fix: pytest warnings

Bumps [pygments](https://github.com/pygments/pygments) from 2.19.2 to 2.20.0.
- [Release notes](https://github.com/pygments/pygments/releases)
- [Changelog](https://github.com/pygments/pygments/blob/master/CHANGES)
- [Commits](https://github.com/pygments/pygments/compare/2.19.2...2.20.0)

---
updated-dependencies:
- dependency-name: pygments
  dependency-version: 2.20.0
  dependency-type: indirect
...

Signed-off-by: dependabot[bot] <support@github.com>
Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com>

.github/workflows/docs.yml

@@ -5,20 +5,15 @@ on:
     types: [published]
 
 permissions:
-  contents: read
-  pages: write
-  id-token: write
+  contents: write
 
 jobs:
   deploy:
-    environment:
-      name: github-pages
-      url: ${{ steps.deployment.outputs.page_url }}
     runs-on: ubuntu-latest
     steps:
-      - uses: actions/configure-pages@v5
-
       - uses: actions/checkout@v6
+        with:
+          fetch-depth: 0
 
       - name: Install uv
         uses: astral-sh/setup-uv@v7
@@ -28,11 +23,40 @@ jobs:
 
       - run: uv sync --group dev
 
-      - run: uv run zensical build --clean
+      - name: Configure git
+        run: |
+          git config user.name "github-actions[bot]"
+          git config user.email "github-actions[bot]@users.noreply.github.com"
 
-      - uses: actions/upload-pages-artifact@v4
-        with:
-          path: site
+      - name: Deploy documentation
+        run: |
+          VERSION=${GITHUB_REF_NAME#v}
+          MAJOR=$(echo "$VERSION" | cut -d. -f1)
+          DEPLOY_VERSION="v$(echo "$VERSION" | cut -d. -f1-2)"
 
-      - uses: actions/deploy-pages@v5
-        id: deployment
+          # On new major: consolidate previous major's feature versions into vX
+          PREV_MAJOR=$((MAJOR - 1))
+          OLD_FEATURE_VERSIONS=$(uv run mike list 2>/dev/null | grep -oE "^v${PREV_MAJOR}\.[0-9]+" || true)
+
+          if [ -n "$OLD_FEATURE_VERSIONS" ]; then
+            LATEST_PREV_TAG=$(git tag -l "v${PREV_MAJOR}.*" | sort -V | tail -1)
+
+            if [ -n "$LATEST_PREV_TAG" ]; then
+              git checkout "$LATEST_PREV_TAG" -- docs/ src/ zensical.toml
+              if ! grep -q '\[project\.extra\.version\]' zensical.toml; then
+                printf '\n[project.extra.version]\nprovider = "mike"\ndefault = "stable"\nalias = true\n' >> zensical.toml
+              fi
+              uv run mike deploy "v${PREV_MAJOR}"
+              git checkout HEAD -- docs/ src/ zensical.toml
+            fi
+
+            # Delete old feature versions
+            echo "$OLD_FEATURE_VERSIONS" | while read -r OLD_V; do
+              echo "Deleting $OLD_V"
+              uv run mike delete "$OLD_V"
+            done
+          fi
+
+          uv run mike deploy --update-aliases "$DEPLOY_VERSION" stable
+          uv run mike set-default stable
+          git push origin gh-pages

README.md

@@ -48,7 +48,8 @@ uv add "fastapi-toolsets[all]"
 - **Database**: Session management, transaction helpers, table locking, and polling-based row change detection
 - **Dependencies**: FastAPI dependency factories (`PathDependency`, `BodyDependency`) for automatic DB lookups from path or body parameters
 - **Fixtures**: Fixture system with dependency management, context support, and pytest integration
-- **Model Mixins**: SQLAlchemy mixins for common column patterns (`UUIDMixin`, `UUIDv7Mixin`, `CreatedAtMixin`, `UpdatedAtMixin`, `TimestampMixin`) and lifecycle callbacks (`WatchedFieldsMixin`, `@watch`) that fire after commit for insert, update, and delete events
+- **Model Mixins**: SQLAlchemy mixins for common column patterns (`UUIDMixin`, `UUIDv7Mixin`, `CreatedAtMixin`, `UpdatedAtMixin`, `TimestampMixin`)
+- **Lifecycle Events**: Post-commit event system (`EventSession`, `listens_for`) that dispatches async/sync callbacks for insert, update, and delete operations
 - **Standardized API Responses**: Consistent response format with `Response`, `ErrorResponse`, `PaginatedResponse`, `CursorPaginatedResponse` and `OffsetPaginatedResponse`.
 - **Exception Handling**: Structured error responses with automatic OpenAPI documentation
 - **Logging**: Logging configuration with uvicorn integration via `configure_logging` and `get_logger`

docs/examples/pagination-search.md

@@ -43,16 +43,16 @@ Declare `searchable_fields`, `facet_fields`, and `order_fields` once on [`CrudFa
 
 ## Routes
 
-```python title="routes.py:1:17"
---8<-- "docs_src/examples/pagination_search/routes.py:1:17"
+```python title="routes.py:1:16"
+--8<-- "docs_src/examples/pagination_search/routes.py:1:16"
 ```
 
 ### Offset pagination
 
 Best for admin panels or any UI that needs a total item count and numbered pages.
 
-```python title="routes.py:20:40"
---8<-- "docs_src/examples/pagination_search/routes.py:20:40"
+```python title="routes.py:19:37"
+--8<-- "docs_src/examples/pagination_search/routes.py:19:37"
 ```
 
 **Example request**
@@ -92,8 +92,8 @@ To skip the `COUNT(*)` query for better performance on large tables, pass `inclu
 
 Best for feeds, infinite scroll, or any high-throughput API where offset performance degrades.
 
-```python title="routes.py:43:63"
---8<-- "docs_src/examples/pagination_search/routes.py:43:63"
+```python title="routes.py:40:58"
+--8<-- "docs_src/examples/pagination_search/routes.py:40:58"
 ```
 
 **Example request**
@@ -132,8 +132,8 @@ Pass `next_cursor` as the `cursor` query parameter on the next request to advanc
 
 [`paginate()`](../module/crud.md#unified-paginate--both-strategies-on-one-endpoint) lets a single endpoint support both strategies via a `pagination_type` query parameter. The `pagination_type` field in the response acts as a discriminator for frontend tooling.
 
-```python title="routes.py:66:90"
---8<-- "docs_src/examples/pagination_search/routes.py:66:90"
+```python title="routes.py:61:79"
+--8<-- "docs_src/examples/pagination_search/routes.py:61:79"
 ```
 
 **Offset request** (default)

docs/index.md

@@ -48,7 +48,8 @@ uv add "fastapi-toolsets[all]"
 - **Database**: Session management, transaction helpers, table locking, and polling-based row change detection
 - **Dependencies**: FastAPI dependency factories (`PathDependency`, `BodyDependency`) for automatic DB lookups from path or body parameters
 - **Fixtures**: Fixture system with dependency management, context support, and pytest integration
-- **Model Mixins**: SQLAlchemy mixins for common column patterns (`UUIDMixin`, `UUIDv7Mixin`, `CreatedAtMixin`, `UpdatedAtMixin`, `TimestampMixin`) and lifecycle callbacks (`WatchedFieldsMixin`) that fire after commit for insert, update, and delete events.
+- **Model Mixins**: SQLAlchemy mixins for common column patterns (`UUIDMixin`, `UUIDv7Mixin`, `CreatedAtMixin`, `UpdatedAtMixin`, `TimestampMixin`).
+- **Lifecycle Events**: Post-commit event system (`EventSession`, `listens_for`) that dispatches async/sync callbacks for insert, update, and delete operations.
 - **Standardized API Responses**: Consistent response format with `Response`, `ErrorResponse`, `PaginatedResponse`, `CursorPaginatedResponse` and `OffsetPaginatedResponse`.
 - **Exception Handling**: Structured error responses with automatic OpenAPI documentation
 - **Logging**: Logging configuration with uvicorn integration via `configure_logging` and `get_logger`

docs/migration/v3.md

@@ -0,0 +1,117 @@
+# Migrating to v3.0
+
+This page covers every breaking change introduced in **v3.0** and the steps required to update your code.
+
+---
+
+## CRUD
+
+### Facet keys now always use the full relationship chain
+
+In `v2`, relationship facet fields used only the terminal column key (e.g. `"name"` for `Role.name`) and only prepended the relationship name when two facet fields shared the same column key. In `v3`, facet keys **always** include the full relationship chain joined by `__`, regardless of collisions.
+
+=== "Before (`v2`)"
+
+    ```
+    User.status -> status
+    (User.role, Role.name) -> name
+    (User.role, Role.permission, Permission.name) -> name
+    ```
+
+=== "Now (`v3`)"
+
+    ```
+    User.status -> status
+    (User.role, Role.name) -> role__name
+    (User.role, Role.permission, Permission.name) -> role__permission__name
+    ```
+
+---
+
+## Models
+
+The lifecycle event system has been rewritten. Callbacks are now registered with a module-level [`listens_for`](../reference/models.md#fastapi_toolsets.models.listens_for) decorator and dispatched by [`EventSession`](../reference/models.md#fastapi_toolsets.models.EventSession), replacing the mixin-based approach from `v2`.
+
+### `WatchedFieldsMixin` and `@watch` removed
+
+Importing `WatchedFieldsMixin` or `watch` will raise `ImportError`.
+
+Model method callbacks (`on_create`, `on_delete`, `on_update`) and the `@watch` decorator are replaced by:
+
+1. **`__watched_fields__`** — a plain class attribute to restrict which field changes trigger `UPDATE` events (replaces `@watch`).
+2. **`@listens_for`** — a module-level decorator to register callbacks for one or more [`ModelEvent`](../reference/models.md#fastapi_toolsets.models.ModelEvent) types (replaces `on_create` / `on_delete` / `on_update` methods).
+
+=== "Before (`v2`)"
+
+    ```python
+    from fastapi_toolsets.models import WatchedFieldsMixin, watch
+
+    @watch("status")
+    class Order(Base, UUIDMixin, WatchedFieldsMixin):
+        __tablename__ = "orders"
+
+        status: Mapped[str]
+
+        async def on_create(self):
+            await notify_new_order(self.id)
+
+        async def on_update(self, changes):
+            if "status" in changes:
+                await notify_status_change(self.id, changes["status"])
+
+        async def on_delete(self):
+            await notify_order_cancelled(self.id)
+    ```
+
+=== "Now (`v3`)"
+
+    ```python
+    from fastapi_toolsets.models import ModelEvent, UUIDMixin, listens_for
+
+    class Order(Base, UUIDMixin):
+        __tablename__ = "orders"
+        __watched_fields__ = ("status",)
+
+        status: Mapped[str]
+
+    @listens_for(Order, [ModelEvent.CREATE])
+    async def on_order_created(order: Order, event_type: ModelEvent, changes: None):
+        await notify_new_order(order.id)
+
+    @listens_for(Order, [ModelEvent.UPDATE])
+    async def on_order_updated(order: Order, event_type: ModelEvent, changes: dict):
+        if "status" in changes:
+            await notify_status_change(order.id, changes["status"])
+
+    @listens_for(Order, [ModelEvent.DELETE])
+    async def on_order_deleted(order: Order, event_type: ModelEvent, changes: None):
+        await notify_order_cancelled(order.id)
+    ```
+
+### `EventSession` now required
+
+Without `EventSession`, lifecycle callbacks will silently stop firing.
+
+Callbacks are now dispatched inside `EventSession.commit()` rather than via background tasks. Pass it as the session class when creating your session factory:
+
+=== "Before (`v2`)"
+
+    ```python
+    from sqlalchemy.ext.asyncio import async_sessionmaker, create_async_engine
+
+    engine = create_async_engine("postgresql+asyncpg://...")
+    SessionLocal = async_sessionmaker(engine, expire_on_commit=False)
+    ```
+
+=== "Now (`v3`)"
+
+    ```python
+    from sqlalchemy.ext.asyncio import async_sessionmaker, create_async_engine
+    from fastapi_toolsets.models import EventSession
+
+    engine = create_async_engine("postgresql+asyncpg://...")
+    SessionLocal = async_sessionmaker(engine, expire_on_commit=False, class_=EventSession)
+    ```
+
+!!! note
+    If you use `create_db_session` from `fastapi_toolsets.pytest`, the session already uses `EventSession` — no changes needed in tests.

docs/module/crud.md

@@ -36,7 +36,7 @@ class UserCrud(AsyncCrud[User]):
     default_load_options = [selectinload(User.role)]
 ```
 
-Subclassing [`AsyncCrud`](../reference/crud.md#fastapi_toolsets.crud.factory.AsyncCrud) directly is the preferred style when you need to add custom methods or when the configuration is complex enough to benefit from a named class body. 
+Subclassing [`AsyncCrud`](../reference/crud.md#fastapi_toolsets.crud.factory.AsyncCrud) directly is the preferred style when you need to add custom methods or when the configuration is complex enough to benefit from a named class body.
 
 ### Adding custom methods
 
@@ -159,18 +159,15 @@ Three pagination methods are available.  All return a typed response whose `pagi
 ### Offset pagination
 
 ```python
+from typing import Annotated
+from fastapi import Depends
+
 @router.get("")
 async def get_users(
     session: SessionDep,
-    items_per_page: int = 50,
-    page: int = 1,
+    params: Annotated[dict, Depends(UserCrud.offset_paginate_params())],
 ) -> OffsetPaginatedResponse[UserRead]:
-    return await UserCrud.offset_paginate(
-        session=session,
-        items_per_page=items_per_page,
-        page=page,
-        schema=UserRead,
-    )
+    return await UserCrud.offset_paginate(session=session, **params, schema=UserRead)
 ```
 
 The [`offset_paginate`](../reference/crud.md#fastapi_toolsets.crud.factory.AsyncCrud.offset_paginate) method returns an [`OffsetPaginatedResponse`](../reference/schemas.md#fastapi_toolsets.schemas.OffsetPaginatedResponse):
@@ -194,32 +191,13 @@ The [`offset_paginate`](../reference/crud.md#fastapi_toolsets.crud.factory.Async
 
 !!! info "Added in `v2.4.1`"
 
-By default `offset_paginate` runs two queries: one for the page items and one `COUNT(*)` for `total_count`. On large tables the `COUNT` can be expensive. Pass `include_total=False` to skip it:
+By default `offset_paginate` runs two queries: one for the page items and one `COUNT(*)` for `total_count`. On large tables the `COUNT` can be expensive. Pass `include_total=False` to `offset_paginate_params()` to skip it:
 
 ```python
-result = await UserCrud.offset_paginate(
-    session=session,
-    page=page,
-    items_per_page=items_per_page,
-    include_total=False,
-    schema=UserRead,
-)
-```
-
-#### Pagination params dependency
-
-!!! info "Added in `v2.4.1`"
-
-Use [`offset_params()`](../reference/crud.md#fastapi_toolsets.crud.factory.AsyncCrud.offset_params) to generate a FastAPI dependency that injects `page` and `items_per_page` from query parameters with configurable defaults and a `max_page_size` cap:
-
-```python
-from typing import Annotated
-from fastapi import Depends
-
 @router.get("")
-async def list_users(
+async def get_users(
     session: SessionDep,
-    params: Annotated[dict, Depends(UserCrud.offset_params(default_page_size=20, max_page_size=100))],
+    params: Annotated[dict, Depends(UserCrud.offset_paginate_params(include_total=False))],
 ) -> OffsetPaginatedResponse[UserRead]:
     return await UserCrud.offset_paginate(session=session, **params, schema=UserRead)
 ```
@@ -230,15 +208,9 @@ async def list_users(
 @router.get("")
 async def list_users(
     session: SessionDep,
-    cursor: str | None = None,
-    items_per_page: int = 20,
+    params: Annotated[dict, Depends(UserCrud.cursor_paginate_params())],
 ) -> CursorPaginatedResponse[UserRead]:
-    return await UserCrud.cursor_paginate(
-        session=session,
-        cursor=cursor,
-        items_per_page=items_per_page,
-        schema=UserRead,
-    )
+    return await UserCrud.cursor_paginate(session=session, **params, schema=UserRead)
 ```
 
 The [`cursor_paginate`](../reference/crud.md#fastapi_toolsets.crud.factory.AsyncCrud.cursor_paginate) method returns a [`CursorPaginatedResponse`](../reference/schemas.md#fastapi_toolsets.schemas.CursorPaginatedResponse):
@@ -291,24 +263,6 @@ PostCrud = CrudFactory(model=Post, cursor_column=Post.id)
 PostCrud = CrudFactory(model=Post, cursor_column=Post.created_at)
 ```
 
-#### Pagination params dependency
-
-!!! info "Added in `v2.4.1`"
-
-Use [`cursor_params()`](../reference/crud.md#fastapi_toolsets.crud.factory.AsyncCrud.cursor_params) to inject `cursor` and `items_per_page` from query parameters with a `max_page_size` cap:
-
-```python
-from typing import Annotated
-from fastapi import Depends
-
-@router.get("")
-async def list_users(
-    session: SessionDep,
-    params: Annotated[dict, Depends(UserCrud.cursor_params(default_page_size=20, max_page_size=100))],
-) -> CursorPaginatedResponse[UserRead]:
-    return await UserCrud.cursor_paginate(session=session, **params, schema=UserRead)
-```
-
 ### Unified endpoint (both strategies)
 
 !!! info "Added in `v2.3.0`"
@@ -316,25 +270,14 @@ async def list_users(
 [`paginate()`](../reference/crud.md#fastapi_toolsets.crud.factory.AsyncCrud.paginate) dispatches to `offset_paginate` or `cursor_paginate` based on a `pagination_type` query parameter, letting you expose **one endpoint** that supports both strategies.  The `pagination_type` field in the response tells clients which strategy was used, enabling frontend discriminated-union typing.
 
 ```python
-from fastapi_toolsets.crud import PaginationType
 from fastapi_toolsets.schemas import PaginatedResponse
 
 @router.get("")
 async def list_users(
     session: SessionDep,
-    pagination_type: PaginationType = PaginationType.OFFSET,
-    page: int = Query(1, ge=1, description="Current page (offset only)"),
-    cursor: str | None = Query(None, description="Cursor token (cursor only)"),
-    items_per_page: int = Query(20, ge=1, le=100),
+    params: Annotated[dict, Depends(UserCrud.paginate_params())],
 ) -> PaginatedResponse[UserRead]:
-    return await UserCrud.paginate(
-        session,
-        pagination_type=pagination_type,
-        page=page,
-        cursor=cursor,
-        items_per_page=items_per_page,
-        schema=UserRead,
-    )
+    return await UserCrud.paginate(session, **params, schema=UserRead)
 ```
 
 ```
@@ -342,25 +285,6 @@ GET /users?pagination_type=offset&page=2&items_per_page=10
 GET /users?pagination_type=cursor&cursor=eyJ2YWx1ZSI6...&items_per_page=10
 ```
 
-#### Pagination params dependency
-
-!!! info "Added in `v2.4.1`"
-
-Use [`paginate_params()`](../reference/crud.md#fastapi_toolsets.crud.factory.AsyncCrud.paginate_params) to inject all parameters at once with configurable defaults and a `max_page_size` cap:
-
-```python
-from typing import Annotated
-from fastapi import Depends
-from fastapi_toolsets.schemas import PaginatedResponse
-
-@router.get("")
-async def list_users(
-    session: SessionDep,
-    params: Annotated[dict, Depends(UserCrud.paginate_params(default_page_size=20, max_page_size=100))],
-) -> PaginatedResponse[UserRead]:
-    return await UserCrud.paginate(session, **params, schema=UserRead)
-```
-
 ## Search
 
 Two search strategies are available, both compatible with [`offset_paginate`](../reference/crud.md#fastapi_toolsets.crud.factory.AsyncCrud.offset_paginate) and [`cursor_paginate`](../reference/crud.md#fastapi_toolsets.crud.factory.AsyncCrud.cursor_paginate).
@@ -406,34 +330,18 @@ This allows searching with both [`offset_paginate`](../reference/crud.md#fastapi
 @router.get("")
 async def get_users(
     session: SessionDep,
-    items_per_page: int = 50,
-    page: int = 1,
-    search: str | None = None,
+    params: Annotated[dict, Depends(UserCrud.offset_paginate_params())],
 ) -> OffsetPaginatedResponse[UserRead]:
-    return await UserCrud.offset_paginate(
-        session=session,
-        items_per_page=items_per_page,
-        page=page,
-        search=search,
-        schema=UserRead,
-    )
+    return await UserCrud.offset_paginate(session=session, **params, schema=UserRead)
 ```
 
 ```python
 @router.get("")
 async def get_users(
     session: SessionDep,
-    cursor: str | None = None,
-    items_per_page: int = 50,
-    search: str | None = None,
+    params: Annotated[dict, Depends(UserCrud.cursor_paginate_params())],
 ) -> CursorPaginatedResponse[UserRead]:
-    return await UserCrud.cursor_paginate(
-        session=session,
-        items_per_page=items_per_page,
-        cursor=cursor,
-        search=search,
-        schema=UserRead,
-    )
+    return await UserCrud.cursor_paginate(session=session, **params, schema=UserRead)
 ```
 
