND tag = ?`);
+    let inserted = 0;
+    let deleted = 0;
+    for (const note of notes) {
+        const want = new Set(tagsOf(note.tags_json));
+        const current = have.get(note.id) ?? new Set();
+        for (const tag of want)
+            if (!current.has(tag))
+                inserted += Number(insert.run(note.id, tag).changes);
+        for (const tag of current)
+            if (!want.has(tag))
+                deleted += Number(del.run(note.id, tag).changes);
+    }
+    return { inserted, deleted };
+}
+const STICKY_TAGS = {
+    table: 'sticky_tags',
+    twin: 'brain_sticky_tags',
+    kvTable: 'brain_schema_meta',
+    tables: ['sticky_tags', 'brain_sticky_tags', 'brain_sticky_notes', 'brain_schema_meta'],
+    bareHashes: (db) => hashSets(stickySets(db, 'sticky_tags')),
+    twinHashes: (db) => hashSets(stickySets(db, 'brain_sticky_tags')),
+    plan: planSticky,
+    apply(db, plan) {
+        const counts = applyStickyRows(db, 'main', plan);
+        const present = db.prepare('SELECT 1 FROM main.brain_sticky_tags WHERE sticky_id = ? AND tag = ?');
+        const has = (row) => {
+            const [id, tag] = row.split('\t');
+            return present.get(id, tag) !== undefined;
+        };
+        const missing = [...plan.set.keys()].filter((r) => !has(r)).length;
+        const lingering = plan.del.filter(has).length;
+        if (missing > 0 || lingering > 0)
+            throw new Error(`sticky_tags collapse did not verify: ${missing} missing, ${lingering} not removed`);
+        const synced = syncStickyJunction(db, 'main');
+        return {
+            inserted: counts.inserted + synced.inserted,
+            replaced: 0,
+            deleted: counts.deleted + synced.deleted,
+        };
+    },
+    shadow(db, plan) {
+        db.exec('CREATE TEMP TABLE IF NOT EXISTS brain_sticky_tags (sticky_id TEXT NOT NULL, tag TEXT NOT NULL, PRIMARY KEY (sticky_id, tag))');
+        db.exec('DELETE FROM temp.brain_sticky_tags');
+        db.exec('INSERT INTO temp.brain_sticky_tags (sticky_id, tag) SELECT sticky_id, tag FROM main.brain_sticky_tags');
+        applyStickyRows(db, 'temp', plan);
+        syncStickyJunction(db, 'temp');
+    },
+};
+/** The pairs this build collapses, in order. */
+const PAIRS = [SCHEMA_META, STICKY_TAGS];
+/** Connections serving TEMP shadows because their collapse failed. */
+const degraded = new WeakMap();
+/**
+ * The failure a connection is degraded by, if any: reads on it are served from
+ * TEMP shadows and writes must be refused (the dispatch write guard does).
+ *
+ * @param db - A project `cleo.db` connection.
+ * @returns The failure, or `undefined` when the store is fully collapsed.
+ * @task T12535
+ */
+export function twinCollapseFailureOf(db) {
+    return degraded.get(db);
+}
+/**
+ * Refuse a write before its first statement when the store is degraded by a
+ * failed twin collapse. The write accessors of the collapsed tables
+ * (`tasks_schema_meta`, and sticky notes and tags) call it at entry, so a
+ * degraded store never takes a partial write (e.g. a sticky's `tags_json`
+ * without its tag rows). The shadows' TEMP triggers stay as the backstop.
+ *
+ * @param handle - The connection the write would use (a `DatabaseSync`, or a
+ *   drizzle instance's `$client`); anything else is not checked.
+ * @throws {CleoError} `E_TWIN_COLLAPSE_FAILED` when the connection is degraded.
+ * @task T12535
+ */
+export function assertTwinCollapseWritable(handle) {
+    if (typeof handle !== 'object' || handle === null)
+        return;
+    const failure = degraded.get(handle);
+    if (failure)
+        throw twinCollapseError(failure);
+}
+/**
+ * Build the `E_TWIN_COLLAPSE_FAILED` error for a failure.
+ *
+ * @param failure - The failure.
+ * @param cause - The underlying error, when there is one.
+ * @returns The error.
