@stonyx/orm 0.3.2-beta.3 → 0.3.2-beta.300

package/src/mysql/mysql-db.ts

@@ -21,6 +21,7 @@ interface PersistContext {
   record?: OrmRecord;
   recordId?: unknown;
   oldState?: Record<string, unknown>;
+  rawData?: Record<string, unknown>;
 }
 
 interface PersistResponse {
@@ -83,6 +84,15 @@ export default class MysqlDB {
   pool!: Pool | null;
   mysqlConfig!: MysqlConfig;
 
+  /**
+   * Promise-chain mutex for write serialization (#156).
+   * All persist() calls chain through this single queue so concurrent
+   * fire-and-forget writes never produce parallel InnoDB transactions
+   * on FK-linked rows (which cause deadlocks).
+   * Reads are NOT affected — only persist() serializes.
+   */
+  private _writeQueue: Promise<void> = Promise.resolve();
+
   constructor(deps: Partial<MysqlDBDeps> = {}) {
     if (MysqlDB.instance) return MysqlDB.instance;
     MysqlDB.instance = this;
@@ -117,7 +127,15 @@ export default class MysqlDB {
     if (pending.length > 0) {
       this.deps.log.db?.(`${pending.length} pending migration(s) found.`);
 
-      const shouldApply = await this.deps.confirm(`${pending.length} pending migration(s) found. Apply now?`);
+      let shouldApply: boolean;
+      if (this.mysqlConfig.autoMigrate === true) {
+        shouldApply = true;
+      } else if (this.mysqlConfig.autoMigrate === false) {
+        shouldApply = false;
+        this.deps.log.warn?.(`autoMigrate is false — skipping ${pending.length} pending migration(s).`);
+      } else {
+        shouldApply = await this.deps.confirm(`${pending.length} pending migration(s) found. Apply now?`);
+      }
 
       if (shouldApply) {
         for (const filename of pending) {
@@ -138,9 +156,17 @@ export default class MysqlDB {
       const modelCount = Object.keys(schemas).length;
 
       if (modelCount > 0) {
-        const shouldGenerate = await this.deps.confirm(
-          `No migrations found but ${modelCount} model(s) detected. Generate and apply initial migration?`
-        );
+        let shouldGenerate: boolean;
+        if (this.mysqlConfig.autoMigrate === true) {
+          shouldGenerate = true;
+        } else if (this.mysqlConfig.autoMigrate === false) {
+          shouldGenerate = false;
+          this.deps.log.warn?.(`autoMigrate is false — skipping initial migration generation for ${modelCount} model(s).`);
+        } else {
+          shouldGenerate = await this.deps.confirm(
+            `No migrations found but ${modelCount} model(s) detected. Generate and apply initial migration?`
+          );
+        }
 
         if (shouldGenerate) {
           const { generateMigration } = await import('./migration-generator.js');
@@ -397,14 +423,21 @@ export default class MysqlDB {
     const Orm = (await import('@stonyx/orm')).default;
     if ((Orm as unknown as { instance?: { isView?: (name: string) => boolean } }).instance?.isView?.(modelName)) return;
 
-    switch (operation) {
-      case 'create':
-        return this._persistCreate(modelName, context, response);
-      case 'update':
-        return this._persistUpdate(modelName, context, response);
-      case 'delete':
-        return this._persistDelete(modelName, context);
-    }
+    const work = async () => {
+      switch (operation) {
+        case 'create':
+          return this._persistCreate(modelName, context, response);
+        case 'update':
+          return this._persistUpdate(modelName, context, response);
+        case 'delete':
+          return this._persistDelete(modelName, context);
+      }
+    };
+
+    // Chain through the write queue — .then(work, work) ensures the queue
+    // advances even when a previous persist rejects (#156).
+    this._writeQueue = this._writeQueue.then(work, work);
+    return this._writeQueue;
   }
 
   private async _persistCreate(modelName: string, context: PersistContext, response: PersistResponse): Promise<void> {
@@ -420,8 +453,10 @@ export default class MysqlDB {
 
     const insertData = this._recordToRow(record, schema);
 