 ### Faceted search
@@ -474,7 +382,7 @@ The distinct values are returned in the `filter_attributes` field of [`Paginated
   "filter_attributes": {
     "status": ["active", "inactive"],
     "country": ["DE", "FR", "US"],
-    "name": ["admin", "editor", "viewer"]
+    "role__name": ["admin", "editor", "viewer"]
   }
 }
 ```
@@ -482,11 +390,11 @@ The distinct values are returned in the `filter_attributes` field of [`Paginated
 Use `filter_by` to pass the client's chosen filter values directly — no need to build SQLAlchemy conditions by hand. Any unknown key raises [`InvalidFacetFilterError`](../reference/exceptions.md#fastapi_toolsets.exceptions.exceptions.InvalidFacetFilterError).
 
 !!! info "The keys in `filter_by` are the same keys the client received in `filter_attributes`."
-    Keys are normally the terminal `column.key` (e.g. `"name"` for `Role.name`). When two facet fields share the same column key (e.g. `(Build.project, Project.name)` and `(Build.os, Os.name)`), the relationship name is prepended automatically: `"project__name"` and `"os__name"`.
+    Keys use `__` as a separator for the full relationship chain. A direct column `User.status` produces `"status"`. A relationship tuple `(User.role, Role.name)` produces `"role__name"`. A deeper chain `(User.role, Role.permission, Permission.name)` produces `"role__permission__name"`.
 
 `filter_by` and `filters` can be combined — both are applied with AND logic.
 
-Use [`filter_params()`](../reference/crud.md#fastapi_toolsets.crud.factory.AsyncCrud.filter_params) to generate a dict with the facet filter values from the query parameters:
+Facet filtering is built into the consolidated params dependencies. When `filter=True` (the default), facet fields are exposed as query parameters and collected into `filter_by` automatically:
 
 ```python
 from typing import Annotated
@@ -501,13 +409,11 @@ UserCrud = CrudFactory(
 @router.get("", response_model_exclude_none=True)
 async def list_users(
     session: SessionDep,
-    page: int = 1,
-    filter_by: Annotated[dict[str, list[str]], Depends(UserCrud.filter_params())],
+    params: Annotated[dict, Depends(UserCrud.offset_paginate_params())],
 ) -> OffsetPaginatedResponse[UserRead]:
     return await UserCrud.offset_paginate(
         session=session,
-        page=page,
-        filter_by=filter_by,
+        **params,
         schema=UserRead,
     )
 ```
@@ -515,9 +421,9 @@ async def list_users(
 Both single-value and multi-value query parameters work:
 
 ```
-GET /users?status=active              → filter_by={"status": ["active"]}
-GET /users?status=active&country=FR   → filter_by={"status": ["active"], "country": ["FR"]}
-GET /users?role=admin&role=editor     → filter_by={"role": ["admin", "editor"]}  (IN clause)
+GET /users?status=active                      → filter_by={"status": ["active"]}
+GET /users?status=active&country=FR           → filter_by={"status": ["active"], "country": ["FR"]}
+GET /users?role__name=admin&role__name=editor → filter_by={"role__name": ["admin", "editor"]}  (IN clause)
 ```
 
 ## Sorting
@@ -536,20 +442,21 @@ UserCrud = CrudFactory(
 )
 ```
 
-Call [`order_params()`](../reference/crud.md#fastapi_toolsets.crud.factory.AsyncCrud.order_params) to generate a FastAPI dependency that maps the query parameters to an [`OrderByClause`](../reference/crud.md#fastapi_toolsets.crud.factory.OrderByClause) expression:
+Ordering is built into the consolidated params dependencies. When `order=True` (the default), `order_by` and `order` query parameters are exposed and resolved into an `OrderByClause` automatically:
 
 ```python
 from typing import Annotated
 
 from fastapi import Depends
-from fastapi_toolsets.crud import OrderByClause
 
 @router.get("")
 async def list_users(
     session: SessionDep,
-    order_by: Annotated[OrderByClause | None, Depends(UserCrud.order_params())],
+    params: Annotated[dict, Depends(UserCrud.offset_paginate_params(
+        default_order_field=User.created_at,
+    ))],
 ) -> OffsetPaginatedResponse[UserRead]:
-    return await UserCrud.offset_paginate(session=session, order_by=order_by, schema=UserRead)
+    return await UserCrud.offset_paginate(session=session, **params, schema=UserRead)
 ```
 
 The dependency adds two query parameters to the endpoint:
@@ -566,10 +473,10 @@ GET /users?order_by=name&order=desc  → ORDER BY users.name DESC
 
 An unknown `order_by` value raises [`InvalidOrderFieldError`](../reference/exceptions.md#fastapi_toolsets.exceptions.exceptions.InvalidOrderFieldError) (HTTP 422).
 
-You can also pass `order_fields` directly to `order_params()` to override the class-level defaults without modifying them:
+You can also pass `order_fields` directly to override the class-level defaults:
 
 ```python
-UserOrderParams = UserCrud.order_params(order_fields=[User.name])
+params = UserCrud.offset_paginate_params(order_fields=[User.name])
 ```
 
 ## Relationship loading
@@ -656,12 +563,11 @@ async def get_user(session: SessionDep, uuid: UUID) -> Response[UserRead]:
     )
 
 @router.get("")
-async def list_users(session: SessionDep, page: int = 1) -> OffsetPaginatedResponse[UserRead]:
-    return await crud.UserCrud.offset_paginate(
-        session=session,
-        page=page,
-        schema=UserRead,
-    )
+async def list_users(
+    session: SessionDep,
+    params: Annotated[dict, Depends(crud.UserCrud.offset_paginate_params())],
+) -> OffsetPaginatedResponse[UserRead]:
+    return await crud.UserCrud.offset_paginate(session=session, **params, schema=UserRead)
 ```
 
 The schema must have `from_attributes=True` (or inherit from [`PydanticBase`](../reference/schemas.md#fastapi_toolsets.schemas.PydanticBase)) so it can be built from SQLAlchemy model instances.

docs/module/fixtures.md

@@ -38,18 +38,20 @@ By context with [`load_fixtures_by_context`](../reference/fixtures.md#fastapi_to
 from fastapi_toolsets.fixtures import load_fixtures_by_context
 
 async with db_context() as session:
-    await load_fixtures_by_context(session=session, registry=fixtures, context=Context.TESTING)
+    await load_fixtures_by_context(session, fixtures, Context.TESTING)
 ```
 
-Directly with [`load_fixtures`](../reference/fixtures.md#fastapi_toolsets.fixtures.utils.load_fixtures):
+Directly by name with [`load_fixtures`](../reference/fixtures.md#fastapi_toolsets.fixtures.utils.load_fixtures):
 
 ```python
 from fastapi_toolsets.fixtures import load_fixtures
 
 async with db_context() as session:
-    await load_fixtures(session=session, registry=fixtures)
+    await load_fixtures(session, fixtures, "roles", "test_users")
 ```
 
+Both functions return a `dict[str, list[...]]` mapping each fixture name to the list of loaded instances.
+
 ## Contexts
 
 [`Context`](../reference/fixtures.md#fastapi_toolsets.fixtures.enum.Context) is an enum with predefined values:
@@ -58,10 +60,60 @@ async with db_context() as session:
 |---------|-------------|
 | `Context.BASE` | Core data required in all environments |
 | `Context.TESTING` | Data only loaded during tests |
+| `Context.DEVELOPMENT` | Data only loaded in development |
 | `Context.PRODUCTION` | Data only loaded in production |
 
 A fixture with no `contexts` defined takes `Context.BASE` by default.
 
+### Custom contexts
+
+Plain strings and any `Enum` subclass are accepted wherever a `Context` enum is expected.
+
+```python
+from enum import Enum
+
+class AppContext(str, Enum):
+    STAGING = "staging"
+    DEMO = "demo"
+
+@fixtures.register(contexts=[AppContext.STAGING])
+def staging_data():
+    return [Config(key="feature_x", enabled=True)]
+
+await load_fixtures_by_context(session, fixtures, AppContext.STAGING)
+```
+
+### Default context for a registry
+
+Pass `contexts` to `FixtureRegistry` to set a default for all fixtures registered in it:
+
+```python
+testing_registry = FixtureRegistry(contexts=[Context.TESTING])
+
+@testing_registry.register          # implicitly contexts=[Context.TESTING]
+def test_orders():
+    return [Order(id=1, total=99)]
+```
+
+### Same fixture name, multiple context variants
+
+The same fixture name may be registered under different (non-overlapping) context sets. When multiple contexts are loaded together, all matching variants are merged:
+
+```python
+@fixtures.register(contexts=[Context.BASE])
+def users():
+    return [User(id=1, username="admin")]
+
+@fixtures.register(contexts=[Context.TESTING])
+def users():
+    return [User(id=2, username="tester")]
+
+# loads both admin and tester
+await load_fixtures_by_context(session, fixtures, Context.BASE, Context.TESTING)
+```
+
+Registering two variants with overlapping context sets raises `ValueError`.
+
 ## Load strategies
 
 [`LoadStrategy`](../reference/fixtures.md#fastapi_toolsets.fixtures.enum.LoadStrategy) controls how the fixture loader handles rows that already exist:
@@ -69,20 +121,44 @@ A fixture with no `contexts` defined takes `Context.BASE` by default.
 | Strategy | Description |
 |----------|-------------|
 | `LoadStrategy.INSERT` | Insert only, fail on duplicates |
-| `LoadStrategy.UPSERT` | Insert or update on conflict |
-| `LoadStrategy.SKIP` | Skip rows that already exist |
+| `LoadStrategy.MERGE` | Insert or update on conflict (default) |
+| `LoadStrategy.SKIP_EXISTING` | Skip rows that already exist |
+
+```python
+await load_fixtures_by_context(
+    session, fixtures, Context.BASE, strategy=LoadStrategy.SKIP_EXISTING
+)
+```
 
 ## Merging registries
 
-Split fixtures definitions across modules and merge them:
+Split fixture definitions across modules and merge them:
 
 ```python
 from myapp.fixtures.dev import dev_fixtures
 from myapp.fixtures.prod import prod_fixtures
 
-fixtures = fixturesRegistry()
+fixtures = FixtureRegistry()
 fixtures.include_registry(registry=dev_fixtures)
 fixtures.include_registry(registry=prod_fixtures)
+```
+
+Fixtures with the same name are allowed as long as their context sets do not overlap. Conflicting contexts raise `ValueError`.
+
+## Looking up fixture instances
+
+[`get_obj_by_attr`](../reference/fixtures.md#fastapi_toolsets.fixtures.utils.get_obj_by_attr) retrieves a specific instance from a fixture function by attribute value — useful when building cross-fixture `depends_on` relationships:
+
+```python
+from fastapi_toolsets.fixtures import get_obj_by_attr
+
+@fixtures.register(depends_on=["roles"])
+def users():
+    admin_role = get_obj_by_attr(roles, "name", "admin")
+    return [User(id=1, username="alice", role_id=admin_role.id)]
+```
+
+Raises `StopIteration` if no matching instance is found.
 
 ## Pytest integration
 
@@ -111,7 +187,6 @@ async def test_user_can_login(fixture_users: list[User], fixture_roles: list[Rol
     ...
 ```
 
-
 The load order is resolved automatically from the `depends_on` declarations in your registry. Each generated fixture receives `db_session` as a dependency and returns the list of loaded model instances.
 
 ## CLI integration

docs/module/models.md

@@ -117,139 +117,118 @@ class Article(Base, UUIDMixin, TimestampMixin):
     title: Mapped[str]
 ```
 
-### [`WatchedFieldsMixin`](../reference/models.md#fastapi_toolsets.models.WatchedFieldsMixin)
+## Lifecycle events
 
-!!! info "Added in `v2.4`"
+The event system provides lifecycle callbacks that fire **after commit**. If the transaction rolls back, no callback fires.
 
-`WatchedFieldsMixin` provides lifecycle callbacks that fire **after commit** — meaning the row is durably persisted when your callback runs. If the transaction rolls back, no callback fires.
+### Setup
 
-Three callbacks are available, each corresponding to a [`ModelEvent`](../reference/models.md#fastapi_toolsets.models.ModelEvent) value:
-
-| Callback | Event | Trigger |
-|---|---|---|
-| `on_create()` | `ModelEvent.CREATE` | After `INSERT` |
-| `on_delete()` | `ModelEvent.DELETE` | After `DELETE` |
-| `on_update(changes)` | `ModelEvent.UPDATE` | After `UPDATE` on a watched field |
-
-Server-side defaults (e.g. `id`, `created_at`) are fully populated in all callbacks. All callbacks support both `async def` and plain `def`. Use `@watch` to restrict which fields trigger `on_update`:
-
-| Decorator | `on_update` behaviour |
-|---|---|
-| `@watch("status", "role")` | Only fires when `status` or `role` changes |
-| *(no decorator)* | Fires when **any** mapped field changes |
-
-`@watch` is inherited through the class hierarchy. If a subclass does not declare its own `@watch`, it uses the filter from the nearest decorated parent. Applying `@watch` on the subclass overrides the parent's filter:
+Event dispatch requires [`EventSession`](../reference/models.md#fastapi_toolsets.models.EventSession). Pass it as the session class when creating your session factory:
 
 ```python
-@watch("status")
-class Order(Base, UUIDMixin, WatchedFieldsMixin):
+from sqlalchemy.ext.asyncio import async_sessionmaker, create_async_engine
+from fastapi_toolsets.models import EventSession
+
+engine = create_async_engine("postgresql+asyncpg://...")
+SessionLocal = async_sessionmaker(engine, expire_on_commit=False, class_=EventSession)
+```
+
+!!! info "Callbacks fire on `session.commit()` only — not on savepoints."
+    Savepoints created by [`get_transaction`](db.md) or `begin_nested()` do **not**
+    trigger callbacks. All events accumulated across flushes are dispatched once
+    when the outermost `commit()` is called.
+
+### Events
+
+Three event types are available, each corresponding to a [`ModelEvent`](../reference/models.md#fastapi_toolsets.models.ModelEvent) value:
+
+| Event | Trigger |
+|---|---|
+| `ModelEvent.CREATE` | After `INSERT` commit |
+| `ModelEvent.DELETE` | After `DELETE` commit |
+| `ModelEvent.UPDATE` | After `UPDATE` commit on a watched field |
+
+!!! warning "Callbacks fire only for ORM-level changes. Rows updated via raw SQL (`UPDATE ... SET ...`) are not detected."
+
+### Watched fields
+
+Set `__watched_fields__` on the model to restrict which field changes trigger `UPDATE` events. It must be a `tuple[str, ...]` — any other type raises `TypeError`:
+
+| Class attribute | `UPDATE` behaviour |
+|---|---|
+| `__watched_fields__ = ("status", "role")` | Only fires when `status` or `role` changes |
+| *(not set)* | Fires when **any** mapped field changes |
+
+`__watched_fields__` is inherited through the class hierarchy via normal Python MRO. A subclass can override it:
+
+```python
+class Order(Base, UUIDMixin):
+    __watched_fields__ = ("status",)
     ...
 
 class UrgentOrder(Order):
-    # inherits @watch("status") — on_update fires only for status changes
+    # inherits __watched_fields__ = ("status",)
     ...
 
-@watch("priority")
 class PriorityOrder(Order):
-    # overrides parent — on_update fires only for priority changes
+    __watched_fields__ = ("priority",)
+    # overrides parent — UPDATE fires only for priority changes
     ...
 ```
 
-#### Option 1 — catch-all with `on_event`
+### Registering handlers
 
-Override `on_event` to handle all event types in one place. The specific methods delegate here by default:
+Register handlers with the [`listens_for`](../reference/models.md#fastapi_toolsets.models.listens_for) decorator. Every callback receives three arguments: the model instance, the [`ModelEvent`](../reference/models.md#fastapi_toolsets.models.ModelEvent) that triggered it, and a `changes` dict (`None` for `CREATE` and `DELETE`):
 
 ```python
-from fastapi_toolsets.models import ModelEvent, UUIDMixin, WatchedFieldsMixin, watch
+from fastapi_toolsets.models import ModelEvent, UUIDMixin, listens_for
 
-@watch("status")
-class Order(Base, UUIDMixin, WatchedFieldsMixin):
+class Order(Base, UUIDMixin):
     __tablename__ = "orders"
+    __watched_fields__ = ("status",)
 
     status: Mapped[str]
 
-    async def on_event(self, event: ModelEvent, changes: dict | None = None) -> None:
-        if event == ModelEvent.CREATE:
-            await notify_new_order(self.id)
-        elif event == ModelEvent.DELETE:
-            await notify_order_cancelled(self.id)
-        elif event == ModelEvent.UPDATE:
-            await notify_status_change(self.id, changes["status"])
+@listens_for(Order, [ModelEvent.CREATE])
+async def on_order_created(order: Order, event_type: ModelEvent, changes: None):
+    await notify_new_order(order.id)
+
+@listens_for(Order, [ModelEvent.DELETE])
+async def on_order_deleted(order: Order, event_type: ModelEvent, changes: None):
+    await notify_order_cancelled(order.id)
+
+@listens_for(Order, [ModelEvent.UPDATE])
+async def on_order_updated(order: Order, event_type: ModelEvent, changes: dict):
+    if "status" in changes:
+        await notify_status_change(order.id, changes["status"])
 ```
 
-#### Option 2 — targeted overrides
+Multiple handlers can be registered for the same model and event. Handlers registered on a parent class also fire for subclass instances.
 
-Override individual methods for more focused logic:
+A single handler can listen for multiple events at once. When `event_types` is omitted, the handler fires for all events:
 
 ```python
-@watch("status")
-class Order(Base, UUIDMixin, WatchedFieldsMixin):
-    __tablename__ = "orders"
+@listens_for(Order, [ModelEvent.CREATE, ModelEvent.UPDATE])
+async def on_order_changed(order: Order, event_type: ModelEvent, changes: dict | None):
+    await invalidate_cache(order.id)
 
-    status: Mapped[str]
-
-    async def on_create(self) -> None:
-        await notify_new_order(self.id)
-
-    async def on_delete(self) -> None:
-        await notify_order_cancelled(self.id)
-
-    async def on_update(self, changes: dict) -> None:
-        if "status" in changes:
-            old = changes["status"]["old"]
-            new = changes["status"]["new"]
-            await notify_status_change(self.id, old, new)
+@listens_for(Order)  # all events
+async def on_any_order_event(order: Order, event_type: ModelEvent, changes: dict | None):
+    await audit_log(order.id, event_type)
 ```
 
-#### Field changes format
+### Field changes format
 
-The `changes` dict maps each watched field that changed to `{"old": ..., "new": ...}`. Only fields that actually changed are included:
+The `changes` dict maps each watched field that changed to `{"old": ..., "new": ...}`. Only fields that actually changed are included. For `CREATE` and `DELETE` events, `changes` is `None`:
 
 ```python
+# CREATE / DELETE → changes is None
 # status changed   → {"status": {"old": "pending", "new": "shipped"}}
 # two fields changed → {"status": {...}, "assigned_to": {...}}
 ```
 
 !!! info "Multiple flushes in one transaction are merged: the earliest `old` and latest `new` are preserved, and `on_update` fires only once per commit."
 
-!!! warning "Callbacks fire only for ORM-level changes. Rows updated via raw SQL (`UPDATE ... SET ...`) are not detected."
-
-!!! warning "Callbacks fire when the **outermost active context** (savepoint or transaction) commits."
-    If you create several related objects using `CrudFactory.create` and need
-    callbacks to see all of them (including associations), wrap the whole
-    operation in a single [`get_transaction`](db.md) or [`lock_tables`](db.md)
-    block. Without it, each `create` call commits its own savepoint and
-    `on_create` fires before the remaining objects exist.
-
-    ```python
-    from fastapi_toolsets.db import get_transaction
-
-    async with get_transaction(session):
-        order = await OrderCrud.create(session, order_data)
-        item  = await ItemCrud.create(session, item_data)
-        await session.refresh(order, attribute_names=["items"])
-        order.items.append(item)
-    # on_create fires here for both order and item,
-    # with the full association already committed.
-    ```
-
-## Composing mixins
-
-All mixins can be combined in any order. The only constraint is that exactly one primary key must be defined — either via `UUIDMixin` or directly on the model.
-
-```python
-from fastapi_toolsets.models import UUIDMixin, TimestampMixin
-
-class Event(Base, UUIDMixin, TimestampMixin):
-    __tablename__ = "events"
-    name: Mapped[str]
-
-class Counter(Base, UpdatedAtMixin):
-    __tablename__ = "counters"
-    id: Mapped[int] = mapped_column(primary_key=True, autoincrement=True)
-    value: Mapped[int]
-```
-
 ---
 
 [:material-api: API Reference](../reference/models.md)

docs/module/pytest.md

@@ -79,9 +79,6 @@ The examples above are already compatible with parallel test execution with `pyt
 
 ## Cleaning up tables
 
-!!! warning
-    Since `V2.1.0` `cleanup_tables` now live in `fastapi_toolsets.db`. For backward compatibility the function is still available in `fastapi_toolsets.pytest`, but this will be remove in `V3.0.0`.
-
 If you want to manually clean up a database you can use [`cleanup_tables`](../reference/db.md#fastapi_toolsets.db.cleanup_tables), this will truncate all tables between tests for fast isolation:
 
 ```python

docs/reference/models.md

@@ -6,17 +6,19 @@ You can import them directly from `fastapi_toolsets.models`:
 
 ```python
 from fastapi_toolsets.models import (
+    EventSession,
     ModelEvent,
     UUIDMixin,
     UUIDv7Mixin,
     CreatedAtMixin,
     UpdatedAtMixin,
     TimestampMixin,
-    WatchedFieldsMixin,
-    watch,
+    listens_for,
 )
 ```
 
+## ::: fastapi_toolsets.models.EventSession
+
 ## ::: fastapi_toolsets.models.ModelEvent
 