+ * @task T12535
+ */
+export function twinCollapseError(failure, cause) {
+    const where = !failure.snapshotPath
+        ? ''
+        : failure.snapshotWritten
+            ? ` Snapshot: ${failure.snapshotPath}.`
+            : ` Snapshot would be written to ${failure.snapshotPath}.`;
+    const space = failure.requiredBytes > 0
+        ? ` The snapshot needs ${failure.requiredBytes} bytes free` +
+            (failure.availableBytes === null ? '.' : ` (${failure.availableBytes} free).`)
+        : '';
+    return new CleoError(ExitCode.TWIN_COLLAPSE_FAILED, `Twin collapse of ${failure.tables.join(', ')} failed (both tables unchanged; reads still ` +
+        `work, writes are refused): ${failure.cause}.${where}${space} Run 'cleo doctor twin-collapse'.`, {
+        fix: "Run 'cleo doctor twin-collapse' for details, clear the cause (free the space, make " +
+            ".cleo/backups/sqlite writable), then 'cleo doctor twin-collapse --retry'.",
+        details: { field: 'twinCollapse', ...failure },
+        cause,
+    });
+}
+/** Record a failure outside the rolled-back transaction; best effort. */
+function recordFailure(db, pairs, failure) {
+    for (const pair of pairs) {
+        try {
+            writeKv(db, pair.kvTable, `${TWIN_COLLAPSE_FAILURE_PREFIX}${pair.table}`, JSON.stringify(failure));
+        }
+        catch {
+            // The store itself may be the problem (disk full, read-only); the
+            // failure is still carried by the degraded state and the error.
+        }
+    }
+}
+/** Message a write into a sealed shadow aborts with. */
+export const SHADOW_WRITE_REFUSED = 'E_TWIN_COLLAPSE_FAILED: store is read-only until the twin collapse succeeds; run cleo doctor twin-collapse';
+const SHADOW_WRITE_OPS = ['INSERT', 'UPDATE', 'DELETE'];
+/** Remove a shadow's write-refusing triggers (before it is rebuilt). */
+function unsealShadow(db, twin) {
+    for (const op of SHADOW_WRITE_OPS) {
+        db.exec(`DROP TRIGGER IF EXISTS temp.${twin}_refuse_${op.toLowerCase()}`);
+    }
+}
+/**
+ * Make a shadow refuse writes. A write through the store accessors that
+ * bypasses the dispatch write guard (a direct SDK call) then fails with
+ * {@link SHADOW_WRITE_REFUSED} instead of landing in the shadow and being lost
+ * when the connection closes. The collapse itself writes `main.<twin>`
+ * (schema-qualified), which these TEMP triggers do not cover.
+ */
+function sealShadow(db, twin) {
+    for (const op of SHADOW_WRITE_OPS) {
+        db.exec(`CREATE TEMP TRIGGER IF NOT EXISTS ${twin}_refuse_${op.toLowerCase()} BEFORE ${op} ON temp.${twin} ` +
+            `BEGIN SELECT RAISE(ABORT, '${SHADOW_WRITE_REFUSED}'); END`);
+    }
+}
+/** Drop the TEMP shadows (and their triggers) of a connection that is no longer degraded. */
+function clearShadows(db) {
+    for (const pair of PAIRS)
+        db.exec(`DROP TABLE IF EXISTS temp.${pair.twin}`);
+    degraded.delete(db);
+}
+function receipt(pair, status, snapshotPath) {
+    return {
+        table: pair.table,
+        twin: pair.twin,
+        status,
+        snapshotPath,
+        inserted: 0,
+        replaced: 0,
+        deleted: 0,
+        skipped: 0,
+        dropped: [],
+        conflicts: [],
+    };
+}
+/** Run one pair's merge in its own transaction. */
+function collapsePair(db, pair, snapshotPath) {
+    db.exec('BEGIN IMMEDIATE');
+    try {
+        // Re-read under the write lock: another process may have merged since.