-    // For auto-increment models, remove the pending ID
-    const isPendingId = record.__data.__pendingSqlId;
+    // For auto-increment models, remove the pending ID.
+    // Check context.rawData (not record.__data) because __pendingSqlId is not a model
+    // attribute and gets lost during serialization.
+    const isPendingId = context.rawData?.__pendingSqlId === true;
 
     if (isPendingId) {
       delete insertData.id;

package/src/normalize-record-id.ts

@@ -0,0 +1,88 @@
+/*
+ * Copyright 2025 Stone Costa
+ *
+ * Licensed under the Apache License, Version 2.0 (the 'License');
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an 'AS IS' BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+/**
+ * The ONE place a URL id is turned into the value a record is resolved by —
+ * abofs/stonyx-orm#270.
+ *
+ * Before this existed this coercion was written out seven times: twice in
+ * README.md, once in docs/usage-patterns.md, and four times inside the
+ * framework. Seven copies, not seven identical copies — the three persistence
+ * ones (abofs/stonyx-orm#282) omit the `if (!id) return ''` guard below, which
+ * is the point: nothing held them together, so they had already drifted.
+ * The framework's copy — `getId()`, module-private in
+ * orm-request.ts, unreachable through the package `exports` map — was the one
+ * that decided which record a request addressed, and a consumer's `access()`
+ * predicate had no way to obtain it. So the framework resolved the record by
+ * one value and asked the consumer to authorize on a different one.
+ *
+ * Measured consequence at the time of filing: with the documented predicate
+ * applied verbatim to a numeric-id model, `GET /animals/007`, `/7.9`, `/7e0`,
+ * `/0x7`, `/%207`, `/%2B7` and `/7%0A` each served the protected record, and
+ * `DELETE /animals/007` destroyed it — because `parseInt` folds every one of
+ * those onto `7` while `'007' === '7'` is false.
+ *
+ * Two things follow from this being a single exported function, and both are
+ * the point:
+ *
+ *  1. A permissive change here is HARMLESS, because both sides move together.
+ *     Lowercasing string ids used to disclose and destroy `owner:angela` with
+ *     the whole suite green; with one implementation the predicate simply sees
+ *     the same lowercased value and still refuses.
+ *  2. A divergence — a second, private normaliser at the resolution site — is
+ *     what is now dangerous, and that is what the tests pin, by observing the
+ *     key the resolution path actually used.
+ *
+ * SEMANTICS ARE UNCHANGED FROM `getId()`, deliberately and byte-for-byte.
+ * Whether `7.9` / `0x7` / `%0A` *should* resolve to record 7 at all is a real
+ * question, and it is a behaviour change on the record-resolution path for
+ * every consumer rather than an authorization fix — it belongs in its own
+ * issue with its own compatibility argument (issue body scope item 4). #270
+ * preserves today's semantics exactly and makes both sides agree on them.
+ *
+ * Two details are load-bearing and are pinned by
+ * test/unit/normalize-record-id-test.ts:
+ *
+ *  - `parseInt` is called with NO RADIX. `parseInt('0x7')` is 7;
+ *    `parseInt('0x7', 10)` is 0. Adding the radix looks like a cleanup and
+ *    silently changes which record every hex-spelled URL addresses.
+ *  - The coercion applies only when the id LOOKS numeric, so a model with
+ *    string ids is passed through untouched, case included.
+ *
+ * Must stay synchronous: `auth()` is invoked without `await` by
+ * @stonyx/rest-server, so a promise here would be handed to `access()` as the
+ * record id.
+ *
+ * FALSY AND `NaN` RETURNS ARE LOAD-BEARING ELSEWHERE — abofs/stonyx-orm#287.
+ * `''` is returned for any falsy id because `store.get(key, undefined)` returns
+ * the whole model Map rather than a record (abofs/stonyx-orm#167). But `''` is
+ * not the only falsy return: `'0'`, `'00'`, `'-0'` and `'0x0'` all normalise to
+ * `0`, and `' '`, `'\t'`, `'\n'`, `'\u00a0'` all normalise to `NaN` (their
+ * `Number()` is `0`, so the `isNaN` guard does not fire and `parseInt` runs).
+ * `store.remove(key, id)` branches on truthiness, so those spellings reach a
+ * fall-through this function does not own. Tracked as #287; every row is pinned
+ * in test/unit/normalize-record-id-test.ts so a cleanup here cannot move the
+ * boundary #287 is measured against.
+ *
+ * @param id the raw, already-URL-decoded id text from `request.params.id`
+ * @returns the value the ORM resolves the record by
+ */
+export default function normalizeRecordId(id?: string | null): string | number {
+  if (!id) return '';
+  if (isNaN(id as unknown as number)) return id;
+
+  return parseInt(id);
+}

package/src/orm-request.ts

@@ -7,11 +7,38 @@ import type { HookContext } from './hooks.js';
 import config from 'stonyx/config';
 import type { OrmRecord } from './types/orm-types.js';
 import { isOrmRecord } from './utils.js';
+import normalizeRecordId from './normalize-record-id.js';
 