 ## ::: fastapi_toolsets.models.UUIDMixin
@@ -29,6 +31,4 @@ from fastapi_toolsets.models import (
 
 ## ::: fastapi_toolsets.models.TimestampMixin
 
-## ::: fastapi_toolsets.models.WatchedFieldsMixin
-
-## ::: fastapi_toolsets.models.watch
+## ::: fastapi_toolsets.models.listens_for

docs_src/examples/pagination_search/routes.py

@@ -2,7 +2,6 @@ from typing import Annotated
 
 from fastapi import APIRouter, Depends
 
-from fastapi_toolsets.crud import OrderByClause
 from fastapi_toolsets.schemas import (
     CursorPaginatedResponse,
     OffsetPaginatedResponse,
@@ -22,21 +21,18 @@ async def list_articles_offset(
     session: SessionDep,
     params: Annotated[
         dict,
-        Depends(ArticleCrud.offset_params(default_page_size=20, max_page_size=100)),
+        Depends(
+            ArticleCrud.offset_paginate_params(
+                default_page_size=20,
+                max_page_size=100,
+                default_order_field=Article.created_at,
+            )
+        ),
     ],
-    filter_by: Annotated[dict[str, list[str]], Depends(ArticleCrud.filter_params())],
-    order_by: Annotated[
-        OrderByClause | None,
-        Depends(ArticleCrud.order_params(default_field=Article.created_at)),
-    ],
-    search: str | None = None,
 ) -> OffsetPaginatedResponse[ArticleRead]:
     return await ArticleCrud.offset_paginate(
         session=session,
         **params,
-        search=search,
-        filter_by=filter_by or None,
-        order_by=order_by,
         schema=ArticleRead,
     )
 
@@ -46,21 +42,18 @@ async def list_articles_cursor(
     session: SessionDep,
     params: Annotated[
         dict,
-        Depends(ArticleCrud.cursor_params(default_page_size=20, max_page_size=100)),
+        Depends(
+            ArticleCrud.cursor_paginate_params(
+                default_page_size=20,
+                max_page_size=100,
+                default_order_field=Article.created_at,
+            )
+        ),
     ],
-    filter_by: Annotated[dict[str, list[str]], Depends(ArticleCrud.filter_params())],
-    order_by: Annotated[
-        OrderByClause | None,
-        Depends(ArticleCrud.order_params(default_field=Article.created_at)),
-    ],
-    search: str | None = None,
 ) -> CursorPaginatedResponse[ArticleRead]:
     return await ArticleCrud.cursor_paginate(
         session=session,
         **params,
-        search=search,
-        filter_by=filter_by or None,
-        order_by=order_by,
         schema=ArticleRead,
     )
 
@@ -70,20 +63,17 @@ async def list_articles(
     session: SessionDep,
     params: Annotated[
         dict,
-        Depends(ArticleCrud.paginate_params(default_page_size=20, max_page_size=100)),
+        Depends(
+            ArticleCrud.paginate_params(
+                default_page_size=20,
+                max_page_size=100,
+                default_order_field=Article.created_at,
+            )
+        ),
     ],
-    filter_by: Annotated[dict[str, list[str]], Depends(ArticleCrud.filter_params())],
-    order_by: Annotated[
-        OrderByClause | None,
-        Depends(ArticleCrud.order_params(default_field=Article.created_at)),
-    ],
-    search: str | None = None,
 ) -> PaginatedResponse[ArticleRead]:
     return await ArticleCrud.paginate(
         session,
         **params,
-        search=search,
-        filter_by=filter_by or None,
-        order_by=order_by,
         schema=ArticleRead,
     )

pyproject.toml

@@ -80,8 +80,9 @@ tests = [
     "pytest>=8.0.0",
 ]
 docs = [
+    "mike",
     "mkdocstrings-python>=2.0.2",
-    "zensical>=0.0.23",
+    "zensical>=0.0.30",
 ]
 
 [build-system]
@@ -104,3 +105,6 @@ exclude_lines = [
     "if TYPE_CHECKING:",
     "raise NotImplementedError",
 ]
+
+[tool.uv.sources]
+mike = { git = "https://github.com/squidfunk/mike.git", tag = "2.2.0+zensical-0.1.0" }

src/fastapi_toolsets/crud/__init__.py

@@ -1,6 +1,11 @@
 """Generic async CRUD operations for SQLAlchemy models."""
 
-from ..exceptions import InvalidFacetFilterError, NoSearchableFieldsError
+from ..exceptions import (
+    InvalidFacetFilterError,
+    InvalidSearchColumnError,
+    NoSearchableFieldsError,
+    UnsupportedFacetTypeError,
+)
 from ..schemas import PaginationType
 from ..types import (
     FacetFieldType,
@@ -18,6 +23,7 @@ __all__ = [
     "FacetFieldType",
     "get_searchable_fields",
     "InvalidFacetFilterError",
+    "InvalidSearchColumnError",
     "JoinType",
     "M2MFieldType",
     "NoSearchableFieldsError",
@@ -25,4 +31,5 @@ __all__ = [
     "PaginationType",
     "SearchConfig",
     "SearchFieldType",
+    "UnsupportedFacetTypeError",
 ]

src/fastapi_toolsets/crud/factory.py

@@ -47,6 +47,7 @@ from .search import (
     build_filter_by,
     build_search_filters,
     facet_keys,
+    search_field_keys,
 )
 
 
@@ -263,118 +264,285 @@ class AsyncCrud(Generic[ModelType]):
         )
 
     @classmethod
-    def filter_params(
+    def _resolve_search_columns(
+        cls: type[Self],
+        search_fields: Sequence[SearchFieldType] | None,
+    ) -> list[str] | None:
+        """Return search column keys, or None if no searchable fields configured."""
+        fields = search_fields if search_fields is not None else cls.searchable_fields
+        if not fields:
+            return None
+        return search_field_keys(fields)
+
+    @classmethod
+    def _build_paginate_params(
         cls: type[Self],
         *,
-        facet_fields: Sequence[FacetFieldType] | None = None,
-    ) -> Callable[..., Awaitable[dict[str, list[str]]]]:
-        """Return a FastAPI dependency that collects facet filter values from query parameters.
+        pagination_params: list[inspect.Parameter],
+        pagination_fixed: dict[str, Any],
+        dep_name: str,
+        search: bool,
+        filter: bool,
+        order: bool,
+        search_fields: Sequence[SearchFieldType] | None,
+        facet_fields: Sequence[FacetFieldType] | None,
+        order_fields: Sequence[QueryableAttribute[Any]] | None,
+        default_order_field: QueryableAttribute[Any] | None,
+        default_order: Literal["asc", "desc"],
+    ) -> Callable[..., Awaitable[dict[str, Any]]]:
+        """Build a consolidated FastAPI dependency that merges pagination, search, filter, and order params."""
+        all_params: list[inspect.Parameter] = list(pagination_params)
+        pagination_param_names = tuple(p.name for p in pagination_params)
+        reserved_names: set[str] = set(pagination_param_names)
 
-        Args:
-            facet_fields: Override the facet fields for this dependency. Falls back to the
-                class-level ``facet_fields`` if not provided.
-
-        Returns:
-            An async dependency function named ``{Model}FilterParams`` that resolves to a
-            ``dict[str, list[str]]`` containing only the keys that were supplied in the
-            request (absent/``None`` parameters are excluded).
-
-        Raises:
-            ValueError: If no facet fields are configured on this CRUD class and none are
-                provided via ``facet_fields``.
-        """
-        fields = cls._resolve_facet_fields(facet_fields)
-        if not fields:
-            raise ValueError(
-                f"{cls.__name__} has no facet_fields configured. "
-                "Pass facet_fields= or set them on CrudFactory."
-            )
-        keys = facet_keys(fields)
-
-        async def dependency(**kwargs: Any) -> dict[str, list[str]]:
-            return {k: v for k, v in kwargs.items() if v is not None}
-
-        dependency.__name__ = f"{cls.model.__name__}FilterParams"
-        dependency.__signature__ = inspect.Signature(  # type: ignore[attr-defined]  # ty:ignore[unresolved-attribute]
-            parameters=[
-                inspect.Parameter(
-                    k,
-                    inspect.Parameter.KEYWORD_ONLY,
-                    annotation=list[str] | None,
-                    default=Query(default=None),
+        search_keys: list[str] | None = None
+        if search:
+            search_keys = cls._resolve_search_columns(search_fields)
+            if search_keys:
+                all_params.extend(
+                    [
+                        inspect.Parameter(
+                            "search",
+                            inspect.Parameter.KEYWORD_ONLY,
+                            annotation=str | None,
+                            default=Query(
+                                default=None, description="Search query string"
+                            ),
+                        ),
+                        inspect.Parameter(
+                            "search_column",
+                            inspect.Parameter.KEYWORD_ONLY,
+                            annotation=str | None,
+                            default=Query(
+                                default=None,
+                                description="Restrict search to a single column",
+                                enum=search_keys,
+                            ),
+                        ),
+                    ]
                 )
-                for k in keys
-            ]
-        )
+                reserved_names.update({"search", "search_column"})
 
+        filter_keys: list[str] | None = None
+        if filter:
+            resolved_facets = cls._resolve_facet_fields(facet_fields)
+            if resolved_facets:
+                filter_keys = facet_keys(resolved_facets)
+                for k in filter_keys:
+                    if k in reserved_names:
+                        raise ValueError(
+                            f"Facet field key {k!r} conflicts with a reserved "
+                            f"parameter name. Reserved names: {sorted(reserved_names)}"
+                        )
+                all_params.extend(
+                    inspect.Parameter(
+                        k,
+                        inspect.Parameter.KEYWORD_ONLY,
+                        annotation=list[str] | None,
+                        default=Query(default=None),
+                    )
+                    for k in filter_keys
+                )
+                reserved_names.update(filter_keys)
+
+        order_field_map: dict[str, QueryableAttribute[Any]] | None = None
+        order_valid_keys: list[str] | None = None
+        if order:
+            resolved_order = (
+                order_fields if order_fields is not None else cls.order_fields
+            )
+            if resolved_order:
+                order_field_map = {f.key: f for f in resolved_order}
+                order_valid_keys = sorted(order_field_map.keys())
+                all_params.extend(
+                    [
+                        inspect.Parameter(
+                            "order_by",
+                            inspect.Parameter.KEYWORD_ONLY,
+                            annotation=str | None,
+                            default=Query(
+                                None,
+                                description=f"Field to order by. Valid values: {order_valid_keys}",
+                                enum=order_valid_keys,
+                            ),
+                        ),
+                        inspect.Parameter(
+                            "order",
+                            inspect.Parameter.KEYWORD_ONLY,
+                            annotation=Literal["asc", "desc"],
+                            default=Query(default_order, description="Sort direction"),
+                        ),
+                    ]
+                )
+
+        async def dependency(**kwargs: Any) -> dict[str, Any]:
+            result: dict[str, Any] = dict(pagination_fixed)
+            for name in pagination_param_names:
+                result[name] = kwargs[name]
+
+            if search_keys is not None:
+                search_val = kwargs.get("search")
+                if search_val is not None:
+                    result["search"] = search_val
+                search_col_val = kwargs.get("search_column")
+                if search_col_val is not None:
+                    result["search_column"] = search_col_val
+
+            if filter_keys is not None:
+                filter_by = {
+                    k: kwargs[k] for k in filter_keys if kwargs.get(k) is not None
+                }
+                result["filter_by"] = filter_by or None
+
+            if order_field_map is not None:
+                order_by_val = kwargs.get("order_by")
+                order_dir = kwargs.get("order", default_order)
+                if order_by_val is None:
+                    field = default_order_field
+                elif order_by_val not in order_field_map:
+                    raise InvalidOrderFieldError(order_by_val, order_valid_keys or [])
+                else:
+                    field = order_field_map[order_by_val]
+                if field is not None:
+                    result["order_by"] = (
+                        field.asc() if order_dir == "asc" else field.desc()
+                    )
+                else:
+                    result["order_by"] = None
+
+            return result
+
+        dependency.__name__ = dep_name
+        dependency.__signature__ = inspect.Signature(  # type: ignore[attr-defined]  # ty:ignore[unresolved-attribute]
+            parameters=all_params,
+        )
         return dependency
 
     @classmethod
-    def offset_params(
+    def offset_paginate_params(
         cls: type[Self],
         *,
         default_page_size: int = 20,
         max_page_size: int = 100,
         include_total: bool = True,
+        search: bool = True,
+        filter: bool = True,
+        order: bool = True,
+        search_fields: Sequence[SearchFieldType] | None = None,
+        facet_fields: Sequence[FacetFieldType] | None = None,
+        order_fields: Sequence[QueryableAttribute[Any]] | None = None,
+        default_order_field: QueryableAttribute[Any] | None = None,
+        default_order: Literal["asc", "desc"] = "asc",
     ) -> Callable[..., Awaitable[dict[str, Any]]]:
-        """Return a FastAPI dependency that collects offset pagination params from query params.
+        """Return a FastAPI dependency that collects all params for :meth:`offset_paginate`.
 
         Args:
-            default_page_size: Default value for the ``items_per_page`` query parameter.
-            max_page_size: Maximum allowed value for ``items_per_page`` (enforced via
-                ``le`` on the ``Query``).
-            include_total: Server-side flag forwarded as-is to ``include_total`` in
-                :meth:`offset_paginate`. Not exposed as a query parameter.
+            default_page_size: Default ``items_per_page`` value.
+            max_page_size: Maximum ``items_per_page`` value.
+            include_total: Whether to include total count (not a query param).
+            search: Enable search query parameters.
+            filter: Enable facet filter query parameters.
+            order: Enable order query parameters.
+            search_fields: Override searchable fields.
+            facet_fields: Override facet fields.
+            order_fields: Override order fields.
+            default_order_field: Default field to order by when ``order_by`` is absent.
+            default_order: Default sort direction.
 
         Returns:
-            An async dependency that resolves to a dict with ``page``,
-            ``items_per_page``, and ``include_total`` keys, ready to be
-            unpacked into :meth:`offset_paginate`.
+            An async dependency that resolves to a dict ready to be unpacked
+            into :meth:`offset_paginate`.
         """
-
-        async def dependency(
-            page: int = Query(1, ge=1, description="Page number (1-indexed)"),
-            items_per_page: int = _page_size_query(default_page_size, max_page_size),
-        ) -> dict[str, Any]:
-            return {
-                "page": page,
-                "items_per_page": items_per_page,
-                "include_total": include_total,
-            }
-
-        dependency.__name__ = f"{cls.model.__name__}OffsetParams"
-        return dependency
+        pagination_params = [
+            inspect.Parameter(
+                "page",
+                inspect.Parameter.KEYWORD_ONLY,
+                annotation=int,
+                default=Query(1, ge=1, description="Page number (1-indexed)"),
+            ),
+            inspect.Parameter(
+                "items_per_page",
+                inspect.Parameter.KEYWORD_ONLY,
+                annotation=int,
+                default=_page_size_query(default_page_size, max_page_size),
+            ),
+        ]
+        return cls._build_paginate_params(
+            pagination_params=pagination_params,
+            pagination_fixed={"include_total": include_total},
+            dep_name=f"{cls.model.__name__}OffsetPaginateParams",
+            search=search,
+            filter=filter,
+            order=order,
+            search_fields=search_fields,
+            facet_fields=facet_fields,
+            order_fields=order_fields,
+            default_order_field=default_order_field,
+            default_order=default_order,
+        )
 
     @classmethod
-    def cursor_params(
+    def cursor_paginate_params(
         cls: type[Self],
         *,
         default_page_size: int = 20,
         max_page_size: int = 100,
+        search: bool = True,
+        filter: bool = True,
+        order: bool = True,
+        search_fields: Sequence[SearchFieldType] | None = None,
+        facet_fields: Sequence[FacetFieldType] | None = None,
+        order_fields: Sequence[QueryableAttribute[Any]] | None = None,
+        default_order_field: QueryableAttribute[Any] | None = None,
+        default_order: Literal["asc", "desc"] = "asc",
     ) -> Callable[..., Awaitable[dict[str, Any]]]:
-        """Return a FastAPI dependency that collects cursor pagination params from query params.
+        """Return a FastAPI dependency that collects all params for :meth:`cursor_paginate`.
 
         Args:
-            default_page_size: Default value for the ``items_per_page`` query parameter.
-            max_page_size: Maximum allowed value for ``items_per_page`` (enforced via
-                ``le`` on the ``Query``).
+            default_page_size: Default ``items_per_page`` value.
+            max_page_size: Maximum ``items_per_page`` value.
+            search: Enable search query parameters.
+            filter: Enable facet filter query parameters.
+            order: Enable order query parameters.
+            search_fields: Override searchable fields.
+            facet_fields: Override facet fields.
+            order_fields: Override order fields.
+            default_order_field: Default field to order by when ``order_by`` is absent.
+            default_order: Default sort direction.
 
         Returns:
-            An async dependency that resolves to a dict with ``cursor`` and
-            ``items_per_page`` keys, ready to be unpacked into
-            :meth:`cursor_paginate`.
+            An async dependency that resolves to a dict ready to be unpacked
+            into :meth:`cursor_paginate`.
         """
-
-        async def dependency(
-            cursor: str | None = Query(
-                None, description="Cursor token from a previous response"
+        pagination_params = [
+            inspect.Parameter(
+                "cursor",
+                inspect.Parameter.KEYWORD_ONLY,
+                annotation=str | None,
+                default=Query(
+                    None, description="Cursor token from a previous response"
+                ),
             ),
-            items_per_page: int = _page_size_query(default_page_size, max_page_size),
-        ) -> dict[str, Any]:
-            return {"cursor": cursor, "items_per_page": items_per_page}
-
-        dependency.__name__ = f"{cls.model.__name__}CursorParams"
-        return dependency
+            inspect.Parameter(
+                "items_per_page",
+                inspect.Parameter.KEYWORD_ONLY,
+                annotation=int,
+                default=_page_size_query(default_page_size, max_page_size),
+            ),
+        ]
+        return cls._build_paginate_params(
+            pagination_params=pagination_params,
+            pagination_fixed={},
+            dep_name=f"{cls.model.__name__}CursorPaginateParams",
+            search=search,
+            filter=filter,
+            order=order,
+            search_fields=search_fields,
+            facet_fields=facet_fields,
+            order_fields=order_fields,
+            default_order_field=default_order_field,
+            default_order=default_order,
+        )
 
     @classmethod
     def paginate_params(
@@ -384,102 +552,81 @@ class AsyncCrud(Generic[ModelType]):
         max_page_size: int = 100,
         default_pagination_type: PaginationType = PaginationType.OFFSET,
         include_total: bool = True,
-    ) -> Callable[..., Awaitable[dict[str, Any]]]:
-        """Return a FastAPI dependency that collects all pagination params from query params.
-
-        Args:
-            default_page_size: Default value for the ``items_per_page`` query parameter.
-            max_page_size: Maximum allowed value for ``items_per_page`` (enforced via
-                ``le`` on the ``Query``).
-            default_pagination_type: Default pagination strategy.
-            include_total: Server-side flag forwarded as-is to ``include_total`` in
-                :meth:`paginate`. Not exposed as a query parameter.
-
-        Returns:
-            An async dependency that resolves to a dict with ``pagination_type``,
-            ``page``, ``cursor``, ``items_per_page``, and ``include_total`` keys,
-            ready to be unpacked into :meth:`paginate`.
-        """
-
-        async def dependency(
-            pagination_type: PaginationType = Query(
-                default_pagination_type, description="Pagination strategy"
-            ),
-            page: int = Query(
-                1, ge=1, description="Page number (1-indexed, offset only)"
-            ),
-            cursor: str | None = Query(
-                None, description="Cursor token from a previous response (cursor only)"
-            ),
-            items_per_page: int = _page_size_query(default_page_size, max_page_size),
-        ) -> dict[str, Any]:
-            return {
-                "pagination_type": pagination_type,
-                "page": page,
-                "cursor": cursor,
-                "items_per_page": items_per_page,
-                "include_total": include_total,
-            }
-
-        dependency.__name__ = f"{cls.model.__name__}PaginateParams"
-        return dependency
-
-    @classmethod
-    def order_params(
-        cls: type[Self],
-        *,
+        search: bool = True,
+        filter: bool = True,
+        order: bool = True,
+        search_fields: Sequence[SearchFieldType] | None = None,
+        facet_fields: Sequence[FacetFieldType] | None = None,
         order_fields: Sequence[QueryableAttribute[Any]] | None = None,
-        default_field: QueryableAttribute[Any] | None = None,
+        default_order_field: QueryableAttribute[Any] | None = None,
         default_order: Literal["asc", "desc"] = "asc",
-    ) -> Callable[..., Awaitable[OrderByClause | None]]:
-        """Return a FastAPI dependency that resolves order query params into an order_by clause.
+    ) -> Callable[..., Awaitable[dict[str, Any]]]:
+        """Return a FastAPI dependency that collects all params for :meth:`paginate`.
 
         Args:
-            order_fields: Override the allowed order fields. Falls back to the class-level
-                ``order_fields`` if not provided.
-            default_field: Field to order by when ``order_by`` query param is absent.
-                If ``None`` and no ``order_by`` is provided, no ordering is applied.
-            default_order: Default order direction when ``order`` is absent
-                (``"asc"`` or ``"desc"``).
+            default_page_size: Default ``items_per_page`` value.
+            max_page_size: Maximum ``items_per_page`` value.
+            default_pagination_type: Default pagination strategy.
+            include_total: Whether to include total count (not a query param).
+            search: Enable search query parameters.
+            filter: Enable facet filter query parameters.
+            order: Enable order query parameters.
+            search_fields: Override searchable fields.
+            facet_fields: Override facet fields.
+            order_fields: Override order fields.
+            default_order_field: Default field to order by when ``order_by`` is absent.
+            default_order: Default sort direction.
 