+        const state = readState(db, pair);
+        if (!bareChanged(db, pair, state)) {
+            db.exec('ROLLBACK');
+            return receipt(pair, 'unchanged', state?.snapshot ?? null);
+        }
+        const plan = pair.plan(db, state);
+        if (state === undefined && snapshotPath === null && (plan.set.size > 0 || plan.del.length > 0))
+            throw new Error(`bare ${pair.table} changed after the snapshot decision; retry the open`);
+        const counts = pair.apply(db, plan);
+        const now = new Date().toISOString();
+        const next = {
+            version: 3,
+            task: 'T12535',
+            collapsedAt: state?.collapsedAt || now,
+            lastMergedAt: now,
+            snapshot: state?.snapshot ?? snapshotPath,
+            hashes: { bare: pair.bareHashes(db), twin: pair.twinHashes(db) },
+            dropped: state === undefined ? plan.dropped : state.dropped,
+            conflicts: plan.conflicts.slice(0, MAX_CONFLICTS),
+            conflictsAt: plan.conflicts.length > 0 ? now : null,
+        };
+        writeKv(db, pair.kvTable, `${TWIN_COLLAPSE_MARKER_PREFIX}${pair.table}`, JSON.stringify(next));
+        db.prepare(`DELETE FROM main.${pair.kvTable} WHERE key = ?`).run(`${TWIN_COLLAPSE_FAILURE_PREFIX}${pair.table}`);
+        db.exec('COMMIT');
+        const done = {
+            ...receipt(pair, state === undefined ? 'initial' : 'incremental', next.snapshot),
+            ...counts,
+            skipped: plan.skipped,
+            dropped: plan.dropped,
+            conflicts: plan.conflicts,
+        };
+        if (counts.inserted + counts.replaced + counts.deleted > 0 || plan.conflicts.length > 0)
+            log.warn(done, `carried bare ${pair.table} into ${pair.twin} (${done.status}, T12535)`);
+        return done;
+    }
+    catch (error) {
+        if (db.isTransaction)
+            db.exec('ROLLBACK');
+        throw error;
+    }
+}
+/**
+ * Collapse every bare/twin pair present in the store: the initial collapse
+ * (snapshot first) where no marker exists yet, an incremental re-merge of
+ * changed bare keys where one does.
+ *
+ * @param nativeDb - The project `cleo.db` connection, outside any transaction.
+ * @param dbPath - Its file path (locates `.cleo/backups/sqlite/`).
+ * @param options - Failure handling.
+ * @returns One receipt per pair.
+ * @throws {CleoError} `E_TWIN_COLLAPSE_FAILED` with `onFailure: 'throw'`.
+ * @task T12535
+ */
+export function collapseTwinTables(nativeDb, dbPath, options = {}) {
+    const onFailure = options.onFailure ?? 'throw';
+    // A bind on a connection already degraded in this process does not retry:
+    // the cause is outside CLEO; `cleo doctor twin-collapse --retry` retries.
+    if (onFailure === 'degrade' && degraded.has(nativeDb))
+        return PAIRS.map((p) => receipt(p, 'degraded', null));
+    const byTable = new Map();
+    const inOrder = () => PAIRS.map((p) => byTable.get(p.table)).filter((r) => r !== undefined);
+    const pending = [];
+    for (const pair of PAIRS) {
+        if (!pair.tables.every((t) => hasMainTable(nativeDb, t))) {
+            byTable.set(pair.table, receipt(pair, 'no-bare-table', null));
+            continue;
+        }
+        const state = readState(nativeDb, pair);
+        if (bareChanged(nativeDb, pair, state))
+            pending.push(pair);
+        else
+            byTable.set(pair.table, receipt(pair, 'unchanged', state?.snapshot ?? null));
+    }
+    if (pending.length === 0) {
+        if (degraded.has(nativeDb))
+            clearShadows(nativeDb);
+        return inOrder();
+    }
+    if (nativeDb.isTransaction)
+        throw new Error('twin collapse needs a connection outside a transaction');
+    const fail = (pairs, failure, cause) => {
+        recordFailure(nativeDb, pairs, failure);
+        log.error(failure, `twin collapse of ${failure.tables.join(', ')} failed (T12535)`);
+        if (onFailure === 'throw')
+            throw twinCollapseError(failure, cause);
+        // Read-only-for-users mode: serve the merged view from TEMP shadows. If
+        // even that cannot be built, reads cannot be served correctly either.
+        try {
+            for (const pair of pairs) {
+                unsealShadow(nativeDb, pair.twin);
+                pair.shadow(nativeDb, pair.plan(nativeDb, readState(nativeDb, pair)));
+                sealShadow(nativeDb, pair.twin);
+            }
+        }
+        catch (shadowError) {
+            throw twinCollapseError(failure, shadowError);
+        }
+        degraded.set(nativeDb, failure);
+        for (const pair of pairs)
+            byTable.set(pair.table, receipt(pair, 'degraded', failure.snapshotPath));
+    };
+    // One snapshot covers every pair whose INITIAL collapse changes its twin.
