Chages

.gitignore

@@ -5,10 +5,13 @@ node_modules
 out
 dist
 *.tgz
+*.sql
+
 
 # code coverage
 coverage
 *.lcov
+*table.json
 
 # logs
 logs
@@ -32,3 +35,6 @@ report.[0-9]_.[0-9]_.[0-9]_.[0-9]_.json
 
 # Finder (MacOS) folder config
 .DS_Store
+
+# Documentation lives in its own repository (ElectronSz/stabilize-docs)
+stabilize-docs/

.gitmodules

@@ -0,0 +1,3 @@
+[submodule "stabilize-cli"]
+	path = stabilize-cli
+	url = https://github.com/ElectronSz/stabilize-cli.git

CHANGELOG.md

@@ -6,6 +6,54 @@ All notable changes to this project will be documented in this file.
 
 - Further features and improvements coming soon.
 
+## [2.2.1] - 2026-09-15
+
+Documentation-only release. No library code changed from 2.2.0.
+
+### Changed
+
+- **`README.md` now documents SQL Server.** 2.2.0 added Microsoft SQL Server as a first-class dialect but the README still described only PostgreSQL, MySQL and SQLite — there was not a single mention of SQL Server anywhere in it. Added a SQL Server badge, corrected the intro and the feature list, and documented that `DBType.MSSQL` uses the `mssql` v12 driver with T-SQL-specific query generation.
+- Corrected the CLI badge (was pinned at 2.1.0) and pointed the license badge at `stabilize-orm`; it previously linked to the `stabilize-cli` repository.
+
+## [2.2.0] - 2026-09-14
+
+### Added
+
+- **Microsoft SQL Server support** - SQL Server is now a first-class dialect alongside SQLite, MySQL and PostgreSQL. `DBType.MSSQL` selects the `mssql` driver (v12), and the placeholder rewriter maps the library's internal `?` placeholders to `@param0`-style named parameters automatically.
+  - T-SQL-specific query generation: `OUTPUT INSERTED.*` for reads-after-write, `OFFSET … FETCH` pagination with an `ORDER BY (SELECT NULL)` fallback when no ordering is supplied, and `MERGE INTO … USING … WHEN MATCHED / WHEN NOT MATCHED … OUTPUT INSERTED.*` for upserts.
+  - Schema generation uses `IF OBJECT_ID(…) IS NULL CREATE TABLE` in place of `CREATE TABLE IF NOT EXISTS`, and a `sys.indexes` probe in place of `CREATE INDEX IF NOT EXISTS`, neither of which SQL Server has.
+  - `poolStats()` reports borrowed / available / size for SQL Server connection pools.
+  - Connection strings use a comma between host and port: `Server=host,port;User Id=sa;Password=…;Database=…;TrustServerCertificate=true`.
+- **Model Relationships** - `OneToOne`, `ManyToOne`, `OneToMany` and `ManyToMany` are declared in the model configuration and eager-loaded either by passing `relations` to a finder or by chaining `.withRelations()` onto the query builder. Relations are resolved in batched follow-up queries rather than per-row.
+- **Many-to-many link management** - `attach()`, `detach()` and `sync()` edit a join table directly, without loading either side of the relationship into memory.
+- **`validateAll()`** - runs every validator on an entity and returns all failures at once, instead of throwing on the first one. `validate()` keeps its fail-fast behaviour.
+- **`findOrFail()` / `firstOrFail()`** - throw a `StabilizeError` with code `NOT_FOUND_ERROR` rather than returning `null`.
+- **Raw clause builders** - `.orderByRaw()`, `.groupByRaw()` and `.havingRaw()` accept expressions that are not bare column names.
+- **`auto-migrate.ts`** - a GORM-style `AutoMigrate` that creates missing tables, adds missing columns and adds missing indexes. It never drops columns and never alters column types - it only adds.
+- **Column encryption** - `utils/encryption.ts` exposes `encrypt()` / `decrypt()`; marking a column as encrypted transparently encrypts on write and decrypts on read, including for rows served from cache.
+- **Connection retry with exponential backoff and jitter** - `retryAttempts` (default 3) and `retryDelay` (default 1000ms) on the client config. Retries apply to read-only statements only; a failed write is never silently replayed.
+- **`auto-migrate.primary-key` and `client.retry` test suites**, plus eight integration suites (`sqlite`, `mysql`, `mariadb`, `postgres`, `mssql`, `models`, `relations`, `write-paths`) that run against real database servers and skip themselves when no server answers.
+
+### Fixed
+
+- **`after*` hooks and both delete hooks never ran.** `getHooks()` looked the model up via `proto.constructor`, but reads return plain rows from the driver, so the prototype was `Object` and the metadata lookup found nothing. The lookup now resolves the caller's model when one is passed, and hooks take a `model` argument for exactly this case. Any code that relied on `afterCreate` / `afterUpdate` / `afterDelete` firing will now see it fire.
+- **`getRepository()` opened a second, disconnected Redis connection on every call.** Repositories are now memoised per model, so repeated calls return the same instance instead of leaking a connection that nothing closed and `getCacheStats()` never counted.
+- **Cached `find()` results were written without their relations.** Rows are now hydrated and have relations loaded before being written to the cache, so a cache hit returns the same shape as a cache miss.
+- **`orderByRaw` / `groupByRaw` / `havingRaw` parameter handling**, and `OFFSET`-based pagination on dialects that require an `ORDER BY`.
+
+### Changed
+
+- `DBType` gains an `MSSQL` member. Since `DBType` is a string enum, existing persisted values are unaffected.
+- `mssql` (v12) is now a runtime dependency.
+
+### Known Issues
+
+- **SQL Server `IDENTITY` columns reject an explicit primary key.** `create({ id: 99 })` fails with *"Cannot insert explicit value for identity column in table 'x' when IDENTITY_INSERT is set to OFF"*, and `upsert({ id: 98 }, ["title"])` fails with *"Cannot update identity column 'id'"*. The cause is that the generated `MERGE` statement places the identity column in both the `UPDATE SET` list and the `INSERT` column list, and SQL Server permits neither. Letting the database assign the key works correctly on every path. A fix requires the upsert builder to skip identity columns in the `UPDATE SET` list and rely on `OUTPUT INSERTED.*` for the generated value; it is documented rather than patched here so that the behaviour change is not shipped silently in a minor release.
+- **`create()` with a JSON object throws on SQLite.** The SQLite write path binds values as given, while MySQL and SQL Server encode explicitly, so an object value reaches `bun:sqlite` unencoded and it rejects it with *"Binding expected string, TypedArray, boolean, number, bigint or null"*. Passing a pre-stringified value works on all three.
+- **Declared column `length`, `precision` and `scale` are not enforced** at write time - they appear in the generated DDL only, and SQLite ignores them entirely.
+- **`getValidators()` reports only `required` and `unique`**; other validators run during `validate()` but are not reflected in that summary.
+- **A `STRING` primary key is generated as `UUID PRIMARY KEY` on PostgreSQL** but as `VARCHAR(255)` / `NVARCHAR(255)` / `TEXT` on the other dialects, so the column types differ across dialects for the same model.
+
 ## [2.1.0] - 2026-04-02
 
 ### Added