         Returns:
-            An async dependency function named ``{Model}OrderParams`` that resolves to an
-            ``OrderByClause`` (or ``None``). Pass it to ``Depends()`` in your route.
-
-        Raises:
-            ValueError: If no order fields are configured on this CRUD class and none are
-                provided via ``order_fields``.
-            InvalidOrderFieldError: When the request provides an unknown ``order_by`` value.
+            An async dependency that resolves to a dict ready to be unpacked
+            into :meth:`paginate`.
         """
-        fields = order_fields if order_fields is not None else cls.order_fields
-        if not fields:
-            raise ValueError(
-                f"{cls.__name__} has no order_fields configured. "
-                "Pass order_fields= or set them on CrudFactory."
-            )
-        field_map: dict[str, QueryableAttribute[Any]] = {f.key: f for f in fields}
-        valid_keys = sorted(field_map.keys())
-
-        async def dependency(
-            order_by: str | None = Query(
-                None, description=f"Field to order by. Valid values: {valid_keys}"
+        pagination_params = [
+            inspect.Parameter(
+                "pagination_type",
+                inspect.Parameter.KEYWORD_ONLY,
+                annotation=PaginationType,
+                default=Query(
+                    default_pagination_type, description="Pagination strategy"
+                ),
             ),
-            order: Literal["asc", "desc"] = Query(
-                default_order, description="Sort direction"
+            inspect.Parameter(
+                "page",
+                inspect.Parameter.KEYWORD_ONLY,
+                annotation=int,
+                default=Query(
+                    1, ge=1, description="Page number (1-indexed, offset only)"
+                ),
             ),
-        ) -> OrderByClause | None:
-            if order_by is None:
-                if default_field is None:
-                    return None
-                field = default_field
-            elif order_by not in field_map:
-                raise InvalidOrderFieldError(order_by, valid_keys)
-            else:
-                field = field_map[order_by]
-            return field.asc() if order == "asc" else field.desc()
-
-        dependency.__name__ = f"{cls.model.__name__}OrderParams"
-        return dependency
+            inspect.Parameter(
+                "cursor",
+                inspect.Parameter.KEYWORD_ONLY,
+                annotation=str | None,
+                default=Query(
+                    None,
+                    description="Cursor token from a previous response (cursor only)",
+                ),
+            ),
+            inspect.Parameter(
+                "items_per_page",
+                inspect.Parameter.KEYWORD_ONLY,
+                annotation=int,
+                default=_page_size_query(default_page_size, max_page_size),
+            ),
+        ]
+        return cls._build_paginate_params(
+            pagination_params=pagination_params,
+            pagination_fixed={"include_total": include_total},
+            dep_name=f"{cls.model.__name__}PaginateParams",
+            search=search,
+            filter=filter,
+            order=order,
+            search_fields=search_fields,
+            facet_fields=facet_fields,
+            order_fields=order_fields,
+            default_order_field=default_order_field,
+            default_order=default_order,
+        )
 
     @overload
     @classmethod
@@ -1056,6 +1203,7 @@ class AsyncCrud(Generic[ModelType]):
         include_total: bool = True,
         search: str | SearchConfig | None = None,
         search_fields: Sequence[SearchFieldType] | None = None,
+        search_column: str | None = None,
         facet_fields: Sequence[FacetFieldType] | None = None,
         filter_by: dict[str, Any] | BaseModel | None = None,
         schema: type[BaseModel],
@@ -1075,6 +1223,7 @@ class AsyncCrud(Generic[ModelType]):
                 ``pagination.total_count`` will be ``None``.
             search: Search query string or SearchConfig object
             search_fields: Fields to search in (overrides class default)
+            search_column: Restrict search to a single column key.
             facet_fields: Columns to compute distinct values for (overrides class default)
             filter_by: Dict of {column_key: value} to filter by declared facet fields.
                 Keys must match the column.key of a facet field. Scalar → equality,
@@ -1097,6 +1246,7 @@ class AsyncCrud(Generic[ModelType]):
                 search,
                 search_fields=search_fields,
                 default_fields=cls.searchable_fields,
+                search_column=search_column,
             )
             filters.extend(search_filters)
             search_joins.extend(new_search_joins)
@@ -1153,6 +1303,7 @@ class AsyncCrud(Generic[ModelType]):
         filter_attributes = await cls._build_filter_attributes(
             session, facet_fields, filters, search_joins
         )
+        search_columns = cls._resolve_search_columns(search_fields)
 
         return OffsetPaginatedResponse(
             data=items,
@@ -1163,6 +1314,7 @@ class AsyncCrud(Generic[ModelType]):
                 has_more=has_more,
             ),
             filter_attributes=filter_attributes,
+            search_columns=search_columns,
         )
 
     @classmethod
@@ -1179,6 +1331,7 @@ class AsyncCrud(Generic[ModelType]):
         items_per_page: int = 20,
         search: str | SearchConfig | None = None,
         search_fields: Sequence[SearchFieldType] | None = None,
+        search_column: str | None = None,
         facet_fields: Sequence[FacetFieldType] | None = None,
         filter_by: dict[str, Any] | BaseModel | None = None,
         schema: type[BaseModel],
@@ -1199,6 +1352,7 @@ class AsyncCrud(Generic[ModelType]):
             items_per_page: Number of items per page (default 20).
             search: Search query string or SearchConfig object.
             search_fields: Fields to search in (overrides class default).
+            search_column: Restrict search to a single column key.
             facet_fields: Columns to compute distinct values for (overrides class default).
             filter_by: Dict of {column_key: value} to filter by declared facet fields.
                 Keys must match the column.key of a facet field. Scalar → equality,
@@ -1238,6 +1392,7 @@ class AsyncCrud(Generic[ModelType]):
                 search,
                 search_fields=search_fields,
                 default_fields=cls.searchable_fields,
+                search_column=search_column,
             )
             filters.extend(search_filters)
             search_joins.extend(new_search_joins)
@@ -1308,6 +1463,7 @@ class AsyncCrud(Generic[ModelType]):
         filter_attributes = await cls._build_filter_attributes(
             session, facet_fields, filters, search_joins
         )
+        search_columns = cls._resolve_search_columns(search_fields)
 
         return CursorPaginatedResponse(
             data=items,
@@ -1318,6 +1474,7 @@ class AsyncCrud(Generic[ModelType]):
                 has_more=has_more,
             ),
             filter_attributes=filter_attributes,
+            search_columns=search_columns,
         )
 
     @overload
@@ -1338,6 +1495,7 @@ class AsyncCrud(Generic[ModelType]):
         include_total: bool = ...,
         search: str | SearchConfig | None = ...,
         search_fields: Sequence[SearchFieldType] | None = ...,
+        search_column: str | None = ...,
         facet_fields: Sequence[FacetFieldType] | None = ...,
         filter_by: dict[str, Any] | BaseModel | None = ...,
         schema: type[BaseModel],
@@ -1361,6 +1519,7 @@ class AsyncCrud(Generic[ModelType]):
         include_total: bool = ...,
         search: str | SearchConfig | None = ...,
         search_fields: Sequence[SearchFieldType] | None = ...,
+        search_column: str | None = ...,
         facet_fields: Sequence[FacetFieldType] | None = ...,
         filter_by: dict[str, Any] | BaseModel | None = ...,
         schema: type[BaseModel],
@@ -1383,6 +1542,7 @@ class AsyncCrud(Generic[ModelType]):
         include_total: bool = True,
         search: str | SearchConfig | None = None,
         search_fields: Sequence[SearchFieldType] | None = None,
+        search_column: str | None = None,
         facet_fields: Sequence[FacetFieldType] | None = None,
         filter_by: dict[str, Any] | BaseModel | None = None,
         schema: type[BaseModel],
@@ -1410,6 +1570,7 @@ class AsyncCrud(Generic[ModelType]):
                 only applies when ``pagination_type`` is ``OFFSET``.
             search: Search query string or :class:`.SearchConfig` object.
             search_fields: Fields to search in (overrides class default).
+            search_column: Restrict search to a single column key.
             facet_fields: Columns to compute distinct values for (overrides
                 class default).
             filter_by: Dict of ``{column_key: value}`` to filter by declared
@@ -1438,6 +1599,7 @@ class AsyncCrud(Generic[ModelType]):
                     items_per_page=items_per_page,
                     search=search,
                     search_fields=search_fields,
+                    search_column=search_column,
                     facet_fields=facet_fields,
                     filter_by=filter_by,
                     schema=schema,
@@ -1457,6 +1619,7 @@ class AsyncCrud(Generic[ModelType]):
                     include_total=include_total,
                     search=search,
                     search_fields=search_fields,
+                    search_column=search_column,
                     facet_fields=facet_fields,
                     filter_by=filter_by,
                     schema=schema,
@@ -1488,7 +1651,7 @@ def CrudFactory(
             responses. Supports direct columns (``User.status``) and relationship tuples
             (``(User.role, Role.name)``). Can be overridden per call.
         order_fields: Optional list of model attributes that callers are allowed to order by
-            via ``order_params()``. Can be overridden per call.
+            via ``offset_paginate_params()``. Can be overridden per call.
         m2m_fields: Optional mapping for many-to-many relationships.
             Maps schema field names (containing lists of IDs) to
             SQLAlchemy relationship attributes.

src/fastapi_toolsets/crud/search.py

@@ -2,17 +2,32 @@
 
 import asyncio
 import functools
-from collections import Counter
 from collections.abc import Sequence
 from dataclasses import dataclass, replace
 from typing import TYPE_CHECKING, Any, Literal
 
-from sqlalchemy import String, and_, or_, select
+from sqlalchemy import String, and_, func, or_, select
 from sqlalchemy.ext.asyncio import AsyncSession
 from sqlalchemy.orm import DeclarativeBase
 from sqlalchemy.orm.attributes import InstrumentedAttribute
+from sqlalchemy.types import (
+    ARRAY,
+    Boolean,
+    Date,
+    DateTime,
+    Enum,
+    Integer,
+    Numeric,
+    Time,
+    Uuid,
+)
 
-from ..exceptions import InvalidFacetFilterError, NoSearchableFieldsError
+from ..exceptions import (
+    InvalidFacetFilterError,
+    InvalidSearchColumnError,
+    NoSearchableFieldsError,
+    UnsupportedFacetTypeError,
+)
 from ..types import FacetFieldType, SearchFieldType
 
 if TYPE_CHECKING:
@@ -82,6 +97,7 @@ def build_search_filters(
     search: str | SearchConfig,
     search_fields: Sequence[SearchFieldType] | None = None,
     default_fields: Sequence[SearchFieldType] | None = None,
+    search_column: str | None = None,
 ) -> tuple[list["ColumnElement[bool]"], list[InstrumentedAttribute[Any]]]:
     """Build SQLAlchemy filter conditions for search.
 
@@ -90,6 +106,8 @@ def build_search_filters(
         search: Search string or SearchConfig
         search_fields: Fields specified per-call (takes priority)
         default_fields: Default fields (from ClassVar)
+        search_column: Optional key to narrow search to a single field.
+            Must match one of the resolved search field keys.
 
     Returns:
         Tuple of (filter_conditions, joins_needed)
@@ -116,6 +134,14 @@ def build_search_filters(
     if not fields:
         raise NoSearchableFieldsError(model)
 
+    # Narrow to a single column when search_column is specified
+    if search_column is not None:
+        keys = search_field_keys(fields)
+        index = {k: f for k, f in zip(keys, fields)}
+        if search_column not in index:
+            raise InvalidSearchColumnError(search_column, sorted(index))
+        fields = [index[search_column]]
+
     query = config.query.strip()
     filters: list[ColumnElement[bool]] = []
     joins: list[InstrumentedAttribute[Any]] = []
@@ -150,8 +176,13 @@ def build_search_filters(
         return filters, joins
 
 
+def search_field_keys(fields: Sequence[SearchFieldType]) -> list[str]:
+    """Return a human-readable key for each search field."""
+    return facet_keys(fields)
+
+
 def facet_keys(facet_fields: Sequence[FacetFieldType]) -> list[str]:
-    """Return a key for each facet field, disambiguating duplicate column keys.
+    """Return a key for each facet field.
 
     Args:
         facet_fields: Sequence of facet fields — either direct columns or
@@ -160,22 +191,12 @@ def facet_keys(facet_fields: Sequence[FacetFieldType]) -> list[str]:
     Returns:
         A list of string keys, one per facet field, in the same order.
     """
-    raw: list[tuple[str, str | None]] = []
+    keys: list[str] = []
     for field in facet_fields:
         if isinstance(field, tuple):
-            rel = field[-2]
-            column = field[-1]
-            raw.append((column.key, rel.key))
+            keys.append("__".join(el.key for el in field))
         else:
-            raw.append((field.key, None))
-
-    counts = Counter(col_key for col_key, _ in raw)
-    keys: list[str] = []
-    for col_key, rel_key in raw:
-        if counts[col_key] > 1 and rel_key is not None:
-            keys.append(f"{rel_key}__{col_key}")
-        else:
-            keys.append(col_key)
+            keys.append(field.key)
     return keys
 
 
@@ -212,7 +233,14 @@ async def build_facets(
             rels = ()
             column = field
 
-        q = select(column).select_from(model).distinct()
+        col_type = column.property.columns[0].type
+        is_array = isinstance(col_type, ARRAY)
+
+        if is_array:
+            unnested = func.unnest(column).label(column.key)
+            q = select(unnested).select_from(model).distinct()
+        else:
+            q = select(column).select_from(model).distinct()
 
         # Apply base joins (already done on main query, but needed here independently)
         for rel in base_joins or []:
@@ -226,7 +254,10 @@ async def build_facets(
         if base_filters:
             q = q.where(and_(*base_filters))
 
-        q = q.order_by(column)
+        if is_array:
+            q = q.order_by(unnested)
+        else:
+            q = q.order_by(column)
         result = await session.execute(q)
         values = [row[0] for row in result.all() if row[0] is not None]
         return key, values
@@ -237,6 +268,10 @@ async def build_facets(
     return dict(pairs)
 
 
+_EQUALITY_TYPES = (String, Integer, Numeric, Date, DateTime, Time, Enum, Uuid)
+"""Column types that support equality / IN filtering in build_filter_by."""
+
+
 def build_filter_by(
     filter_by: dict[str, Any],
     facet_fields: Sequence[FacetFieldType],
@@ -282,9 +317,23 @@ def build_filter_by(
                 joins.append(rel)
                 added_join_keys.add(rel_key)
 
-        if isinstance(value, list):
-            filters.append(column.in_(value))
+        col_type = column.property.columns[0].type
+        if isinstance(col_type, ARRAY):
+            if isinstance(value, list):
+                filters.append(column.overlap(value))
+            else:
+                filters.append(column.any(value))
+        elif isinstance(col_type, Boolean):
+            if isinstance(value, list):
+                filters.append(column.in_(value))
+            else:
+                filters.append(column.is_(value))
+        elif isinstance(col_type, _EQUALITY_TYPES):
+            if isinstance(value, list):
+                filters.append(column.in_(value))
+            else:
+                filters.append(column == value)
         else:
-            filters.append(column == value)
+            raise UnsupportedFacetTypeError(key, type(col_type).__name__)
 
     return filters, joins

src/fastapi_toolsets/db.py

@@ -24,9 +24,12 @@ __all__ = [
 ]
 
 
+_SessionT = TypeVar("_SessionT", bound=AsyncSession)
+
+
 def create_db_dependency(
-    session_maker: async_sessionmaker[AsyncSession],
-) -> Callable[[], AsyncGenerator[AsyncSession, None]]:
+    session_maker: async_sessionmaker[_SessionT],
+) -> Callable[[], AsyncGenerator[_SessionT, None]]:
     """Create a FastAPI dependency for database sessions.
 
     Creates a dependency function that yields a session and auto-commits
@@ -54,7 +57,7 @@ def create_db_dependency(
         ```
     """
 
-    async def get_db() -> AsyncGenerator[AsyncSession, None]:
+    async def get_db() -> AsyncGenerator[_SessionT, None]:
         async with session_maker() as session:
             await session.connection()
             yield session
@@ -65,8 +68,8 @@ def create_db_dependency(
 
 
 def create_db_context(
-    session_maker: async_sessionmaker[AsyncSession],
-) -> Callable[[], AbstractAsyncContextManager[AsyncSession]]:
+    session_maker: async_sessionmaker[_SessionT],
+) -> Callable[[], AbstractAsyncContextManager[_SessionT]]:
     """Create a context manager for database sessions.
 
     Creates a context manager for use outside of FastAPI request handlers,

src/fastapi_toolsets/exceptions/__init__.py

@@ -7,9 +7,11 @@ from .exceptions import (
     ForbiddenError,
     InvalidFacetFilterError,
     InvalidOrderFieldError,
+    InvalidSearchColumnError,
     NoSearchableFieldsError,
     NotFoundError,
     UnauthorizedError,
+    UnsupportedFacetTypeError,
     generate_error_responses,
 )
 from .handler import init_exceptions_handlers
@@ -23,7 +25,9 @@ __all__ = [
     "init_exceptions_handlers",
     "InvalidFacetFilterError",
     "InvalidOrderFieldError",
+    "InvalidSearchColumnError",
     "NoSearchableFieldsError",
     "NotFoundError",
     "UnauthorizedError",
+    "UnsupportedFacetTypeError",
 ]

src/fastapi_toolsets/exceptions/exceptions.py

@@ -144,6 +144,61 @@ class InvalidFacetFilterError(ApiException):
         )
 
 
+class UnsupportedFacetTypeError(ApiException):
+    """Raised when a facet field has a column type not supported by filter_by."""
+
+    api_error = ApiError(
+        code=400,
+        msg="Unsupported Facet Type",
+        desc="The column type is not supported for facet filtering.",
+        err_code="FACET-TYPE-400",
+    )
+
+    def __init__(self, key: str, col_type: str) -> None:
+        """Initialize the exception.
+
+        Args:
+            key: The facet field key.
+            col_type: The unsupported column type name.
+        """
+        self.key = key
+        self.col_type = col_type
+        super().__init__(
+            desc=(
+                f"Facet field '{key}' has unsupported column type '{col_type}'. "
+                f"Supported types: String, Integer, Numeric, Boolean, "
+                f"Date, DateTime, Time, Enum, Uuid, ARRAY."
+            )
+        )
+
+
+class InvalidSearchColumnError(ApiException):
+    """Raised when search_column is not one of the configured searchable fields."""
+
+    api_error = ApiError(
+        code=400,
+        msg="Invalid Search Column",
+        desc="The requested search column is not a configured searchable field.",
+        err_code="SEARCH-COL-400",
+    )
+
+    def __init__(self, column: str, valid_columns: list[str]) -> None:
+        """Initialize the exception.
+
+        Args:
+            column: The unknown search column provided by the caller.
+            valid_columns: List of valid search column keys.
+        """
+        self.column = column
+        self.valid_columns = valid_columns
+        super().__init__(
+            desc=(
+                f"'{column}' is not a searchable column. "
+                f"Valid columns: {valid_columns}."
+            )
+        )
+
+
 class InvalidOrderFieldError(ApiException):
     """Raised when order_by contains a field not in the allowed order fields."""
 

src/fastapi_toolsets/exceptions/handler.py

@@ -122,7 +122,7 @@ def _format_validation_error(
     )
 
     return JSONResponse(
-        status_code=status.HTTP_422_UNPROCESSABLE_ENTITY,
+        status_code=status.HTTP_422_UNPROCESSABLE_CONTENT,
         content=error_response.model_dump(),
     )
 

src/fastapi_toolsets/fixtures/registry.py

@@ -2,6 +2,7 @@
 
 from collections.abc import Callable, Sequence
 from dataclasses import dataclass, field
+from enum import Enum
 from typing import Any, cast
 
 from sqlalchemy.orm import DeclarativeBase
@@ -12,6 +13,13 @@ from .enum import Context
 logger = get_logger()
 
 
+def _normalize_contexts(
+    contexts: list[str | Enum] | tuple[str | Enum, ...],
+) -> list[str]:
+    """Convert a sequence of any Enum subclass and/or plain strings to a list of strings."""
+    return [c.value if isinstance(c, Enum) else c for c in contexts]
+
+
 @dataclass
 class Fixture:
     """A fixture definition with metadata."""
@@ -50,26 +58,51 @@ class FixtureRegistry:
                 Post(id=1, title="Test", user_id=1),
             ]
         ```
+
+    Fixtures with the same name may be registered for **different** contexts.
+    When multiple contexts are loaded together, their instances are merged:
+
+        ```python
+        @fixtures.register(contexts=[Context.BASE])
+        def users():
+            return [User(id=1, username="admin")]
+
+        @fixtures.register(contexts=[Context.TESTING])
+        def users():
+            return [User(id=2, username="tester")]
+        # load_fixtures_by_context(..., Context.BASE, Context.TESTING)
+        # → loads both User(admin) and User(tester) under the "users" name
+        ```
     """
 
     def __init__(
         self,
-        contexts: list[str | Context] | None = None,
+        contexts: list[str | Enum] | None = None,
     ) -> None:
-        self._fixtures: dict[str, Fixture] = {}
+        self._fixtures: dict[str, list[Fixture]] = {}
         self._default_contexts: list[str] | None = (
-            [c.value if isinstance(c, Context) else c for c in contexts]
-            if contexts
-            else None
+            _normalize_contexts(contexts) if contexts else None
         )
 
+    def _validate_no_context_overlap(self, name: str, new_contexts: list[str]) -> None:
+        """Raise ``ValueError`` if any existing variant for *name* overlaps."""
+        existing_variants = self._fixtures.get(name, [])
+        new_set = set(new_contexts)
+        for variant in existing_variants:
+            if set(variant.contexts) & new_set:
+                raise ValueError(
+                    f"Fixture '{name}' already exists in the current registry "
+                    f"with overlapping contexts. Use distinct context sets for "
+                    f"each variant of the same fixture name."
+                )
+
     def register(
         self,
         func: Callable[[], Sequence[DeclarativeBase]] | None = None,
         *,
         name: str | None = None,
         depends_on: list[str] | None = None,
-        contexts: list[str | Context] | None = None,
+        contexts: list[str | Enum] | None = None,
     ) -> Callable[..., Any]:
         """Register a fixture function.
 
@@ -79,7 +112,8 @@ class FixtureRegistry:
             func: Fixture function returning list of model instances
             name: Fixture name (defaults to function name)
             depends_on: List of fixture names this depends on
-            contexts: List of contexts this fixture belongs to
+            contexts: List of contexts this fixture belongs to.  Both
+                :class:`Context` enum values and plain strings are accepted.
 
