akm-cli 0.9.16-alpha.1 → 0.9.16

package/dist/storage/repositories/files-repository.js

@@ -1,181 +0,0 @@
-// This Source Code Form is subject to the terms of the Mozilla Public
-// License, v. 2.0. If a copy of the MPL was not distributed with this
-// file, You can obtain one at https://mozilla.org/MPL/2.0/.
-import { SQLITE_CHUNK_SIZE } from "./index-sql.js";
-// ── Schema ──────────────────────────────────────────────────────────────────
-/**
- * Create `files`, `unit_texts` and `units_fts` if they do not already exist.
- * Idempotent; never drops or rewrites an existing row. Safe to call on every
- * schema ensure, same as `units-repository.ts`'s `ensureUnitTables`.
- */
-export function ensureFileAndUnitTextTables(db) {
-    db.exec(`
-    CREATE TABLE IF NOT EXISTS files (
-      path       TEXT PRIMARY KEY,
-      bundle_id  TEXT NOT NULL,
-      size       INTEGER NOT NULL,
-      mtime_ms   REAL NOT NULL,
-      ctime_ms   REAL NOT NULL DEFAULT -1,
-      blob_hash  TEXT NOT NULL,
-      adapter_id TEXT NOT NULL DEFAULT ''
-    );
-    CREATE INDEX IF NOT EXISTS files_bundle ON files(bundle_id);
-
-    CREATE TABLE IF NOT EXISTS unit_texts (
-      unit_hash TEXT PRIMARY KEY,
-      kind      TEXT NOT NULL CHECK (kind IN ('card','fragment')),
-      text      TEXT NOT NULL
-    );
-    -- index-redesign-contract.md B5f item 2: search runs the lexical query as
-    -- two kind-scoped lists (card, fragment) so a name/description match
-    -- ranks in its own small pool instead of competing with body text in one
-    -- BM25 pool. That kind filter runs on every search, so it needs an index.
-    CREATE INDEX IF NOT EXISTS unit_texts_kind ON unit_texts(kind);
-
-    CREATE VIRTUAL TABLE IF NOT EXISTS units_fts USING fts5(
-      unit_hash UNINDEXED, text, tokenize='porter unicode61'
-    );
-  `);
-    ensureFilesAdapterIdColumn(db);
-    ensureFilesCtimeColumn(db);
-}
-/**
- * `adapter_id` was added after `files`' first release, so a database created
- * before it needs an `ALTER TABLE` (`CREATE TABLE IF NOT EXISTS` only shapes
- * a fresh table). Idempotent. A pre-existing row's default `''` matches no
- * real adapter id, so the very next reconcile sees it as a mismatch and
- * re-parses that one file under its current adapter — a one-time,
- * self-healing cost, not a correctness gap.
- */
-function ensureFilesAdapterIdColumn(db) {
-    const columns = db.prepare("PRAGMA table_info(files)").all();
-    if (!columns.some((column) => column.name === "adapter_id")) {
-        db.exec("ALTER TABLE files ADD COLUMN adapter_id TEXT NOT NULL DEFAULT ''");
-    }
-}
-/**
- * `ctime_ms` was added after `files`' first release, the same additive,
- * self-healing way `adapter_id` was (see {@link ensureFilesAdapterIdColumn}):
- * a database created before it needs an `ALTER TABLE`. A pre-existing row's
- * default `-1` matches no real `ctimeMs` (always >= 0), so the very next
- * reconcile sees it as a mismatch and re-parses that one file — a one-time,
- * self-healing cost, not a correctness gap.
- */
-function ensureFilesCtimeColumn(db) {
-    const columns = db.prepare("PRAGMA table_info(files)").all();
-    if (!columns.some((column) => column.name === "ctime_ms")) {
-        db.exec("ALTER TABLE files ADD COLUMN ctime_ms REAL NOT NULL DEFAULT -1");
-    }
-}
-// ── files ───────────────────────────────────────────────────────────────────
-function rowToFileState(row) {
-    return {
-        path: row.path,
-        bundleId: row.bundle_id,
-        size: row.size,
-        mtimeMs: row.mtime_ms,
-        ctimeMs: row.ctime_ms,
-        blobHash: row.blob_hash,
-        adapterId: row.adapter_id,
-    };
-}
-/** The stored stat/hash row for one path, or `undefined` if it has never been reconciled. */
-export function getFileState(db, path) {
-    const row = db
-        .prepare("SELECT path, bundle_id, size, mtime_ms, ctime_ms, blob_hash, adapter_id FROM files WHERE path = ?")
-        .get(path);
-    return row ? rowToFileState(row) : undefined;
-}
-/** Every stored `files` row for one bundle — the stat cache `reconcileRoots` diffs one root's walk against. */
-export function getFileStatesByBundle(db, bundleId) {
-    const rows = db
-        .prepare("SELECT path, bundle_id, size, mtime_ms, ctime_ms, blob_hash, adapter_id FROM files WHERE bundle_id = ?")
-        .all(bundleId);
-    return rows.map(rowToFileState);
-}
-/** Insert or replace one file's stat/hash row. */
-export function upsertFileState(db, row) {
-    db.prepare(`INSERT INTO files (path, bundle_id, size, mtime_ms, ctime_ms, blob_hash, adapter_id) VALUES (?, ?, ?, ?, ?, ?, ?)