-interface OrmRequest$ extends Request {
+/**
+ * The request object a consumer's `access()` predicate receives.
+ *
+ * Exported because `recordId` is public API — README's `access()` samples
+ * destructure it — and a public runtime field with an unreachable type asks the
+ * consumer to re-declare something the framework already knows, which is
+ * abofs/stonyx-orm#270's own defect shape one layer over into the type surface.
+ * `HookContext` (src/hooks.ts) is this repo's precedent for exporting the
+ * interface a consumer is handed. Re-exported from the root barrel as
+ * `OrmAccessRequest` (src/index.ts) — NOT as `OrmRequest`, which is already the
+ * default-exported CLASS in this file and means something else. The
+ * `./orm-request` subpath is not in the `exports` map, so the barrel is the
+ * only reachable spelling.
+ */
+export interface OrmRequest$ extends Request {
   protocol?: string;
+  // Express sets this to the path the router was mounted at, e.g. '/api/animals'
+  // when orm.restServer.route is '/api'. Optional because non-Express callers
+  // (unit tests, programmatic handler invocation) do not supply it.
+  baseUrl?: string;
   method: string;
   params: { [key: string]: string };
+  // Attached by auth() before access() runs — abofs/stonyx-orm#270. This is the
+  // value the ORM resolves the record by; `params.id` remains the raw client
+  // text it was parsed from.
+  //
+  // `undefined` is part of the contract, not an absence: it is how a collection
+  // route is told from a record route, and both documented samples branch on
+  // it. Spelled out rather than left to `?:` for that reason.
+  recordId?: string | number | undefined;
   body?: { [key: string]: unknown };
   query?: { [key: string]: string };
   get(header: string): string;
@@ -77,19 +104,49 @@ function getModelRelationships(modelName: string): { [key: string]: Relationship
   return relationships;
 }
 
-// Helper to build base URL from request
-function getBaseUrl(request: OrmRequest$): string {
+/**
+ * Build the absolute base URL that every advertised link hangs off — origin
+ * plus the prefix the ORM's routes are actually mounted at.
+ *
+ * The prefix is derived from the *request*, not from
+ * `config.orm.restServer.route`. Express sets `request.baseUrl` to the real
+ * mountpath registered by `RestServer.mountRoute`, which for this module is
+ * always `<prefix>/<pluralizedModel>` (see setup-rest-server.ts). Stripping the
+ * trailing model segment therefore yields the prefix by construction, and the
+ * link builder cannot drift from the mount registrar the way a second,
+ * independent normalisation of `route` would.
+ *
+ * When `request.baseUrl` is absent or does not end in the model segment the
+ * prefix is empty, which reproduces the previous origin-only behaviour.
+ */
+function getBaseUrl(request: OrmRequest$, pluralizedModel: string): string {
   const protocol = request.protocol || 'http';
   const host = request.get('host');
-  return `${protocol}://${host}`;
+
+  const modelSegment = `/${pluralizedModel}`;
+  const mountPath = request.baseUrl ?? '';
+  const prefix = mountPath.endsWith(modelSegment) ? mountPath.slice(0, -modelSegment.length) : '';
+
+  return `${protocol}://${host}${prefix}`;
 }
 
+// Kept as a name because twelve `getId(...)` call sites read it. It is now a
+// thin delegate: there is exactly ONE normalisation of a URL id in the repo,
+// and it is the exported one a consumer can import.
+//
+// "of a URL id" is the load-bearing qualifier, and it is the same one
+// src/normalize-record-id.ts:18 carries. Three copies of the coercion survive
+// at this head — src/orm-request.ts (the create-response path),
+// src/postgres/postgres-db.ts and src/mysql/mysql-db.ts — but each normalises a
+// RESPONSE id (`response?.data?.id`), not a URL id, and each omits this
+// function's `if (!id) return ''` guard. They are tracked as
+// abofs/stonyx-orm#282 and enumerated by name in AC-5's allowlist
+// (test/integration/readme-sample-test.ts).
+//
+// A second implementation of the URL-id normalisation here is the defect
+// abofs/stonyx-orm#270 exists to remove — see src/normalize-record-id.ts.
 function getId(params: { id?: string; [key: string]: unknown }): string | number {
-  const id = params.id;
-  if (!id) return '';
-  if (isNaN(id as unknown as number)) return id;
-
-  return parseInt(id);
+  return normalizeRecordId(params.id);
 }
 
 function buildResponse(
@@ -278,7 +335,7 @@ export default class OrmRequest extends Request {
       if (accessFilter) recordsToReturn = recordsToReturn.filter(accessFilter as (record: OrmRecord) => boolean);
       if (queryFilterPredicate) recordsToReturn = recordsToReturn.filter(queryFilterPredicate as (record: OrmRecord) => boolean);
 