         Example:
             ```python
@@ -90,7 +124,6 @@ class FixtureRegistry:
             @fixtures.register(depends_on=["roles"], contexts=[Context.TESTING])
             def test_users():
                 return [User(id=1, username="test", role_id=1)]
-            ```
         """
 
         def decorator(
@@ -98,19 +131,20 @@ class FixtureRegistry:
         ) -> Callable[[], Sequence[DeclarativeBase]]:
             fixture_name = name or cast(Any, fn).__name__
             if contexts is not None:
-                fixture_contexts = [
-                    c.value if isinstance(c, Context) else c for c in contexts
-                ]
+                fixture_contexts = _normalize_contexts(contexts)
             elif self._default_contexts is not None:
                 fixture_contexts = self._default_contexts
             else:
                 fixture_contexts = [Context.BASE.value]
 
-            self._fixtures[fixture_name] = Fixture(
-                name=fixture_name,
-                func=fn,
-                depends_on=depends_on or [],
-                contexts=fixture_contexts,
+            self._validate_no_context_overlap(fixture_name, fixture_contexts)
+            self._fixtures.setdefault(fixture_name, []).append(
+                Fixture(
+                    name=fixture_name,
+                    func=fn,
+                    depends_on=depends_on or [],
+                    contexts=fixture_contexts,
+                )
             )
             return fn
 
@@ -121,11 +155,14 @@ class FixtureRegistry:
     def include_registry(self, registry: "FixtureRegistry") -> None:
         """Include another `FixtureRegistry` in the same current `FixtureRegistry`.
 
+        Fixtures with the same name are allowed as long as their context sets
+        do not overlap.  Conflicting contexts raise :class:`ValueError`.
+
         Args:
             registry: The `FixtureRegistry` to include
 
         Raises:
-            ValueError: If a fixture name already exists in the current registry
+            ValueError: If a fixture name already exists with overlapping contexts
 
         Example:
             ```python
@@ -139,31 +176,73 @@ class FixtureRegistry:
             registry.include_registry(registry=dev_registry)
             ```
         """
-        for name, fixture in registry._fixtures.items():
-            if name in self._fixtures:
-                raise ValueError(
-                    f"Fixture '{name}' already exists in the current registry"
-                )
-            self._fixtures[name] = fixture
+        for name, variants in registry._fixtures.items():
+            for fixture in variants:
+                self._validate_no_context_overlap(name, fixture.contexts)
+                self._fixtures.setdefault(name, []).append(fixture)
 
     def get(self, name: str) -> Fixture:
-        """Get a fixture by name."""
+        """Get a fixture by name.
+
+        Raises:
+            KeyError: If no fixture with *name* is registered.
+            ValueError: If the fixture has multiple context variants — use
+                :meth:`get_variants` in that case.
+        """
         if name not in self._fixtures:
             raise KeyError(f"Fixture '{name}' not found")
-        return self._fixtures[name]
+        variants = self._fixtures[name]
+        if len(variants) > 1:
+            raise ValueError(
+                f"Fixture '{name}' has {len(variants)} context variants. "
+                f"Use get_variants('{name}') to retrieve them."
+            )
+        return variants[0]
+
+    def get_variants(self, name: str, *contexts: str | Enum) -> list[Fixture]:
+        """Return all registered variants for *name*, optionally filtered by context.
+
+        Args:
+            name: Fixture name.
+            *contexts: If given, only return variants whose context set
+                intersects with these values.  Both :class:`Context` enum
+                values and plain strings are accepted.
+
+        Returns:
+            List of matching :class:`Fixture` objects (may be empty when a
+            context filter is applied and nothing matches).
+
+        Raises:
+            KeyError: If no fixture with *name* is registered.
+        """
+        if name not in self._fixtures:
+            raise KeyError(f"Fixture '{name}' not found")
+        variants = self._fixtures[name]
+        if not contexts:
+            return list(variants)
+        context_values = set(_normalize_contexts(contexts))
+        return [v for v in variants if set(v.contexts) & context_values]
 
     def get_all(self) -> list[Fixture]:
-        """Get all registered fixtures."""
-        return list(self._fixtures.values())
+        """Get all registered fixtures (all variants of all names)."""
+        return [f for variants in self._fixtures.values() for f in variants]
 
-    def get_by_context(self, *contexts: str | Context) -> list[Fixture]:
+    def get_by_context(self, *contexts: str | Enum) -> list[Fixture]:
         """Get fixtures for specific contexts."""
-        context_values = {c.value if isinstance(c, Context) else c for c in contexts}
-        return [f for f in self._fixtures.values() if set(f.contexts) & context_values]
+        context_values = set(_normalize_contexts(contexts))
+        return [
+            f
+            for variants in self._fixtures.values()
+            for f in variants
+            if set(f.contexts) & context_values
+        ]
 
     def resolve_dependencies(self, *names: str) -> list[str]:
         """Resolve fixture dependencies in topological order.
 
+        When a fixture name has multiple context variants, the union of all
+        variants' ``depends_on`` lists is used.
+
         Args:
             *names: Fixture names to resolve
 
@@ -185,9 +264,20 @@ class FixtureRegistry:
                 raise ValueError(f"Circular dependency detected: {name}")
 
             visiting.add(name)
-            fixture = self.get(name)
+            variants = self._fixtures.get(name)
+            if variants is None:
+                raise KeyError(f"Fixture '{name}' not found")
 
-            for dep in fixture.depends_on:
+            # Union of depends_on across all variants, preserving first-seen order.
+            seen_deps: set[str] = set()
+            all_deps: list[str] = []
+            for variant in variants:
+                for dep in variant.depends_on:
+                    if dep not in seen_deps:
+                        all_deps.append(dep)
+                        seen_deps.add(dep)
+
+            for dep in all_deps:
                 visit(dep)
 
             visiting.remove(name)
@@ -199,7 +289,7 @@ class FixtureRegistry:
 
         return resolved
 
-    def resolve_context_dependencies(self, *contexts: str | Context) -> list[str]:
+    def resolve_context_dependencies(self, *contexts: str | Enum) -> list[str]:
         """Resolve all fixtures for contexts with dependencies.
 
         Args:
@@ -209,7 +299,9 @@ class FixtureRegistry:
             List of fixture names in load order
         """
         context_fixtures = self.get_by_context(*contexts)
-        names = [f.name for f in context_fixtures]
+        # Deduplicate names while preserving first-seen order (a name can
+        # appear multiple times if it has variants in different contexts).
+        names = list(dict.fromkeys(f.name for f in context_fixtures))
 
         all_deps: set[str] = set()
         for name in names:

src/fastapi_toolsets/fixtures/utils.py

@@ -1,8 +1,11 @@
 """Fixture loading utilities for database seeding."""
 
 from collections.abc import Callable, Sequence
+from enum import Enum
 from typing import Any
 
+from sqlalchemy import inspect as sa_inspect
+from sqlalchemy.dialects.postgresql import insert as pg_insert
 from sqlalchemy.ext.asyncio import AsyncSession
 from sqlalchemy.orm import DeclarativeBase
 
@@ -10,23 +13,177 @@ from ..db import get_transaction
 from ..logger import get_logger
 from ..types import ModelType
 from .enum import LoadStrategy
-from .registry import Context, FixtureRegistry
+from .registry import FixtureRegistry, _normalize_contexts
 
 logger = get_logger()
 
 
+def _instance_to_dict(instance: DeclarativeBase) -> dict[str, Any]:
+    """Extract column values from a model instance, skipping unset server-default columns."""
+    state = sa_inspect(instance)
+    state_dict = state.dict
+    result: dict[str, Any] = {}
+    for prop in state.mapper.column_attrs:
+        if prop.key not in state_dict:
+            continue
+        val = state_dict[prop.key]
+        if val is None:
+            col = prop.columns[0]
+
+            if (
+                col.server_default is not None
+                or (col.default is not None and col.default.is_callable)
+                or col.autoincrement is True
+            ):
+                continue
+        result[prop.key] = val
+    return result
+
+
+def _group_by_type(
+    instances: list[DeclarativeBase],
+) -> list[tuple[type[DeclarativeBase], list[DeclarativeBase]]]:
+    """Group instances by their concrete model class, preserving insertion order."""
+    groups: dict[type[DeclarativeBase], list[DeclarativeBase]] = {}
+    for instance in instances:
+        groups.setdefault(type(instance), []).append(instance)
+    return list(groups.items())
+
+
+def _group_by_column_set(
+    dicts: list[dict[str, Any]],
+    instances: list[DeclarativeBase],
+) -> list[tuple[list[dict[str, Any]], list[DeclarativeBase]]]:
+    """Group (dict, instance) pairs by their dict key sets."""
+    groups: dict[
+        frozenset[str], tuple[list[dict[str, Any]], list[DeclarativeBase]]
+    ] = {}
+    for d, inst in zip(dicts, instances):
+        key = frozenset(d)
+        if key not in groups:
+            groups[key] = ([], [])
+        groups[key][0].append(d)
+        groups[key][1].append(inst)
+    return list(groups.values())
+
+
+async def _batch_insert(
+    session: AsyncSession,
+    model_cls: type[DeclarativeBase],
+    instances: list[DeclarativeBase],
+) -> None:
+    """INSERT all instances — raises on conflict (no duplicate handling)."""
+    dicts = [_instance_to_dict(i) for i in instances]
+    for group_dicts, _ in _group_by_column_set(dicts, instances):
+        await session.execute(pg_insert(model_cls).values(group_dicts))
+
+
+async def _batch_merge(
+    session: AsyncSession,
+    model_cls: type[DeclarativeBase],
+    instances: list[DeclarativeBase],
+) -> None:
+    """UPSERT: insert new rows, update existing ones with the provided values."""
+    mapper = model_cls.__mapper__
+    pk_names = [col.name for col in mapper.primary_key]
+    pk_names_set = set(pk_names)
+    non_pk_cols = [
+        prop.key
+        for prop in mapper.column_attrs
+        if not any(col.name in pk_names_set for col in prop.columns)
+    ]
+
+    dicts = [_instance_to_dict(i) for i in instances]
+    for group_dicts, _ in _group_by_column_set(dicts, instances):
+        stmt = pg_insert(model_cls).values(group_dicts)
+
+        inserted_keys = set(group_dicts[0])
+        update_cols = [col for col in non_pk_cols if col in inserted_keys]
+
+        if update_cols:
+            stmt = stmt.on_conflict_do_update(
+                index_elements=pk_names,
+                set_={col: stmt.excluded[col] for col in update_cols},
+            )
+        else:
+            stmt = stmt.on_conflict_do_nothing(index_elements=pk_names)
+
+        await session.execute(stmt)
+
+
+async def _batch_skip_existing(
+    session: AsyncSession,
+    model_cls: type[DeclarativeBase],
+    instances: list[DeclarativeBase],
+) -> list[DeclarativeBase]:
+    """INSERT only rows that do not already exist; return the inserted ones."""
+    mapper = model_cls.__mapper__
+    pk_names = [col.name for col in mapper.primary_key]
+
+    no_pk: list[DeclarativeBase] = []
+    with_pk_pairs: list[tuple[DeclarativeBase, Any]] = []
+    for inst in instances:
+        pk = _get_primary_key(inst)
+        if pk is None:
+            no_pk.append(inst)
+        else:
+            with_pk_pairs.append((inst, pk))
+
+    loaded: list[DeclarativeBase] = list(no_pk)
+    if no_pk:
+        no_pk_dicts = [_instance_to_dict(i) for i in no_pk]
+        for group_dicts, _ in _group_by_column_set(no_pk_dicts, no_pk):
+            await session.execute(pg_insert(model_cls).values(group_dicts))
+
+    if with_pk_pairs:
+        with_pk = [i for i, _ in with_pk_pairs]
+        with_pk_dicts = [_instance_to_dict(i) for i in with_pk]
+        for group_dicts, group_insts in _group_by_column_set(with_pk_dicts, with_pk):
+            stmt = (
+                pg_insert(model_cls)
+                .values(group_dicts)
+                .on_conflict_do_nothing(index_elements=pk_names)
+            )
+            result = await session.execute(stmt.returning(*mapper.primary_key))
+            inserted_pks = {
+                row[0] if len(pk_names) == 1 else tuple(row) for row in result
+            }
+            loaded.extend(
+                inst
+                for inst, pk in zip(
+                    group_insts, [_get_primary_key(i) for i in group_insts]
+                )
+                if pk in inserted_pks
+            )
+
+    return loaded
+
+
 async def _load_ordered(
     session: AsyncSession,
     registry: FixtureRegistry,
     ordered_names: list[str],
     strategy: LoadStrategy,
+    contexts: tuple[str, ...] | None = None,
 ) -> dict[str, list[DeclarativeBase]]:
-    """Load fixtures in order."""
+    """Load fixtures in order using batch Core INSERT statements."""
     results: dict[str, list[DeclarativeBase]] = {}
 
     for name in ordered_names:
-        fixture = registry.get(name)
-        instances = list(fixture.func())
+        variants = (
+            registry.get_variants(name, *contexts)
+            if contexts is not None
+            else registry.get_variants(name)
+        )
+
+        if contexts is not None and not variants:
+            variants = registry.get_variants(name)
+
+        if not variants:
+            results[name] = []
+            continue
+
+        instances = [inst for v in variants for inst in v.func()]
 
         if not instances:
             results[name] = []
@@ -36,25 +193,17 @@ async def _load_ordered(
         loaded: list[DeclarativeBase] = []
 
         async with get_transaction(session):
-            for instance in instances:
-                if strategy == LoadStrategy.INSERT:
-                    session.add(instance)
-                    loaded.append(instance)
-
-                elif strategy == LoadStrategy.MERGE:
-                    merged = await session.merge(instance)
-                    loaded.append(merged)
-
-                else:  # LoadStrategy.SKIP_EXISTING
-                    pk = _get_primary_key(instance)
-                    if pk is not None:
-                        existing = await session.get(type(instance), pk)
-                        if existing is None:
-                            session.add(instance)
-                            loaded.append(instance)
-                    else:
-                        session.add(instance)
-                        loaded.append(instance)
+            for model_cls, group in _group_by_type(instances):
+                match strategy:
+                    case LoadStrategy.INSERT:
+                        await _batch_insert(session, model_cls, group)
+                        loaded.extend(group)
+                    case LoadStrategy.MERGE:
+                        await _batch_merge(session, model_cls, group)
+                        loaded.extend(group)
+                    case LoadStrategy.SKIP_EXISTING:
+                        inserted = await _batch_skip_existing(session, model_cls, group)
+                        loaded.extend(inserted)
 
         results[name] = loaded
         logger.info(f"Loaded fixture '{name}': {len(loaded)} {model_name}(s)")
@@ -109,6 +258,8 @@ async def load_fixtures(
 ) -> dict[str, list[DeclarativeBase]]:
     """Load specific fixtures by name with dependencies.
 
+    All context variants of each requested fixture are loaded and merged.
+
     Args:
         session: Database session
         registry: Fixture registry
@@ -125,7 +276,7 @@ async def load_fixtures(
 async def load_fixtures_by_context(
     session: AsyncSession,
     registry: FixtureRegistry,
-    *contexts: str | Context,
+    *contexts: str | Enum,
     strategy: LoadStrategy = LoadStrategy.MERGE,
 ) -> dict[str, list[DeclarativeBase]]:
     """Load all fixtures for specific contexts.
@@ -133,11 +284,15 @@ async def load_fixtures_by_context(
     Args:
         session: Database session
         registry: Fixture registry
-        *contexts: Contexts to load (e.g., Context.BASE, Context.TESTING)
+        *contexts: Contexts to load (e.g., ``Context.BASE``, ``Context.TESTING``,
+            or plain strings for custom contexts)
         strategy: How to handle existing records
 
     Returns:
         Dict mapping fixture names to loaded instances
     """
+    context_strings = tuple(_normalize_contexts(contexts))
     ordered = registry.resolve_context_dependencies(*contexts)
-    return await _load_ordered(session, registry, ordered, strategy)
+    return await _load_ordered(
+        session, registry, ordered, strategy, contexts=context_strings
+    )

src/fastapi_toolsets/metrics/handler.py

@@ -1,6 +1,6 @@
 """Prometheus metrics endpoint for FastAPI applications."""
 
-import asyncio
+import inspect
 import os
 
 from fastapi import FastAPI
@@ -55,10 +55,10 @@ def init_metrics(
 
     # Partition collectors and cache env check at startup — both are stable for the app lifetime.
     async_collectors = [
-        c for c in registry.get_collectors() if asyncio.iscoroutinefunction(c.func)
+        c for c in registry.get_collectors() if inspect.iscoroutinefunction(c.func)
     ]
     sync_collectors = [
-        c for c in registry.get_collectors() if not asyncio.iscoroutinefunction(c.func)
+        c for c in registry.get_collectors() if not inspect.iscoroutinefunction(c.func)
     ]
     multiprocess_mode = _is_multiprocess()
 

src/fastapi_toolsets/models/__init__.py

@@ -7,15 +7,15 @@ from .columns import (
     UUIDv7Mixin,
     UpdatedAtMixin,
 )
-from .watched import ModelEvent, WatchedFieldsMixin, watch
+from .watched import EventSession, ModelEvent, listens_for
 
 __all__ = [
+    "EventSession",
     "ModelEvent",
     "UUIDMixin",
     "UUIDv7Mixin",
     "CreatedAtMixin",
     "UpdatedAtMixin",
     "TimestampMixin",
-    "WatchedFieldsMixin",
-    "watch",
+    "listens_for",
 ]

src/fastapi_toolsets/models/columns.py

@@ -6,14 +6,6 @@ from datetime import datetime
 from sqlalchemy import DateTime, Uuid, text
 from sqlalchemy.orm import Mapped, mapped_column
 
-__all__ = [
-    "UUIDMixin",
-    "UUIDv7Mixin",
-    "CreatedAtMixin",
-    "UpdatedAtMixin",
-    "TimestampMixin",
-]
-
 
 class UUIDMixin:
     """Mixin that adds a UUID primary key auto-generated by the database."""

src/fastapi_toolsets/models/watched.py

@@ -1,11 +1,9 @@
 """Field-change monitoring via SQLAlchemy session events."""
 
-import asyncio
 import inspect
-import weakref
-from collections.abc import Awaitable
+from collections.abc import Callable
 from enum import Enum
-from typing import Any, TypeVar
+from typing import Any
 
 from sqlalchemy import event
 from sqlalchemy import inspect as sa_inspect
@@ -14,65 +12,114 @@ from sqlalchemy.orm.attributes import set_committed_value as _sa_set_committed_v
 
 from ..logger import get_logger
 
-__all__ = ["ModelEvent", "WatchedFieldsMixin", "watch"]
-
 _logger = get_logger()
-_T = TypeVar("_T")
-_CALLBACK_ERROR_MSG = "WatchedFieldsMixin callback raised an unhandled exception"
-_WATCHED_FIELDS: weakref.WeakKeyDictionary[type, list[str]] = (
-    weakref.WeakKeyDictionary()
-)
-_SESSION_PENDING_NEW = "_ft_pending_new"
-_SESSION_CREATES = "_ft_creates"
-_SESSION_DELETES = "_ft_deletes"
-_SESSION_UPDATES = "_ft_updates"
-_SESSION_SAVEPOINT_DEPTH = "_ft_sp_depth"
 
 
 class ModelEvent(str, Enum):
-    """Event types emitted by :class:`WatchedFieldsMixin`."""
+    """Event types dispatched by :class:`EventSession`."""
 
     CREATE = "create"
     DELETE = "delete"
     UPDATE = "update"
 
 
-def watch(*fields: str) -> Any:
-    """Class decorator to filter which fields trigger ``on_update``.
+_CALLBACK_ERROR_MSG = "Event callback raised an unhandled exception"
+_SESSION_CREATES = "_ft_creates"
+_SESSION_DELETES = "_ft_deletes"
+_SESSION_UPDATES = "_ft_updates"
+_DEFERRED_STRATEGY_KEY = (("deferred", True), ("instrument", True))
+_EVENT_HANDLERS: dict[tuple[type, ModelEvent], list[Callable[..., Any]]] = {}
+_WATCHED_MODELS: set[type] = set()
+_WATCHED_CACHE: dict[type, bool] = {}
+_HANDLER_CACHE: dict[tuple[type, ModelEvent], list[Callable[..., Any]]] = {}
+
+
+def _invalidate_caches() -> None:
+    """Clear lookup caches after handler registration."""
+    _WATCHED_CACHE.clear()
+    _HANDLER_CACHE.clear()
+
+
+def listens_for(
+    model_class: type,
+    event_types: list[ModelEvent] | None = None,
+) -> Callable[[Callable[..., Any]], Callable[..., Any]]:
+    """Register a callback for one or more model lifecycle events.
 
     Args:
-        *fields: One or more field names to watch. At least one name is required.
-
-    Raises:
-        ValueError: If called with no field names.
+        model_class: The SQLAlchemy model class to listen on.
+        event_types: List of :class:`ModelEvent` values to listen for.
+            Defaults to all event types.
     """
-    if not fields:
-        raise ValueError("@watch requires at least one field name.")
+    evs = event_types if event_types is not None else list(ModelEvent)
 
-    def decorator(cls: type[_T]) -> type[_T]:
-        _WATCHED_FIELDS[cls] = list(fields)
-        return cls
+    def decorator(fn: Callable[..., Any]) -> Callable[..., Any]:
+        for ev in evs:
+            _EVENT_HANDLERS.setdefault((model_class, ev), []).append(fn)
+        _WATCHED_MODELS.add(model_class)
+        _invalidate_caches()
+        return fn
 
     return decorator
 
 
+def _is_watched(obj: Any) -> bool:
+    """Return True if *obj*'s type (or any ancestor) has registered handlers."""
+    cls = type(obj)
+    try:
+        return _WATCHED_CACHE[cls]
+    except KeyError:
+        result = any(klass in _WATCHED_MODELS for klass in cls.__mro__)
+        _WATCHED_CACHE[cls] = result
+        return result
+
+
+def _get_handlers(cls: type, ev: ModelEvent) -> list[Callable[..., Any]]:
+    """Return registered handlers for *cls* and *ev*, walking the MRO."""
+    key = (cls, ev)
+    try:
+        return _HANDLER_CACHE[key]
+    except KeyError:
+        handlers: list[Callable[..., Any]] = []
+        for klass in cls.__mro__:
+            handlers.extend(_EVENT_HANDLERS.get((klass, ev), []))
+        _HANDLER_CACHE[key] = handlers
+        return handlers
+
+
 def _snapshot_column_attrs(obj: Any) -> dict[str, Any]:
     """Read currently-loaded column values into a plain dict."""
     state = sa_inspect(obj)  # InstanceState
     state_dict = state.dict
-    return {
-        prop.key: state_dict[prop.key]
-        for prop in state.mapper.column_attrs
-        if prop.key in state_dict
-    }
+    snapshot: dict[str, Any] = {}
+    for prop in state.mapper.column_attrs:
+        if prop.key in state_dict:
+            snapshot[prop.key] = state_dict[prop.key]
+        elif (  # pragma: no cover
+            not state.expired
+            and prop.strategy_key != _DEFERRED_STRATEGY_KEY
+            and all(
+                col.nullable
+                and col.server_default is None
+                and col.server_onupdate is None
+                for col in prop.columns
+            )
+        ):
+            snapshot[prop.key] = None
+    return snapshot
 
 
-def _get_watched_fields(cls: type) -> list[str] | None:
-    """Return the watched fields for *cls*, walking the MRO to inherit from parents."""
-    for klass in cls.__mro__:
-        if klass in _WATCHED_FIELDS:
-            return _WATCHED_FIELDS[klass]
-    return None
+def _get_watched_fields(cls: type) -> tuple[str, ...] | None:
+    """Return the watched fields for *cls*."""
+    fields = getattr(cls, "__watched_fields__", None)
+    if fields is not None and (
+        not isinstance(fields, tuple) or not all(isinstance(f, str) for f in fields)
+    ):
+        raise TypeError(
+            f"{cls.__name__}.__watched_fields__ must be a tuple[str, ...], "
+            f"got {type(fields).__name__}"
+        )
+    return fields
 
 
 def _upsert_changes(
@@ -93,50 +140,32 @@ def _upsert_changes(
         pending[key] = (obj, changes)
 
 
-@event.listens_for(AsyncSession.sync_session_class, "after_transaction_create")
-def _after_transaction_create(session: Any, transaction: Any) -> None:
-    if transaction.nested:
-        session.info[_SESSION_SAVEPOINT_DEPTH] = (
-            session.info.get(_SESSION_SAVEPOINT_DEPTH, 0) + 1
-        )
-
-
-@event.listens_for(AsyncSession.sync_session_class, "after_transaction_end")
-def _after_transaction_end(session: Any, transaction: Any) -> None:
-    if transaction.nested:
-        depth = session.info.get(_SESSION_SAVEPOINT_DEPTH, 0)
-        if depth > 0:  # pragma: no branch
-            session.info[_SESSION_SAVEPOINT_DEPTH] = depth - 1
-
-
 @event.listens_for(AsyncSession.sync_session_class, "after_flush")
 def _after_flush(session: Any, flush_context: Any) -> None:
-    # New objects: capture references while session.new is still populated.
-    # Values are read in _after_flush_postexec once RETURNING has been processed.
+    # New objects: capture reference. Attributes will be refreshed after commit.
     for obj in session.new:
-        if isinstance(obj, WatchedFieldsMixin):
-            session.info.setdefault(_SESSION_PENDING_NEW, []).append(obj)
+        if _is_watched(obj):
+            session.info.setdefault(_SESSION_CREATES, []).append(obj)
 
-    # Deleted objects: capture before they leave the identity map.
+    # Deleted objects: snapshot now while attributes are still loaded.
     for obj in session.deleted:
-        if isinstance(obj, WatchedFieldsMixin):
-            session.info.setdefault(_SESSION_DELETES, []).append(obj)
+        if _is_watched(obj):
+            snapshot = _snapshot_column_attrs(obj)
+            session.info.setdefault(_SESSION_DELETES, []).append((obj, snapshot))
 
     # Dirty objects: read old/new from SQLAlchemy attribute history.
     for obj in session.dirty:
-        if not isinstance(obj, WatchedFieldsMixin):
+        if not _is_watched(obj):
             continue
 
-        # None = not in dict = watch all fields; list = specific fields only
         watched = _get_watched_fields(type(obj))
         changes: dict[str, dict[str, Any]] = {}
 
+        inst_attrs = sa_inspect(obj).attrs
         attrs = (
-            # Specific fields
-            ((field, sa_inspect(obj).attrs[field]) for field in watched)
+            ((field, inst_attrs[field]) for field in watched)
             if watched is not None
-            # All mapped fields
-            else ((s.key, s) for s in sa_inspect(obj).attrs)
+            else ((s.key, s) for s in inst_attrs)
         )
         for field, attr_state in attrs:
             history = attr_state.history
@@ -154,116 +183,108 @@ def _after_flush(session: Any, flush_context: Any) -> None:
             )
 
 
-@event.listens_for(AsyncSession.sync_session_class, "after_flush_postexec")
-def _after_flush_postexec(session: Any, flush_context: Any) -> None:
-    # New objects are now persistent and RETURNING values have been applied,
-    # so server defaults (id, created_at, …) are available via getattr.
-    pending_new: list[Any] = session.info.pop(_SESSION_PENDING_NEW, [])
-    if not pending_new:
-        return
-    session.info.setdefault(_SESSION_CREATES, []).extend(pending_new)
-
-
 @event.listens_for(AsyncSession.sync_session_class, "after_rollback")
 def _after_rollback(session: Any) -> None:
-    session.info.pop(_SESSION_PENDING_NEW, None)
+    if session.in_transaction():
+        return
     session.info.pop(_SESSION_CREATES, None)
     session.info.pop(_SESSION_DELETES, None)
     session.info.pop(_SESSION_UPDATES, None)
 
 
-def _task_error_handler(task: asyncio.Task[Any]) -> None:
-    if not task.cancelled() and (exc := task.exception()):
-        _logger.error(_CALLBACK_ERROR_MSG, exc_info=exc)
-
-
-def _schedule_with_snapshot(
-    loop: asyncio.AbstractEventLoop, obj: Any, fn: Any, *args: Any
+async def _invoke_callback(
+    fn: Callable[..., Any],
+    obj: Any,
+    event_type: ModelEvent,
+    changes: dict[str, dict[str, Any]] | None,
 ) -> None:
-    """Snapshot *obj*'s column attrs now (before expire_on_commit wipes them),
-    then schedule a coroutine that restores the snapshot and calls *fn*.
-    """
-    snapshot = _snapshot_column_attrs(obj)
-
-    async def _run(
-        obj: Any = obj,
-        fn: Any = fn,
-        snapshot: dict[str, Any] = snapshot,
-        args: tuple = args,
-    ) -> None:
-        for key, value in snapshot.items():
-            _sa_set_committed_value(obj, key, value)
-        try:
-            result = fn(*args)
-            if inspect.isawaitable(result):
-                await result
-        except Exception as exc:
-            _logger.error(_CALLBACK_ERROR_MSG, exc_info=exc)
-
-    task = loop.create_task(_run())
-    task.add_done_callback(_task_error_handler)
+    """Call *fn* and await the result if it is awaitable."""
+    result = fn(obj, event_type, changes)
+    if inspect.isawaitable(result):
+        await result
 
 
-@event.listens_for(AsyncSession.sync_session_class, "after_commit")
-def _after_commit(session: Any) -> None:
-    if session.info.get(_SESSION_SAVEPOINT_DEPTH, 0) > 0:
-        return
+class EventSession(AsyncSession):
+    """AsyncSession subclass that dispatches lifecycle callbacks after commit."""
 
-    creates: list[Any] = session.info.pop(_SESSION_CREATES, [])
-    deletes: list[Any] = session.info.pop(_SESSION_DELETES, [])
-    field_changes: dict[int, tuple[Any, dict[str, dict[str, Any]]]] = session.info.pop(
-        _SESSION_UPDATES, {}
-    )
+    async def commit(self) -> None:  # noqa: C901
+        await super().commit()
 
-    if creates and deletes:
-        transient_ids = {id(o) for o in creates} & {id(o) for o in deletes}
-        if transient_ids:
-            creates = [o for o in creates if id(o) not in transient_ids]
-            deletes = [o for o in deletes if id(o) not in transient_ids]
+        creates: list[Any] = self.info.pop(_SESSION_CREATES, [])
+        deletes: list[tuple[Any, dict[str, Any]]] = self.info.pop(_SESSION_DELETES, [])
+        field_changes: dict[int, tuple[Any, dict[str, dict[str, Any]]]] = self.info.pop(
+            _SESSION_UPDATES, {}
+        )
+
+        if not creates and not deletes and not field_changes:
+            return
+
+        # Suppress transient objects (created + deleted in same transaction).
+        if creates and deletes:
+            created_ids = {id(o) for o in creates}
+            deleted_ids = {id(o) for o, _ in deletes}
+            transient_ids = created_ids & deleted_ids
+            if transient_ids:
+                creates = [o for o in creates if id(o) not in transient_ids]
+                deletes = [(o, s) for o, s in deletes if id(o) not in transient_ids]
+                field_changes = {
+                    k: v for k, v in field_changes.items() if k not in transient_ids
+                }
+
+        # Suppress updates for deleted objects (row is gone, refresh would fail).
+        if deletes and field_changes:
+            deleted_ids = {id(o) for o, _ in deletes}
             field_changes = {
-                k: v for k, v in field_changes.items() if k not in transient_ids
+                k: v for k, v in field_changes.items() if k not in deleted_ids
             }
 
-    if not creates and not deletes and not field_changes:
-        return
+        # Suppress updates for newly created objects (CREATE-only semantics).
+        if creates and field_changes:
+            create_ids = {id(o) for o in creates}
+            field_changes = {
+                k: v for k, v in field_changes.items() if k not in create_ids
+            }
 
-    try:
-        loop = asyncio.get_running_loop()
-    except RuntimeError:
-        return
+        # Dispatch CREATE callbacks.
+        for obj in creates:
+            try:
+                state = sa_inspect(obj, raiseerr=False)
+                if (
+                    state is None or state.detached or state.transient
+                ):  # pragma: no cover
+                    continue
+                await self.refresh(obj)
+                for handler in _get_handlers(type(obj), ModelEvent.CREATE):
+                    await _invoke_callback(handler, obj, ModelEvent.CREATE, None)
+            except Exception as exc:
+                _logger.error(_CALLBACK_ERROR_MSG, exc_info=exc)
 
-    for obj in creates:
-        _schedule_with_snapshot(loop, obj, obj.on_create)
+        # Dispatch DELETE callbacks (restore snapshot; row is gone).
+        for obj, snapshot in deletes:
+            try:
+                for key, value in snapshot.items():
+                    _sa_set_committed_value(obj, key, value)
+                for handler in _get_handlers(type(obj), ModelEvent.DELETE):
+                    await _invoke_callback(handler, obj, ModelEvent.DELETE, None)
+            except Exception as exc:
+                _logger.error(_CALLBACK_ERROR_MSG, exc_info=exc)
 
-    for obj in deletes:
-        _schedule_with_snapshot(loop, obj, obj.on_delete)
+        # Dispatch UPDATE callbacks.
+        for obj, changes in field_changes.values():
+            try:
+                state = sa_inspect(obj, raiseerr=False)
+                if (
+                    state is None or state.detached or state.transient
+                ):  # pragma: no cover
+                    continue
+                await self.refresh(obj)
+                for handler in _get_handlers(type(obj), ModelEvent.UPDATE):
+                    await _invoke_callback(handler, obj, ModelEvent.UPDATE, changes)
+            except Exception as exc:
+                _logger.error(_CALLBACK_ERROR_MSG, exc_info=exc)
 
-    for obj, changes in field_changes.values():
-        _schedule_with_snapshot(loop, obj, obj.on_update, changes)
-
-
-class WatchedFieldsMixin:
-    """Mixin that enables lifecycle callbacks for SQLAlchemy models."""
-
-    def on_event(
-        self, event: ModelEvent, changes: dict[str, dict[str, Any]] | None = None
-    ) -> Awaitable[None] | None:
-        """Catch-all callback fired for every lifecycle event.
-
-        Args:
-            event: The event type (:attr:`ModelEvent.CREATE`, :attr:`ModelEvent.DELETE`,
-                or :attr:`ModelEvent.UPDATE`).
-            changes: Field changes for :attr:`ModelEvent.UPDATE`, ``None`` otherwise.
-        """
-
-    def on_create(self) -> Awaitable[None] | None:
-        """Called after INSERT commit."""
-        return self.on_event(ModelEvent.CREATE)
-
-    def on_delete(self) -> Awaitable[None] | None:
-        """Called after DELETE commit."""
-        return self.on_event(ModelEvent.DELETE)
-
-    def on_update(self, changes: dict[str, dict[str, Any]]) -> Awaitable[None] | None:
-        """Called after UPDATE commit when watched fields change."""
-        return self.on_event(ModelEvent.UPDATE, changes=changes)
+    async def rollback(self) -> None:
+        await super().rollback()
+        self.info.pop(_SESSION_CREATES, None)
+        self.info.pop(_SESSION_DELETES, None)
+        self.info.pop(_SESSION_UPDATES, None)

src/fastapi_toolsets/pytest/utils.py

@@ -1,12 +1,12 @@
 """Pytest helper utilities for FastAPI testing."""
 
 import os
-import warnings
 from collections.abc import AsyncGenerator, Callable
 from contextlib import asynccontextmanager
 from typing import Any
 
 from httpx import ASGITransport, AsyncClient
+from sqlalchemy import text
 from sqlalchemy.engine import make_url
 from sqlalchemy.ext.asyncio import (
     AsyncSession,
@@ -15,33 +15,8 @@ from sqlalchemy.ext.asyncio import (
 )
 from sqlalchemy.orm import DeclarativeBase
 
-from sqlalchemy import text
-
-from ..db import (
-    cleanup_tables as _cleanup_tables,
-    create_database,
-    create_db_context,
-)
-
-
-async def cleanup_tables(
-    session: AsyncSession,
-    base: type[DeclarativeBase],
-) -> None:
-    """Truncate all tables for fast between-test cleanup.
-
-    .. deprecated::
-        Import ``cleanup_tables`` from ``fastapi_toolsets.db`` instead.
-        This re-export will be removed in v3.0.0.
-    """
-    warnings.warn(
-        "Importing cleanup_tables from fastapi_toolsets.pytest is deprecated "
-        "and will be removed in v3.0.0. "
-        "Use 'from fastapi_toolsets.db import cleanup_tables' instead.",
-        DeprecationWarning,
-        stacklevel=2,
-    )
-    await _cleanup_tables(session=session, base=base)
+from ..db import cleanup_tables, create_database
+from ..models.watched import EventSession
 
 
 def _get_xdist_worker(default_test_db: str) -> str:
@@ -269,15 +244,14 @@ async def create_db_session(
         async with engine.begin() as conn:
             await conn.run_sync(base.metadata.create_all)
 
-        # Create session using existing db context utility
-        session_maker = async_sessionmaker(engine, expire_on_commit=expire_on_commit)
-        get_session = create_db_context(session_maker)
-
-        async with get_session() as session:
+        session_maker = async_sessionmaker(
+            engine, expire_on_commit=expire_on_commit, class_=EventSession
+        )
+        async with session_maker() as session:
             yield session
 
             if cleanup:
-                await cleanup_tables(session, base)
+                await cleanup_tables(session=session, base=base)
 
         if drop_tables:
             async with engine.begin() as conn:

src/fastapi_toolsets/schemas.py

@@ -162,6 +162,7 @@ class PaginatedResponse(BaseResponse, Generic[DataT]):
     pagination: OffsetPagination | CursorPagination
     pagination_type: PaginationType | None = None
     filter_attributes: dict[str, list[Any]] | None = None
+    search_columns: list[str] | None = None
 
     _discriminated_union_cache: ClassVar[dict[Any, Any]] = {}
 

tests/conftest.py

@@ -14,11 +14,13 @@ from sqlalchemy import (
     DateTime,
     ForeignKey,
     Integer,
+    JSON,
     Numeric,
     String,
     Table,
     Uuid,
 )
+from sqlalchemy.dialects.postgresql import ARRAY
 from sqlalchemy.ext.asyncio import async_sessionmaker, create_async_engine
 from sqlalchemy.orm import DeclarativeBase, Mapped, mapped_column, relationship
 
@@ -57,6 +59,7 @@ class User(Base):
     username: Mapped[str] = mapped_column(String(50), unique=True)
     email: Mapped[str] = mapped_column(String(100), unique=True)
     is_active: Mapped[bool] = mapped_column(default=True)
+    notes: Mapped[str | None]
     role_id: Mapped[uuid.UUID | None] = mapped_column(
         ForeignKey("roles.id"), nullable=True
     )
@@ -136,6 +139,17 @@ class Post(Base):
     tags: Mapped[list[Tag]] = relationship(secondary=post_tags)
 
 
+class Article(Base):
+    """Test article model with ARRAY and JSON columns."""
+
+    __tablename__ = "articles"
+
+    id: Mapped[uuid.UUID] = mapped_column(Uuid, primary_key=True, default=uuid.uuid4)
+    title: Mapped[str] = mapped_column(String(200))
+    labels: Mapped[list[str]] = mapped_column(ARRAY(String))
+    metadata_: Mapped[dict | None] = mapped_column("metadata", JSON, nullable=True)
+
+
 class RoleCreate(BaseModel):
     """Schema for creating a role."""
 
@@ -270,6 +284,23 @@ class ProductCreate(BaseModel):
     price: decimal.Decimal
 
 
+class ArticleCreate(BaseModel):
+    """Schema for creating an article."""
+
+    id: uuid.UUID | None = None
+    title: str
+    labels: list[str] = []
+
+
+class ArticleRead(PydanticBase):
+    """Schema for reading an article."""
+
+    id: uuid.UUID
+    title: str
+    labels: list[str]
+
+
+ArticleCrud = CrudFactory(Article)
 RoleCrud = CrudFactory(Role)
 RoleCursorCrud = CrudFactory(Role, cursor_column=Role.id)
 IntRoleCursorCrud = CrudFactory(IntRole, cursor_column=IntRole.id)

tests/test_crud.py

@@ -211,6 +211,38 @@ class TestResolveLoadOptions:
         assert crud._resolve_load_options([]) == []
 
 
+class TestResolveSearchColumns:
+    """Tests for _resolve_search_columns logic."""
+
+    def test_returns_none_when_no_searchable_fields(self):
+        """Returns None when cls.searchable_fields is None and no search_fields passed."""
+
+        class AbstractCrud(AsyncCrud[User]):
+            pass
+
+        assert AbstractCrud._resolve_search_columns(None) is None
+
+    def test_returns_none_when_empty_search_fields_passed(self):
+        """Returns None when an empty list is passed explicitly."""
+        crud = CrudFactory(User)
+        assert crud._resolve_search_columns([]) is None
+
+    def test_returns_keys_from_class_searchable_fields(self):
+        """Returns column keys from cls.searchable_fields when no override passed."""
+        crud = CrudFactory(User, searchable_fields=[User.username])
+        result = crud._resolve_search_columns(None)
+        assert result is not None
+        assert "username" in result
+
+    def test_search_fields_override_takes_priority(self):
+        """Explicit search_fields override cls.searchable_fields."""
+        crud = CrudFactory(User, searchable_fields=[User.username])
+        result = crud._resolve_search_columns([User.email])
+        assert result is not None
+        assert "email" in result
+        assert "username" not in result
+
+
 class TestDefaultLoadOptionsIntegration:
     """Integration tests for default_load_options with real DB queries."""
 

tests/test_example_pagination_search.py

@@ -97,7 +97,7 @@ class TestAppSessionDep:
         gen = get_db()
         session = await gen.__anext__()
         assert isinstance(session, AsyncSession)
-        await session.close()
+        await gen.aclose()
 
 
 class TestOffsetPagination:
@@ -182,8 +182,7 @@ class TestOffsetPagination:
         body = resp.json()
         fa = body["filter_attributes"]
         assert set(fa["status"]) == {"draft", "published"}
-        # "name" is unique across all facet fields — no prefix needed
-        assert set(fa["name"]) == {"backend", "python"}