+    const needSnapshot = pending.filter((p) => {
+        if (readState(nativeDb, p) !== undefined)
+            return false;
+        const plan = p.plan(nativeDb, undefined);
+        return plan.set.size > 0 || plan.del.length > 0;
+    });
+    let snapshotPath = null;
+    if (needSnapshot.length > 0) {
+        const plan = planMigrationSnapshot(nativeDb, dbPath);
+        try {
+            snapshotPath = writeMigrationSnapshot(nativeDb, plan, `T12535 twin collapse of ${needSnapshot.map((p) => p.table).join(', ')} (store before the merge)`);
+        }
+        catch (error) {
+            fail(needSnapshot, {
+                tables: needSnapshot.map((p) => p.table),
+                cause: `snapshot not written: ${error instanceof Error ? error.message : String(error)}`,
+                snapshotPath: plan.snapshotPath,
+                snapshotWritten: false,
+                requiredBytes: plan.requiredBytes,
+                availableBytes: plan.availableBytes,
+                failedAt: new Date().toISOString(),
+            }, error);
+            // Pairs that need no snapshot (incremental, or an initial no-op) still run.
+            const rest = pending.filter((p) => !needSnapshot.includes(p));
+            pending.length = 0;
+            pending.push(...rest);
+        }
+    }
+    const failed = [];
+    let firstError;
+    for (const pair of pending) {
+        try {
+            byTable.set(pair.table, collapsePair(nativeDb, pair, snapshotPath));
+        }
+        catch (error) {
+            failed.push(pair);
+            firstError ??= error;
+        }
+    }
+    if (failed.length > 0) {
+        fail(failed, {
+            tables: failed.map((p) => p.table),
+            cause: firstError instanceof Error ? firstError.message : String(firstError),
+            snapshotPath,
+            snapshotWritten: snapshotPath !== null,
+            requiredBytes: 0,
+            availableBytes: null,
+            failedAt: new Date().toISOString(),
+        }, firstError);
+    }
+    else if (degraded.has(nativeDb) && byTable.size === PAIRS.length) {
+        if ([...byTable.values()].every((r) => r.status !== 'degraded'))
+            clearShadows(nativeDb);
+    }
+    return inOrder();
+}
+/**
+ * Report every pair's collapse state without writing (for `cleo doctor`).
+ *
+ * @param db - A connection on the project `cleo.db`; read-only is enough.
+ * @returns One status per pair.
+ * @task T12535
+ */
+export function inspectTwinCollapse(db) {
+    return PAIRS.map((pair) => {
+        const empty = {
+            table: pair.table,
+            twin: pair.twin,
+            wouldChangeTwin: false,
+            snapshotPath: null,
+            collapsedAt: null,
+            lastMergedAt: null,
+            changedSinceMerge: 0,
+            conflicts: [],
+            conflictsAt: null,
+            failure: null,
+        };
+        if (!pair.tables.every((t) => hasMainTable(db, t)))
+            return { ...empty, state: 'no-bare-table' };
+        let failure = null;
+        try {
+            const raw = readKv(db, pair.kvTable, `${TWIN_COLLAPSE_FAILURE_PREFIX}${pair.table}`);
+            failure = raw === undefined ? null : JSON.parse(raw);
+        }
+        catch {
+            failure = null;
+        }
+        const state = readState(db, pair);
+        const now = pair.bareHashes(db);
+        const prev = state?.hashes?.bare ?? {};
+        let changed = 0;
+        for (const k of new Set([...Object.keys(now), ...Object.keys(prev)])) {
+            if (now[k] !== prev[k])
+                changed++;
+        }
+        let wouldChangeTwin = false;
+        if (state === undefined) {
+            const plan = pair.plan(db, undefined);
+            wouldChangeTwin = plan.set.size > 0 || plan.del.length > 0;
+        }
+        return {
+            ...empty,
+            state: failure !== null
+                ? 'failed'
+                : state === undefined
+                    ? 'pending'
+                    : changed > 0 || state.hashes === null
+                        ? 'bare-changed'
+                        : 'collapsed',
+            wouldChangeTwin,
+            snapshotPath: state?.snapshot ?? failure?.snapshotPath ?? null,
+            collapsedAt: state?.collapsedAt || null,
+            lastMergedAt: state?.lastMergedAt || null,
+            changedSinceMerge: state === undefined ? 0 : changed,
+            conflicts: state?.conflicts ?? [],
+            conflictsAt: state?.conflictsAt ?? null,
+            failure,
+        };
+    });
+}
+//# sourceMappingURL=twin-collapse.js.map