-      const baseUrl = getBaseUrl(request);
+      const baseUrl = getBaseUrl(request, pluralizedModel);
       const data = recordsToReturn.map(record => record.toJSON?.({ fields: modelFields, baseUrl }));
 
       return buildResponse(data, request.query?.include, recordsToReturn, {
@@ -294,7 +351,7 @@ export default class OrmRequest extends Request {
       const fieldsMap = parseFields(request.query);
       const modelFields = fieldsMap.get(pluralizedModel) || fieldsMap.get(model);
 
-      const baseUrl = getBaseUrl(request);
+      const baseUrl = getBaseUrl(request, pluralizedModel);
       return buildResponse(record.toJSON?.({ fields: modelFields, baseUrl }), request.query?.include, record, {
         links: { self: `${baseUrl}/${pluralizedModel}/${request.params.id}` },
         baseUrl
@@ -330,7 +387,7 @@ export default class OrmRequest extends Request {
       }
 
       const recordAttributes = id !== undefined ? { id, ...sanitizedAttributes } : sanitizedAttributes;
-      const created = createRecord(model, recordAttributes as { [key: string]: unknown }, { serialize: false });
+      const created = createRecord(model, recordAttributes as { [key: string]: unknown }, { serialize: false, _skipAutoPersist: true });
       const record = isOrmRecord(created) ? created : null;
       if (!record) return 500;
 
@@ -368,7 +425,7 @@ export default class OrmRequest extends Request {
           }
         }
         if (Object.keys(relUpdates).length > 0) {
-          updateRecord(record as never, relUpdates);
+          updateRecord(record as never, relUpdates, { _skipAutoPersist: true });
         }
       }
 
@@ -376,7 +433,7 @@ export default class OrmRequest extends Request {
     };
 
     const deleteHandler: HandlerFn = ({ params }) => {
-      store.remove(model, getId(params));
+      store.remove(model, getId(params), { _skipAutoPersist: true });
       return 204;
     };
 
@@ -443,7 +500,12 @@ export default class OrmRequest extends Request {
       // Execute main handler
       const response = await handler(request, state);
 
-      // Persist to SQL database for write operations
+      // Set context.record for update BEFORE persist so SQL drivers can read it
+      if (operation === 'update' && (response as JsonApiResponse)?.data) {
+        context.record = store.get(this.model, getId(request.params));
+      }
+
+      // Persist to SQL database for all write operations (create/update/delete)
       const sqlDb = Orm.instance.sqlDb;
       if (sqlDb && WRITE_OPERATIONS.has(operation)) {
         await sqlDb.persist(operation, this.model, context, response);
@@ -461,8 +523,6 @@ export default class OrmRequest extends Request {
         const responseData = (response as { data: { id: string | number } }).data;
         const recordId = isNaN(responseData.id as unknown as number) ? responseData.id : parseInt(responseData.id as string);
         context.record = store.get(this.model, recordId);
-      } else if (operation === 'update' && (response as JsonApiResponse)?.data) {
-        context.record = store.get(this.model, getId(request.params));
       } else if (operation === 'delete') {
         // For delete, the record may no longer exist, but we have oldState
         context.recordId = getId(request.params);
@@ -499,7 +559,7 @@ export default class OrmRequest extends Request {
         if (!record) return 404;
 
         const relatedData = record.__relationships[relationshipName];
-        const baseUrl = getBaseUrl(request);
+        const baseUrl = getBaseUrl(request, pluralizedModel);
 
         let data: unknown;
         if (info.isArray) {
@@ -523,7 +583,7 @@ export default class OrmRequest extends Request {
         if (!record) return 404;
 
         const relatedData = record.__relationships[relationshipName];
-        const baseUrl = getBaseUrl(request);
+        const baseUrl = getBaseUrl(request, pluralizedModel);
 
         let data: unknown;
         if (info.isArray) {
@@ -572,6 +632,37 @@ export default class OrmRequest extends Request {
   }
 