+        assert set(fa["category__name"]) == {"backend", "python"}
 
     @pytest.mark.anyio
     async def test_filter_attributes_scoped_to_filter(

tests/test_fixtures.py

@@ -1,6 +1,7 @@
 """Tests for fastapi_toolsets.fixtures module."""
 
 import uuid
+from enum import Enum
 
 import pytest
 from sqlalchemy.ext.asyncio import AsyncSession
@@ -13,10 +14,22 @@ from fastapi_toolsets.fixtures import (
     load_fixtures,
     load_fixtures_by_context,
 )
+from fastapi_toolsets.fixtures.utils import _get_primary_key, _instance_to_dict
 
-from fastapi_toolsets.fixtures.utils import _get_primary_key
+from .conftest import IntRole, Permission, Role, RoleCreate, RoleCrud, User, UserCrud
 
-from .conftest import IntRole, Permission, Role, RoleCrud, User, UserCrud
+
+class AppContext(str, Enum):
+    """Example user-defined str+Enum context."""
+
+    STAGING = "staging"
+    DEMO = "demo"
+
+
+class PlainEnumContext(Enum):
+    """Example user-defined plain Enum context (no str mixin)."""
+
+    STAGING = "staging"
 
 
 class TestContext:
@@ -39,6 +52,86 @@ class TestContext:
         assert Context.TESTING.value == "testing"
 
 
+class TestCustomEnumContext:
+    """Custom Enum types are accepted wherever Context/str are expected."""
+
+    def test_cannot_subclass_context_with_members(self):
+        """Python prohibits extending an Enum that already has members."""
+        with pytest.raises(TypeError):
+
+            class MyContext(Context):  # noqa: F841  # ty: ignore[subclass-of-final-class]
+                STAGING = "staging"
+
+    def test_custom_enum_values_interchangeable_with_context(self):
+        """A custom enum with the same .value as a built-in Context member is
+        treated as the same context — fixtures registered under one are found
+        by the other."""
+
+        class AppContextFull(str, Enum):
+            BASE = "base"
+            STAGING = "staging"
+
+        registry = FixtureRegistry()
+
+        @registry.register(contexts=[Context.BASE])
+        def roles():
+            return []
+
+        # AppContextFull.BASE has value "base" — same as Context.BASE
+        fixtures = registry.get_by_context(AppContextFull.BASE)
+        assert len(fixtures) == 1
+
+    def test_custom_enum_registry_default_contexts(self):
+        """FixtureRegistry(contexts=[...]) accepts a custom Enum."""
+        registry = FixtureRegistry(contexts=[AppContext.STAGING])
+
+        @registry.register
+        def data():
+            return []
+
+        fixture = registry.get("data")
+        assert fixture.contexts == ["staging"]
+
+    def test_custom_enum_resolve_context_dependencies(self):
+        """resolve_context_dependencies accepts a custom Enum context."""
+        registry = FixtureRegistry()
+
+        @registry.register(contexts=[AppContext.STAGING])
+        def staging_roles():
+            return []
+
+        order = registry.resolve_context_dependencies(AppContext.STAGING)
+        assert "staging_roles" in order
+
+    @pytest.mark.anyio
+    async def test_custom_enum_e2e(self, db_session: AsyncSession):
+        """End-to-end: register with custom Enum, load with the same Enum."""
+        registry = FixtureRegistry()
+
+        @registry.register(contexts=[AppContext.STAGING])
+        def staging_roles():
+            return [Role(id=uuid.uuid4(), name="staging-admin")]
+
+        result = await load_fixtures_by_context(
+            db_session, registry, AppContext.STAGING
+        )
+        assert len(result["staging_roles"]) == 1
+
+    @pytest.mark.anyio
+    async def test_plain_enum_e2e(self, db_session: AsyncSession):
+        """End-to-end: register with plain Enum, load with the same Enum."""
+        registry = FixtureRegistry()
+
+        @registry.register(contexts=[PlainEnumContext.STAGING])
+        def staging_roles():
+            return [Role(id=uuid.uuid4(), name="plain-staging-admin")]
+
+        result = await load_fixtures_by_context(
+            db_session, registry, PlainEnumContext.STAGING
+        )
+        assert len(result["staging_roles"]) == 1
+
+
 class TestLoadStrategy:
     """Tests for LoadStrategy enum."""
 
@@ -407,6 +500,37 @@ class TestDependencyResolution:
         with pytest.raises(ValueError, match="Circular dependency"):
             registry.resolve_dependencies("a")
 
+    def test_resolve_raises_for_unknown_dependency(self):
+        """KeyError when depends_on references an unregistered fixture."""
+        registry = FixtureRegistry()
+
+        @registry.register(depends_on=["ghost"])
+        def users():
+            return []
+
+        with pytest.raises(KeyError, match="ghost"):
+            registry.resolve_dependencies("users")
+
+    def test_resolve_deduplicates_shared_depends_on_across_variants(self):
+        """A dep shared by two same-name variants appears only once in the order."""
+        registry = FixtureRegistry()
+
+        @registry.register(contexts=[Context.BASE])
+        def roles():
+            return []
+
+        @registry.register(depends_on=["roles"], contexts=[Context.BASE])
+        def items():
+            return []
+
+        @registry.register(depends_on=["roles"], contexts=[Context.TESTING])
+        def items():  # noqa: F811
+            return []
+
+        order = registry.resolve_dependencies("items")
+        assert order.count("roles") == 1
+        assert order.index("roles") < order.index("items")
+
     def test_resolve_context_dependencies(self):
         """Resolve all fixtures for a context with dependencies."""
         registry = FixtureRegistry()
@@ -496,6 +620,52 @@ class TestLoadFixtures:
         count = await RoleCrud.count(db_session)
         assert count == 1
 
+    @pytest.mark.anyio
+    async def test_merge_does_not_overwrite_omitted_nullable_columns(
+        self, db_session: AsyncSession
+    ):
+        """MERGE must not clear nullable columns that the fixture didn't set.
+
+        When a fixture omits a nullable column (e.g. role_id or notes), a re-merge
+        must leave the existing DB value untouched — not overwrite it with NULL.
+        """
+        registry = FixtureRegistry()
+        admin = await RoleCrud.create(db_session, RoleCreate(name="admin"))
+        uid = uuid.uuid4()
+
+        # First load: user has role_id and notes set
+        @registry.register
+        def users():
+            return [
+                User(
+                    id=uid,
+                    username="alice",
+                    email="a@test.com",
+                    role_id=admin.id,
+                    notes="original",
+                )
+            ]
+
+        await load_fixtures(db_session, registry, "users", strategy=LoadStrategy.MERGE)
+
+        # Second load: fixture omits role_id and notes
+        registry2 = FixtureRegistry()
+
+        @registry2.register
+        def users():  # noqa: F811
+            return [User(id=uid, username="alice-updated", email="a@test.com")]
+
+        await load_fixtures(db_session, registry2, "users", strategy=LoadStrategy.MERGE)
+
+        from sqlalchemy import select
+
+        row = (
+            await db_session.execute(select(User).where(User.id == uid))
+        ).scalar_one()
+        assert row.username == "alice-updated"  # updated column changed
+        assert row.role_id == admin.id  # omitted → preserved
+        assert row.notes == "original"  # omitted → preserved
+
     @pytest.mark.anyio
     async def test_load_with_skip_existing_strategy(self, db_session: AsyncSession):
         """Load fixtures with SKIP_EXISTING strategy."""
@@ -795,3 +965,511 @@ class TestGetPrimaryKey:
         instance = Permission(subject="post")  # action is None
         pk = _get_primary_key(instance)
         assert pk is None
+
+
+class TestRegistryGetVariants:
+    """Tests for FixtureRegistry.get and get_variants edge cases."""
+
+    def test_get_raises_value_error_for_multi_variant(self):
+        """get() raises ValueError when the fixture has multiple context variants."""
+        registry = FixtureRegistry()
+
+        @registry.register(contexts=[Context.BASE])
+        def items():
+            return []
+
+        @registry.register(contexts=[Context.TESTING])
+        def items():  # noqa: F811
+            return []
+
+        with pytest.raises(ValueError, match="get_variants"):
+            registry.get("items")
+
+    def test_get_variants_raises_key_error_for_unknown(self):
+        """get_variants() raises KeyError for an unregistered name."""
+        registry = FixtureRegistry()
+        with pytest.raises(KeyError, match="not found"):
+            registry.get_variants("no_such_fixture")
+
+
+class TestInstanceToDict:
+    """Unit tests for the _instance_to_dict helper."""
+
+    def test_explicit_values_included(self):
+        """All explicitly set column values appear in the result."""
+        role_id = uuid.uuid4()
+        instance = Role(id=role_id, name="admin")
+        d = _instance_to_dict(instance)
+        assert d["id"] == role_id
+        assert d["name"] == "admin"
+
+    def test_callable_default_none_excluded(self):
+        """A column whose value is None but has a callable Python-side default
+        (e.g. ``default=uuid.uuid4``) is excluded so the DB generates it."""
+        instance = Role(id=None, name="admin")
+        d = _instance_to_dict(instance)
+        assert "id" not in d
+        assert d["name"] == "admin"
+
+    def test_autoincrement_none_excluded(self):
+        """A column whose value is None but has autoincrement=True is excluded
+        so the DB generates the value via its sequence."""
+        instance = IntRole(id=None, name="admin")
+        d = _instance_to_dict(instance)
+        assert "id" not in d
+        assert d["name"] == "admin"
+
+    def test_nullable_none_included(self):
+        """None on a nullable column with no default is kept (explicit NULL)."""
+        instance = User(id=uuid.uuid4(), username="u", email="e@e.com", role_id=None)
+        d = _instance_to_dict(instance)
+        assert "role_id" in d
+        assert d["role_id"] is None
+
+    def test_nullable_str_no_default_omitted_not_in_dict(self):
+        """Mapped[str | None] with no default, not provided in constructor, is absent from dict."""
+        instance = User(id=uuid.uuid4(), username="u", email="e@e.com")
+        d = _instance_to_dict(instance)
+        assert "notes" not in d
+
+    def test_nullable_str_no_default_explicit_none_included(self):
+        """Mapped[str | None] with no default, explicitly set to None, is included as NULL."""
+        instance = User(id=uuid.uuid4(), username="u", email="e@e.com", notes=None)
+        d = _instance_to_dict(instance)
+        assert "notes" in d
+        assert d["notes"] is None
+
+    def test_nullable_str_no_default_with_value_included(self):
+        """Mapped[str | None] with no default and a value set is included normally."""
+        instance = User(id=uuid.uuid4(), username="u", email="e@e.com", notes="hello")
+        d = _instance_to_dict(instance)
+        assert d["notes"] == "hello"
+
+    @pytest.mark.anyio
+    async def test_nullable_str_no_default_insert_roundtrip(
+        self, db_session: AsyncSession
+    ):
+        """Fixture loading works for models with Mapped[str | None] (no default).
+
+        Both the omitted-value (→ NULL) and explicit-None paths must insert without error.
+        """
+        registry = FixtureRegistry()
+
+        uid_a = uuid.uuid4()
+        uid_b = uuid.uuid4()
+        uid_c = uuid.uuid4()
+
+        @registry.register
+        def users():
+            return [
+                User(
+                    id=uid_a, username="no_notes", email="a@test.com"
+                ),  # notes omitted
+                User(
+                    id=uid_b, username="null_notes", email="b@test.com", notes=None
+                ),  # explicit None
+                User(
+                    id=uid_c, username="has_notes", email="c@test.com", notes="hi"
+                ),  # value set
+            ]
+
+        result = await load_fixtures(db_session, registry, "users")
+
+        from sqlalchemy import select
+
+        rows = (
+            (await db_session.execute(select(User).order_by(User.username)))
+            .scalars()
+            .all()
+        )
+        by_username = {r.username: r for r in rows}
+
+        assert by_username["no_notes"].notes is None
+        assert by_username["null_notes"].notes is None
+        assert by_username["has_notes"].notes == "hi"
+        assert len(result["users"]) == 3
+
+
+class TestBatchMergeNonPkColumns:
+    """Batch MERGE on a model with no non-PK columns (PK-only table)."""
+
+    @pytest.mark.anyio
+    async def test_merge_pk_only_model(self, db_session: AsyncSession):
+        """MERGE strategy on a PK-only model uses on_conflict_do_nothing."""
+        registry = FixtureRegistry()
+
+        @registry.register
+        def permissions():
+            return [
+                Permission(subject="post", action="read"),
+                Permission(subject="post", action="write"),
+            ]
+
+        result = await load_fixtures(
+            db_session, registry, "permissions", strategy=LoadStrategy.MERGE
+        )
+        assert len(result["permissions"]) == 2
+
+        # Run again — conflicts are silently ignored.
+        result2 = await load_fixtures(
+            db_session, registry, "permissions", strategy=LoadStrategy.MERGE
+        )
+        assert len(result2["permissions"]) == 2
+
+
+class TestBatchNullableColumnEdgeCases:
+    """Deep tests for nullable column handling during batch import."""
+
+    @pytest.mark.anyio
+    async def test_insert_batch_mixed_nullable_fk(self, db_session: AsyncSession):
+        """INSERT batch where some rows set a nullable FK and others don't.
+
+        After normalization the omitted role_id becomes None.  For INSERT this
+        is acceptable — both rows should insert successfully with the correct
+        values (one with FK, one with NULL).
+        """
+        registry = FixtureRegistry()
+        admin = await RoleCrud.create(db_session, RoleCreate(name="admin"))
+        uid1 = uuid.uuid4()
+        uid2 = uuid.uuid4()
+
+        @registry.register
+        def users():
+            return [
+                User(
+                    id=uid1, username="with_role", email="a@test.com", role_id=admin.id
+                ),
+                User(id=uid2, username="no_role", email="b@test.com"),
+            ]
+
+        await load_fixtures(db_session, registry, "users", strategy=LoadStrategy.INSERT)
+
+        from sqlalchemy import select
+
+        rows = {
+            r.username: r
+            for r in (await db_session.execute(select(User))).scalars().all()
+        }
+        assert rows["with_role"].role_id == admin.id
+        assert rows["no_role"].role_id is None
+
+    @pytest.mark.anyio
+    async def test_insert_batch_mixed_nullable_notes(self, db_session: AsyncSession):
+        """INSERT batch where some rows have notes and others don't.
+
+        Ensures normalization doesn't break the insert and that each row gets
+        the intended value.
+        """
+        registry = FixtureRegistry()
+        uid1 = uuid.uuid4()
+        uid2 = uuid.uuid4()
+        uid3 = uuid.uuid4()
+
+        @registry.register
+        def users():
+            return [
+                User(
+                    id=uid1,
+                    username="has_notes",
+                    email="a@test.com",
+                    notes="important",
+                ),
+                User(id=uid2, username="no_notes", email="b@test.com"),
+                User(id=uid3, username="null_notes", email="c@test.com", notes=None),
+            ]
+
+        await load_fixtures(db_session, registry, "users", strategy=LoadStrategy.INSERT)
+
+        from sqlalchemy import select
+
+        rows = {
+            r.username: r
+            for r in (await db_session.execute(select(User))).scalars().all()
+        }
+        assert rows["has_notes"].notes == "important"
+        assert rows["no_notes"].notes is None
+        assert rows["null_notes"].notes is None
+
+    @pytest.mark.anyio
+    async def test_merge_batch_mixed_nullable_does_not_overwrite(
+        self, db_session: AsyncSession
+    ):
+        """MERGE batch where one row sets a nullable column and another omits it.
+
+        If both rows already exist in DB, the row that omits the column must
+        NOT have its existing value overwritten with NULL.
+
+        This is the core normalization bug: _normalize_rows fills missing keys
+        with None, and then MERGE's SET clause includes that column for ALL rows.
+        """
+        from sqlalchemy import select
+
+        admin = await RoleCrud.create(db_session, RoleCreate(name="admin"))
+        uid1 = uuid.uuid4()
+        uid2 = uuid.uuid4()
+
+        # Pre-populate: both users have role_id and notes
+        registry_initial = FixtureRegistry()
+
+        @registry_initial.register
+        def users():
+            return [
+                User(
+                    id=uid1,
+                    username="alice",
+                    email="a@test.com",
+                    role_id=admin.id,
+                    notes="alice notes",
+                ),
+                User(
+                    id=uid2,
+                    username="bob",
+                    email="b@test.com",
+                    role_id=admin.id,
+                    notes="bob notes",
+                ),
+            ]
+
+        await load_fixtures(
+            db_session, registry_initial, "users", strategy=LoadStrategy.INSERT
+        )
+
+        # Re-merge: alice updates notes, bob omits notes entirely
+        registry_merge = FixtureRegistry()
+
+        @registry_merge.register
+        def users():  # noqa: F811
+            return [
+                User(
+                    id=uid1,
+                    username="alice",
+                    email="a@test.com",
+                    role_id=admin.id,
+                    notes="updated",
+                ),
+                User(
+                    id=uid2,
+                    username="bob",
+                    email="b@test.com",
+                    role_id=admin.id,
+                ),  # notes omitted
+            ]
+
+        await load_fixtures(
+            db_session, registry_merge, "users", strategy=LoadStrategy.MERGE
+        )
+
+        rows = {
+            r.username: r
+            for r in (await db_session.execute(select(User))).scalars().all()
+        }
+        assert rows["alice"].notes == "updated"
+        # Bob's notes must be preserved, NOT overwritten with NULL
+        assert rows["bob"].notes == "bob notes"
+
+    @pytest.mark.anyio
+    async def test_merge_batch_mixed_nullable_fk_preserves_existing(
+        self, db_session: AsyncSession
+    ):
+        """MERGE batch where one row sets role_id and another omits it.
+
+        The row that omits role_id must keep its existing DB value.
+        """
+        from sqlalchemy import select
+
+        admin = await RoleCrud.create(db_session, RoleCreate(name="admin"))
+        editor = await RoleCrud.create(db_session, RoleCreate(name="editor"))
+        uid1 = uuid.uuid4()
+        uid2 = uuid.uuid4()
+
+        # Pre-populate
+        registry_initial = FixtureRegistry()
+
+        @registry_initial.register
+        def users():
+            return [
+                User(
+                    id=uid1,
+                    username="alice",
+                    email="a@test.com",
+                    role_id=admin.id,
+                ),
+                User(
+                    id=uid2,
+                    username="bob",
+                    email="b@test.com",
+                    role_id=editor.id,
+                ),
+            ]
+
+        await load_fixtures(
+            db_session, registry_initial, "users", strategy=LoadStrategy.INSERT
+        )
+
+        # Re-merge: alice changes role, bob omits role_id
+        registry_merge = FixtureRegistry()
+
+        @registry_merge.register
+        def users():  # noqa: F811
+            return [
+                User(
+                    id=uid1,
+                    username="alice",
+                    email="a@test.com",
+                    role_id=editor.id,
+                ),
+                User(id=uid2, username="bob", email="b@test.com"),  # role_id omitted
+            ]
+
+        await load_fixtures(
+            db_session, registry_merge, "users", strategy=LoadStrategy.MERGE
+        )
+
+        rows = {
+            r.username: r
+            for r in (await db_session.execute(select(User))).scalars().all()
+        }
+        assert rows["alice"].role_id == editor.id  # updated
+        assert rows["bob"].role_id == editor.id  # must be preserved, NOT NULL
+
+    @pytest.mark.anyio
+    async def test_insert_batch_mixed_pk_presence(self, db_session: AsyncSession):
+        """INSERT batch where some rows have explicit PK and others rely on
+        the callable default (uuid.uuid4).
+
+        Normalization adds the PK key with None to rows that omitted it,
+        which can cause NOT NULL violations on the PK column.
+        """
+        registry = FixtureRegistry()