-       ON CONFLICT(path) DO UPDATE SET bundle_id = excluded.bundle_id, size = excluded.size,
-         mtime_ms = excluded.mtime_ms, ctime_ms = excluded.ctime_ms, blob_hash = excluded.blob_hash,
-         adapter_id = excluded.adapter_id`).run(row.path, row.bundleId, row.size, row.mtimeMs, row.ctimeMs, row.blobHash, row.adapterId);
-}
-/** Remove `files` rows for paths that no longer have an entry (gone, or the adapter no longer recognizes them). */
-export function deleteFileStates(db, paths) {
-    for (let offset = 0; offset < paths.length; offset += SQLITE_CHUNK_SIZE) {
-        const chunk = paths.slice(offset, offset + SQLITE_CHUNK_SIZE);
-        const placeholders = chunk.map(() => "?").join(",");
-        db.prepare(`DELETE FROM files WHERE path IN (${placeholders})`).run(...chunk);
-    }
-}
-// ── unit_texts / units_fts ───────────────────────────────────────────────────
-/**
- * Write the text for every hash in `units` reconcile has not stored before.
- * `unit_texts` dedupes itself (`INSERT OR IGNORE` on its `unit_hash` PRIMARY
- * KEY); a `units_fts` mirror row is written ONLY when the `unit_texts` insert
- * actually happened (FTS5 has no PRIMARY KEY to `OR IGNORE` against, so
- * inserting unconditionally would duplicate a hash already indexed).
- */
-export function insertNewUnitTexts(db, units) {
-    if (units.length === 0)
-        return { inserted: 0 };
-    const insertText = db.prepare("INSERT OR IGNORE INTO unit_texts (unit_hash, kind, text) VALUES (?, ?, ?)");
-    const insertFts = db.prepare("INSERT INTO units_fts (unit_hash, text) VALUES (?, ?)");
-    let inserted = 0;
-    for (const unit of units) {
-        const result = insertText.run(unit.hash, unit.kind, unit.text);
-        if (Number(result.changes) > 0) {
-            insertFts.run(unit.hash, unit.text);
-            inserted++;
-        }
-    }
-    return { inserted };
-}
-/**
- * Delete `unit_texts`/`units_fts` rows no `entry_units` row references any
- * more — a single `NOT EXISTS` sweep, run once at the end of `reconcileRoots`
- * (never `reconcilePaths`, whose small known-paths scope makes a full-table
- * sweep wasteful for the write-path caller it serves). Vectors
- * (`units`/`units_vec`) are never touched here — dropping the text does not
- * drop the embedding, so a hash that comes back later resumes serving search
- * without re-embedding.
- */
-export function pruneOrphanUnitTexts(db) {
-    const before = db.prepare("SELECT COUNT(*) AS n FROM unit_texts").get().n;
-    db.exec("DELETE FROM units_fts WHERE unit_hash IN (SELECT unit_hash FROM unit_texts WHERE NOT EXISTS " +
-        "(SELECT 1 FROM entry_units WHERE entry_units.unit_hash = unit_texts.unit_hash))");
-    db.exec("DELETE FROM unit_texts WHERE NOT EXISTS (SELECT 1 FROM entry_units WHERE entry_units.unit_hash = unit_texts.unit_hash)");
-    const after = db.prepare("SELECT COUNT(*) AS n FROM unit_texts").get().n;
-    return { removed: before - after };
-}
-/**
- * Delete `unit_texts`/`units_fts` rows for exactly the hashes in `hashes`
- * that no `entry_units` row references any more — the narrow counterpart to
- * {@link pruneOrphanUnitTexts}'s whole-table sweep, for a caller that already
- * knows precisely which hashes its own write just orphaned (`reconcile.ts`'s
- * `applyChange`, run after every write — including `reconcilePaths`, which
- * previously never pruned anything and so grew `unit_texts`/`units_fts`
- * without bound across repeated edits) rather than sweeping every row in the
- * table on every single write. Vectors (`units`/`units_vec`) are never
- * touched here either, for the same reason `pruneOrphanUnitTexts` leaves
- * them alone: dropping a unit's text must not drop its vector, so a hash
- * that returns resumes serving search without re-embedding.
- */
-export function pruneOrphanUnitTextsForHashes(db, hashes) {
-    const unique = [...new Set(hashes)];
-    if (unique.length === 0)
-        return { removed: 0 };
-    let removed = 0;
-    for (let offset = 0; offset < unique.length; offset += SQLITE_CHUNK_SIZE) {
-        const chunk = unique.slice(offset, offset + SQLITE_CHUNK_SIZE);
-        const placeholders = chunk.map(() => "?").join(",");
-        db.prepare(`DELETE FROM units_fts WHERE unit_hash IN (${placeholders}) AND NOT EXISTS ` +
-            `(SELECT 1 FROM entry_units WHERE entry_units.unit_hash = units_fts.unit_hash)`).run(...chunk);
-        const result = db
-            .prepare(`DELETE FROM unit_texts WHERE unit_hash IN (${placeholders}) AND NOT EXISTS ` +
-            `(SELECT 1 FROM entry_units WHERE entry_units.unit_hash = unit_texts.unit_hash)`)
-            .run(...chunk);
-        removed += Number(result.changes);
-    }
-    return { removed };
-}