   auth(request: OrmRequest$, state: { [key: string]: unknown }): number | undefined {
+    // abofs/stonyx-orm#270 — hand the predicate the value the record is
+    // ACTUALLY resolved by, rather than the raw text it was parsed from.
+    //
+    // This is the default path precisely because it requires the consumer to
+    // remember nothing. Exporting `normalizeRecordId` alone would leave the
+    // framework owning resolution while asking every consumer to call one
+    // function, with no signal when they forget — the same silent fail-open,
+    // one step over. The export is the escape hatch for hooks and custom
+    // handlers; this line is the contract.
+    //
+    // `request.params` is deliberately NOT mutated. Twelve `getId(...)` call
+    // sites read it, and `_withHooks` assigns `params: request.params` onto the
+    // hook context, so every consumer hook reads the same object. Changing
+    // `params.id` from string to number underneath them is a silent behaviour
+    // change on paths this issue is not about.
+    //
+    // Synchronous by necessity: @stonyx/rest-server calls auth() without
+    // awaiting it.
+    //
+    // The `undefined` branch is a PRESENCE check, not a second normalisation:
+    // it answers "does this route carry an :id at all", which is how both
+    // documented samples tell a collection request from a record request. It
+    // cannot be folded into normalizeRecordId, because that function must keep
+    // returning '' for a falsy id — `store.get(key, undefined)` returns the
+    // whole model Map rather than a record (abofs/stonyx-orm#167, pinned by
+    // test/unit/store-get-falsy-id-test.ts), so the resolution path depends on
+    // the '' it returns today.
+    const rawId = request.params?.id;
+
+    request.recordId = rawId === undefined ? undefined : normalizeRecordId(rawId);
+
     const access = this.access(request);
 
     if (!access) return 403;

package/src/postgres/column-rules.ts

@@ -0,0 +1,37 @@
+import type { ColumnRule } from '../column-conversion.js';
+
+/**
+ * PostgreSQL's write-path conversion rule (#292).
+ *
+ * JSONB columns must receive a JSON *string*. node-pg's `prepareValue`
+ * (`pg/lib/utils.js`) checks `Array.isArray` before its object fallback, so a
+ * raw array is rendered as a Postgres array literal (`{"1","2","3"}`) rather
+ * than JSON. `[]` becomes `{}`, which is valid JSON, so it lands silently
+ * corrupted with no error.
+ *
+ * Two behaviours this rule must keep:
+ *
+ *  - `typeof value === 'string'` passes through untouched. A JSONB column
+ *    already holding a JSON string (`'{"a":1}'`) must store the object it
+ *    describes; an unconditional `JSON.stringify` would double-encode it.
+ *  - An `undefined` column type is a no-op. `updated_at`, `id` and FK columns
+ *    are added outside the attribute loop and are absent from
+ *    `schema.columns`.
+ *
+ * The `'JSONB'` sentinel is the exact literal emitted by
+ * `src/postgres/type-map.ts`, the only module that produces the values in
+ * `schema.columns`. The two producing functions are `getPgType` and
+ * `getVectorType` (`src/postgres/schema-introspector.ts` writes `columns[key]`
+ * at exactly three sites, all routed through those two); `mysqlTypeToPg` is not
+ * exported and is reachable only from inside `getPgType`.
+ *
+ * One documented gap: `getPgType` returns a consumer transform's declared
+ * `pgType` verbatim, so a consumer writing `pgType: 'jsonb'` puts a value in
+ * `schema.columns` that type-map never authored and this exact-match sentinel
+ * misses. It misses identically on the create path, pre-fix and post-fix --
+ * pre-existing and symmetric, not introduced here.
+ */
+export const postgresJsonRule: ColumnRule = (value, columnType) =>
+  columnType === 'JSONB' && typeof value !== 'string'
+    ? JSON.stringify(value)
+    : value;

package/src/postgres/connection.ts

@@ -8,6 +8,7 @@ interface PgConfig {
   password: string;
   database: string;
   connectionLimit: number;
+  [key: string]: unknown;
 }
 
 let pool: PgPool | null = null;
@@ -20,15 +21,18 @@ export async function getPool(pgConfig: PgConfig, extensions: string[] = ['vecto
 
   const { default: pg } = await import('pg');
 
+  const { host, port, user, password, database, connectionLimit, migrationsDir, migrationsTable, autoMigrate, ...poolOpts } = pgConfig;
+
   pool = new pg.Pool({
-    host: pgConfig.host,
-    port: pgConfig.port,
-    user: pgConfig.user,
-    password: pgConfig.password,
-    database: pgConfig.database,
-    max: pgConfig.connectionLimit,
+    host,
+    port,
+    user,
+    password,
+    database,
+    max: connectionLimit,
     idleTimeoutMillis: 30000,
     connectionTimeoutMillis: 10000,
+    ...poolOpts,
   });
 
   // Enable requested PostgreSQL extensions