+        explicit_id = uuid.uuid4()
+
+        @registry.register
+        def roles():
+            return [
+                Role(id=explicit_id, name="admin"),
+                Role(name="user"),  # PK omitted, relies on default=uuid.uuid4
+            ]
+
+        await load_fixtures(db_session, registry, "roles", strategy=LoadStrategy.INSERT)
+
+        from sqlalchemy import select
+
+        rows = (await db_session.execute(select(Role))).scalars().all()
+        assert len(rows) == 2
+        names = {r.name for r in rows}
+        assert names == {"admin", "user"}
+        # The "admin" row must have the explicit ID
+        admin = next(r for r in rows if r.name == "admin")
+        assert admin.id == explicit_id
+        # The "user" row must have a generated UUID (not None)
+        user = next(r for r in rows if r.name == "user")
+        assert user.id is not None
+
+    @pytest.mark.anyio
+    async def test_skip_existing_batch_mixed_nullable(self, db_session: AsyncSession):
+        """SKIP_EXISTING with mixed nullable columns inserts correctly.
+
+        Only new rows are inserted; existing rows are untouched regardless of
+        which columns the fixture provides.
+        """
+        from sqlalchemy import select
+
+        admin = await RoleCrud.create(db_session, RoleCreate(name="admin"))
+        uid1 = uuid.uuid4()
+        uid2 = uuid.uuid4()
+
+        # Pre-populate uid1 with notes
+        registry_initial = FixtureRegistry()
+
+        @registry_initial.register
+        def users():
+            return [
+                User(
+                    id=uid1,
+                    username="alice",
+                    email="a@test.com",
+                    role_id=admin.id,
+                    notes="keep me",
+                ),
+            ]
+
+        await load_fixtures(
+            db_session, registry_initial, "users", strategy=LoadStrategy.INSERT
+        )
+
+        # Load again with SKIP_EXISTING: uid1 already exists, uid2 is new
+        registry_skip = FixtureRegistry()
+
+        @registry_skip.register
+        def users():  # noqa: F811
+            return [
+                User(id=uid1, username="alice-updated", email="a@test.com"),  # exists
+                User(
+                    id=uid2,
+                    username="bob",
+                    email="b@test.com",
+                    notes="new user",
+                ),  # new
+            ]
+
+        result = await load_fixtures(
+            db_session, registry_skip, "users", strategy=LoadStrategy.SKIP_EXISTING
+        )
+        assert len(result["users"]) == 1  # only bob inserted
+
+        rows = {
+            r.username: r
+            for r in (await db_session.execute(select(User))).scalars().all()
+        }
+        # alice untouched
+        assert rows["alice"].role_id == admin.id
+        assert rows["alice"].notes == "keep me"
+        # bob inserted correctly
+        assert rows["bob"].notes == "new user"
+
+    @pytest.mark.anyio
+    async def test_insert_batch_every_row_different_nullable_columns(
+        self, db_session: AsyncSession
+    ):
+        """Each row in the batch sets a different combination of nullable columns.
+
+        Tests that normalization produces valid SQL for all rows.
+        """
+        registry = FixtureRegistry()
+        admin = await RoleCrud.create(db_session, RoleCreate(name="admin"))
+        uid1 = uuid.uuid4()
+        uid2 = uuid.uuid4()
+        uid3 = uuid.uuid4()
+
+        @registry.register
+        def users():
+            return [
+                User(
+                    id=uid1,
+                    username="all_set",
+                    email="a@test.com",
+                    role_id=admin.id,
+                    notes="full",
+                ),
+                User(
+                    id=uid2, username="only_role", email="b@test.com", role_id=admin.id
+                ),
+                User(
+                    id=uid3, username="only_notes", email="c@test.com", notes="partial"
+                ),
+            ]
+
+        await load_fixtures(db_session, registry, "users", strategy=LoadStrategy.INSERT)
+
+        from sqlalchemy import select
+
+        rows = {
+            r.username: r
+            for r in (await db_session.execute(select(User))).scalars().all()
+        }
+        assert rows["all_set"].role_id == admin.id
+        assert rows["all_set"].notes == "full"
+        assert rows["only_role"].role_id == admin.id
+        assert rows["only_role"].notes is None
+        assert rows["only_notes"].role_id is None
+        assert rows["only_notes"].notes == "partial"

tests/test_pytest.py

@@ -7,9 +7,10 @@ from fastapi import Depends, FastAPI
 from httpx import AsyncClient
 from sqlalchemy import select, text
 from sqlalchemy.engine import make_url
-from sqlalchemy.ext.asyncio import AsyncSession, create_async_engine
+from sqlalchemy.ext.asyncio import AsyncSession, async_sessionmaker, create_async_engine
 from sqlalchemy.orm import selectinload
 
+from fastapi_toolsets.db import get_transaction
 from fastapi_toolsets.fixtures import Context, FixtureRegistry
 from fastapi_toolsets.pytest import (
     create_async_client,
@@ -336,6 +337,42 @@ class TestCreateDbSession:
             result = await session.execute(select(Role))
             assert result.all() == []
 
+    @pytest.mark.anyio
+    async def test_get_transaction_commits_visible_to_separate_session(self):
+        """Data written via get_transaction() is committed and visible to other sessions."""
+        role_id = uuid.uuid4()
+
+        async with create_db_session(DATABASE_URL, Base, drop_tables=False) as session:
+            # Simulate what _create_fixture_function does: insert via get_transaction
+            # with no explicit commit afterward.
+            async with get_transaction(session):
+                role = Role(id=role_id, name="visible_to_other_session")
+                session.add(role)
+
+            # The data must have been committed (begin/commit, not a savepoint),
+            # so a separate engine/session can read it.
+            other_engine = create_async_engine(DATABASE_URL, echo=False)
+            try:
+                other_session_maker = async_sessionmaker(
+                    other_engine, expire_on_commit=False
+                )
+                async with other_session_maker() as other:
+                    result = await other.execute(select(Role).where(Role.id == role_id))
+                    fetched = result.scalar_one_or_none()
+                    assert fetched is not None, (
+                        "Fixture data inserted via get_transaction() must be committed "
+                        "and visible to a separate session. If create_db_session uses "
+                        "create_db_context, auto-begin forces get_transaction() into "
+                        "savepoints instead of real commits."
+                    )
+                    assert fetched.name == "visible_to_other_session"
+            finally:
+                await other_engine.dispose()
+
+        # Cleanup
+        async with create_db_session(DATABASE_URL, Base, drop_tables=True) as _:
+            pass
+
 
 class TestGetXdistWorker:
     """Tests for _get_xdist_worker helper."""

uv.lock

@@ -285,6 +285,7 @@ dev = [
     { name = "coverage" },
     { name = "fastapi-toolsets", extra = ["all"] },
     { name = "httpx" },
+    { name = "mike" },
     { name = "mkdocstrings-python" },
     { name = "prek" },
     { name = "pytest" },
@@ -296,6 +297,7 @@ dev = [
     { name = "zensical" },
 ]
 docs = [
+    { name = "mike" },
     { name = "mkdocstrings-python" },
     { name = "zensical" },
 ]
@@ -328,6 +330,7 @@ dev = [
     { name = "coverage", specifier = ">=7.0.0" },
     { name = "fastapi-toolsets", extras = ["all"] },
     { name = "httpx", specifier = ">=0.25.0" },
+    { name = "mike", git = "https://github.com/squidfunk/mike.git?tag=2.2.0%2Bzensical-0.1.0" },
     { name = "mkdocstrings-python", specifier = ">=2.0.2" },
     { name = "prek", specifier = ">=0.3.8" },
     { name = "pytest", specifier = ">=8.0.0" },
@@ -336,11 +339,12 @@ dev = [
     { name = "pytest-xdist", specifier = ">=3.0.0" },
     { name = "ruff", specifier = ">=0.1.0" },
     { name = "ty", specifier = ">=0.0.1a0" },
-    { name = "zensical", specifier = ">=0.0.23" },
+    { name = "zensical", specifier = ">=0.0.30" },
 ]
 docs = [
+    { name = "mike", git = "https://github.com/squidfunk/mike.git?tag=2.2.0%2Bzensical-0.1.0" },
     { name = "mkdocstrings-python", specifier = ">=2.0.2" },
-    { name = "zensical", specifier = ">=0.0.23" },
+    { name = "zensical", specifier = ">=0.0.30" },
 ]
 tests = [
     { name = "coverage", specifier = ">=7.0.0" },
@@ -604,6 +608,17 @@ wheels = [
     { url = "https://files.pythonhosted.org/packages/2c/19/04f9b178c2d8a15b076c8b5140708fa6ffc5601fb6f1e975537072df5b2a/mergedeep-1.3.4-py3-none-any.whl", hash = "sha256:70775750742b25c0d8f36c55aed03d24c3384d17c951b3175d898bd778ef0307", size = 6354, upload-time = "2021-02-05T18:55:29.583Z" },
 ]
 
+[[package]]
+name = "mike"
+version = "2.2.0+zensical.0.1.0"
+source = { git = "https://github.com/squidfunk/mike.git?tag=2.2.0%2Bzensical-0.1.0#0f62791256ebeba60d20d2f1d8fe6ec3b7d1e2b3" }
+dependencies = [
+    { name = "jinja2" },
+    { name = "pyparsing" },
+    { name = "verspec" },
+    { name = "zensical" },
+]
+
 [[package]]
 name = "mkdocs"
 version = "1.6.1"
@@ -870,24 +885,33 @@ wheels = [
 
 [[package]]
 name = "pygments"
-version = "2.19.2"
+version = "2.20.0"
 source = { registry = "https://pypi.org/simple" }
-sdist = { url = "https://files.pythonhosted.org/packages/b0/77/a5b8c569bf593b0140bde72ea885a803b82086995367bf2037de0159d924/pygments-2.19.2.tar.gz", hash = "sha256:636cb2477cec7f8952536970bc533bc43743542f70392ae026374600add5b887", size = 4968631, upload-time = "2025-06-21T13:39:12.283Z" }
+sdist = { url = "https://files.pythonhosted.org/packages/c3/b2/bc9c9196916376152d655522fdcebac55e66de6603a76a02bca1b6414f6c/pygments-2.20.0.tar.gz", hash = "sha256:6757cd03768053ff99f3039c1a36d6c0aa0b263438fcab17520b30a303a82b5f", size = 4955991, upload-time = "2026-03-29T13:29:33.898Z" }
 wheels = [
-    { url = "https://files.pythonhosted.org/packages/c7/21/705964c7812476f378728bdf590ca4b771ec72385c533964653c68e86bdc/pygments-2.19.2-py3-none-any.whl", hash = "sha256:86540386c03d588bb81d44bc3928634ff26449851e99741617ecb9037ee5ec0b", size = 1225217, upload-time = "2025-06-21T13:39:07.939Z" },
+    { url = "https://files.pythonhosted.org/packages/f4/7e/a72dd26f3b0f4f2bf1dd8923c85f7ceb43172af56d63c7383eb62b332364/pygments-2.20.0-py3-none-any.whl", hash = "sha256:81a9e26dd42fd28a23a2d169d86d7ac03b46e2f8b59ed4698fb4785f946d0176", size = 1231151, upload-time = "2026-03-29T13:29:30.038Z" },
 ]
 
 [[package]]
 name = "pymdown-extensions"
-version = "10.21"
+version = "10.21.2"
 source = { registry = "https://pypi.org/simple" }
 dependencies = [
     { name = "markdown" },
     { name = "pyyaml" },
 ]
-sdist = { url = "https://files.pythonhosted.org/packages/ba/63/06673d1eb6d8f83c0ea1f677d770e12565fb516928b4109c9e2055656a9e/pymdown_extensions-10.21.tar.gz", hash = "sha256:39f4a020f40773f6b2ff31d2cd2546c2c04d0a6498c31d9c688d2be07e1767d5", size = 853363, upload-time = "2026-02-15T20:44:06.748Z" }
+sdist = { url = "https://files.pythonhosted.org/packages/df/08/f1c908c581fd11913da4711ea7ba32c0eee40b0190000996bb863b0c9349/pymdown_extensions-10.21.2.tar.gz", hash = "sha256:c3f55a5b8a1d0edf6699e35dcbea71d978d34ff3fa79f3d807b8a5b3fa90fbdc", size = 853922, upload-time = "2026-03-29T15:01:55.233Z" }
 wheels = [
-    { url = "https://files.pythonhosted.org/packages/6f/2c/5b079febdc65e1c3fb2729bf958d18b45be7113828528e8a0b5850dd819a/pymdown_extensions-10.21-py3-none-any.whl", hash = "sha256:91b879f9f864d49794c2d9534372b10150e6141096c3908a455e45ca72ad9d3f", size = 268877, upload-time = "2026-02-15T20:44:05.464Z" },
+    { url = "https://files.pythonhosted.org/packages/f7/27/a2fc51a4a122dfd1015e921ae9d22fee3d20b0b8080d9a704578bf9deece/pymdown_extensions-10.21.2-py3-none-any.whl", hash = "sha256:5c0fd2a2bea14eb39af8ff284f1066d898ab2187d81b889b75d46d4348c01638", size = 268901, upload-time = "2026-03-29T15:01:53.244Z" },
+]
+
+[[package]]
+name = "pyparsing"
+version = "3.3.2"
+source = { registry = "https://pypi.org/simple" }
+sdist = { url = "https://files.pythonhosted.org/packages/f3/91/9c6ee907786a473bf81c5f53cf703ba0957b23ab84c264080fb5a450416f/pyparsing-3.3.2.tar.gz", hash = "sha256:c777f4d763f140633dcb6d8a3eda953bf7a214dc4eff598413c070bcdc117cbc", size = 6851574, upload-time = "2026-01-21T03:57:59.36Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/10/bd/c038d7cc38edc1aa5bf91ab8068b63d4308c66c4c8bb3cbba7dfbc049f9c/pyparsing-3.3.2-py3-none-any.whl", hash = "sha256:850ba148bd908d7e2411587e247a1e4f0327839c40e2e5e6d05a007ecc69911d", size = 122781, upload-time = "2026-01-21T03:57:55.912Z" },
 ]
 
 [[package]]
@@ -1262,6 +1286,15 @@ wheels = [
     { url = "https://files.pythonhosted.org/packages/dc/9b/47798a6c91d8bdb567fe2698fe81e0c6b7cb7ef4d13da4114b41d239f65d/typing_inspection-0.4.2-py3-none-any.whl", hash = "sha256:4ed1cacbdc298c220f1bd249ed5287caa16f34d44ef4e9c3d0cbad5b521545e7", size = 14611, upload-time = "2025-10-01T02:14:40.154Z" },
 ]
 
+[[package]]
+name = "verspec"
+version = "0.1.0"
+source = { registry = "https://pypi.org/simple" }
+sdist = { url = "https://files.pythonhosted.org/packages/e7/44/8126f9f0c44319b2efc65feaad589cadef4d77ece200ae3c9133d58464d0/verspec-0.1.0.tar.gz", hash = "sha256:c4504ca697b2056cdb4bfa7121461f5a0e81809255b41c03dda4ba823637c01e", size = 27123, upload-time = "2020-11-30T02:24:09.646Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/a4/ce/3b6fee91c85626eaf769d617f1be9d2e15c1cca027bbdeb2e0d751469355/verspec-0.1.0-py3-none-any.whl", hash = "sha256:741877d5633cc9464c45a469ae2a31e801e6dbbaa85b9675d481cda100f11c31", size = 19640, upload-time = "2020-11-30T02:24:08.387Z" },
+]
+
 [[package]]
 name = "watchdog"
 version = "6.0.0"
@@ -1291,7 +1324,7 @@ wheels = [
 
 [[package]]
 name = "zensical"
-version = "0.0.29"
+version = "0.0.30"
 source = { registry = "https://pypi.org/simple" }
 dependencies = [
     { name = "click" },
@@ -1301,18 +1334,18 @@ dependencies = [
     { name = "pymdown-extensions" },
     { name = "pyyaml" },
 ]
-sdist = { url = "https://files.pythonhosted.org/packages/78/bd/5786ab618a60bd7469ab243a7fd2c9eecb0790c85c784abb8b97edb77a54/zensical-0.0.29.tar.gz", hash = "sha256:0d6282be7cb551e12d5806badf5e94c54a5e2f2cf07057a3e36d1eaf97c33ada", size = 3842641, upload-time = "2026-03-24T13:37:27.587Z" }
+sdist = { url = "https://files.pythonhosted.org/packages/1d/53/5e551f8912718816733a75adcb53a0787b2d2edca5869c156325aaf82e24/zensical-0.0.30.tar.gz", hash = "sha256:408b531683f6bcb6cc5ab928146d2c68afbc16fac4eda87ae3dd20af1498180f", size = 3844287, upload-time = "2026-03-28T17:55:52.836Z" }
 wheels = [
-    { url = "https://files.pythonhosted.org/packages/4b/9c/8b681daa024abca9763017bec09ecee8008e110cae1254217c8dd22cc339/zensical-0.0.29-cp310-abi3-macosx_10_12_x86_64.whl", hash = "sha256:20ae0709ea14fce25ab33d0a82acdaf454a7a2e232a9ee20c019942205174476", size = 12311399, upload-time = "2026-03-24T13:36:53.809Z" },
-    { url = "https://files.pythonhosted.org/packages/81/ae/4ebb4d8bb2ef0164d473698b92f11caf431fc436e1625524acd5641102ca/zensical-0.0.29-cp310-abi3-macosx_11_0_arm64.whl", hash = "sha256:599af3ba66fcd0146d7019f3493ed3c316051fae6c4d5599bc59f3a8f4b8a6f0", size = 12191845, upload-time = "2026-03-24T13:36:56.909Z" },
-    { url = "https://files.pythonhosted.org/packages/d5/35/67f89db06571a52283b3ecbe3bcf32fd3115ca50436b3ae177a948b83ea7/zensical-0.0.29-cp310-abi3-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:eea7e48a00a71c0586e875079b5f83a070c33a147e52ad4383e4b63ab524332b", size = 12554105, upload-time = "2026-03-24T13:36:59.945Z" },
-    { url = "https://files.pythonhosted.org/packages/7c/f6/ac79e5d9c18b28557c9ff1c7c23d695fbdd82645d69bfe02292f46d935e7/zensical-0.0.29-cp310-abi3-manylinux_2_17_armv7l.manylinux2014_armv7l.whl", hash = "sha256:59a57db35542e98d2896b833de07d199320f8ada3b4e7ddccb7fe892292d8b74", size = 12498643, upload-time = "2026-03-24T13:37:02.376Z" },
-    { url = "https://files.pythonhosted.org/packages/b1/70/5c22a96a69e0e91e569c26236918bb9bab1170f59b29ad04105ead64f199/zensical-0.0.29-cp310-abi3-manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:d42c2b2a96a80cf64c98ba7242f59ef95109914bd4c9499d7ebc12544663852c", size = 12854531, upload-time = "2026-03-24T13:37:04.962Z" },
-    { url = "https://files.pythonhosted.org/packages/79/25/e32237a8fcb0ceae1ef8e192e7f8db53b38f1e48f1c7cdbacd0a7b713892/zensical-0.0.29-cp310-abi3-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:6b2fca39c5f6b1782c77cf6591cf346357cabee85ebdb956c5ddc0fd5169f3d9", size = 12596828, upload-time = "2026-03-24T13:37:07.817Z" },
-    { url = "https://files.pythonhosted.org/packages/ff/74/89ac909cbb258903ea53802c184e4986c17ce0ba79b1c7f77b7e78a2dce3/zensical-0.0.29-cp310-abi3-musllinux_1_2_aarch64.whl", hash = "sha256:dfc23a74ef672aa51088c080286319da1dc0b989cd5051e9e5e6d7d4abbc2fc1", size = 12732059, upload-time = "2026-03-24T13:37:11.651Z" },
-    { url = "https://files.pythonhosted.org/packages/8c/31/2429de6a9328eed4acc7e9a3789f160294a15115be15f9870a0d02649302/zensical-0.0.29-cp310-abi3-musllinux_1_2_armv7l.whl", hash = "sha256:c9336d4e4b232e3c9a70e30258e916dd7e60c0a2a08c8690065e60350c302028", size = 12768542, upload-time = "2026-03-24T13:37:14.39Z" },
-    { url = "https://files.pythonhosted.org/packages/10/8a/55588b2a1dcbe86dad0404506c9ba367a06c663b1ff47147c84d26f7510e/zensical-0.0.29-cp310-abi3-musllinux_1_2_i686.whl", hash = "sha256:30661148f0681199f3b598cbeb1d54f5cba773e54ae840bac639250d85907b84", size = 12917991, upload-time = "2026-03-24T13:37:16.795Z" },
-    { url = "https://files.pythonhosted.org/packages/ec/5d/653901f0d3a3ca72daebc62746a148797f4e422cc3a2b66a4e6718e4398f/zensical-0.0.29-cp310-abi3-musllinux_1_2_x86_64.whl", hash = "sha256:6a566ac1fd4bfac5d711a7bd1ae06666712127c2718daa5083c7bf3f107e8578", size = 12868392, upload-time = "2026-03-24T13:37:19.42Z" },
-    { url = "https://files.pythonhosted.org/packages/29/58/d7449bc88a174b98daa3f2fbdfbdac3493768a557d8987e88bdaa6c78b1a/zensical-0.0.29-cp310-abi3-win32.whl", hash = "sha256:a231a3a02a3851741dc4d2de8910b5c39fe81e55bf026d8edf4d803e91a922fb", size = 11905486, upload-time = "2026-03-24T13:37:22.154Z" },
-    { url = "https://files.pythonhosted.org/packages/f5/09/3fd082d016497c4d26ff20f42a8be2cc91e27191c0c5f3cd6507827f666f/zensical-0.0.29-cp310-abi3-win_amd64.whl", hash = "sha256:7145c5504380a344b8cd4586da815cdde77ef4a42319fa4f35e78250f01985af", size = 12101510, upload-time = "2026-03-24T13:37:24.77Z" },
+    { url = "https://files.pythonhosted.org/packages/1b/e3/ac0eb77a8a7f793613813de68bde26776d0da68d8041fa9eb8d0b986a449/zensical-0.0.30-cp310-abi3-macosx_10_12_x86_64.whl", hash = "sha256:b67fca8bfcd71c94b331045a591bf6e24fe123a66fba94587aa3379faf521a16", size = 12313786, upload-time = "2026-03-28T17:55:18.839Z" },
+    { url = "https://files.pythonhosted.org/packages/a5/6a/73e461dfa27d3bc415e48396f83a3287b43df2fd3361e25146bc86360aab/zensical-0.0.30-cp310-abi3-macosx_11_0_arm64.whl", hash = "sha256:8ceadfece1153edc26506e8ddf68d9818afe8517cf3bcdb6bfe4cb2793ae247b", size = 12186136, upload-time = "2026-03-28T17:55:21.836Z" },
+    { url = "https://files.pythonhosted.org/packages/a3/bc/9022156b4c28c1b95209acb64319b1e5cd0af2e97035bdd461e58408cb46/zensical-0.0.30-cp310-abi3-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:e100b2b654337ac5306ba12818f3c5336c66d0d34c593ef05e316c124a5819cb", size = 12556115, upload-time = "2026-03-28T17:55:24.849Z" },
+    { url = "https://files.pythonhosted.org/packages/0b/29/9e8f5bd6d33b35f4c368ae8b13d431dc42b2de17ea6eccbd71d48122eba6/zensical-0.0.30-cp310-abi3-manylinux_2_17_armv7l.manylinux2014_armv7l.whl", hash = "sha256:bdf641ffddaf21c6971b91a4426b81cd76271c5b1adb7176afcce3f1508328b1", size = 12498121, upload-time = "2026-03-28T17:55:27.637Z" },
+    { url = "https://files.pythonhosted.org/packages/c4/e1/b8dfa0769050e62cd731358145fdeb67af35e322197bd7e7727250596e7b/zensical-0.0.30-cp310-abi3-manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:1fd909a0c2116e26190c7f3ec4fb55837c417b7a8d99ebf4f3deb26b07b97e49", size = 12854142, upload-time = "2026-03-28T17:55:30.54Z" },
+    { url = "https://files.pythonhosted.org/packages/04/11/62a36cfb81522b6108db8f9e96d36da8cccb306b02c15ad19e1b333fa7c8/zensical-0.0.30-cp310-abi3-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:16fd2da09fe4e5cbec2ca74f31abc70f32f7330d56593b647e0a114bb329171a", size = 12598341, upload-time = "2026-03-28T17:55:32.988Z" },
+    { url = "https://files.pythonhosted.org/packages/a7/a4/8c7a6725fb226aa71d19209403d974e45f39d757e725f9558c6ed8d350a5/zensical-0.0.30-cp310-abi3-musllinux_1_2_aarch64.whl", hash = "sha256:896b36eaef7fed5f8fc6f2c8264b2751aad63c2d66d3d8650e38481b6b4f6f7b", size = 12732307, upload-time = "2026-03-28T17:55:35.618Z" },
+    { url = "https://files.pythonhosted.org/packages/5e/a1/7858fb3f6ac67d7d24a8acbe834cbe26851d6bd151ece6fba3fc88b0f878/zensical-0.0.30-cp310-abi3-musllinux_1_2_armv7l.whl", hash = "sha256:a1f515ec67a0d0250e53846327bf0c69635a1f39749da3b04feb68431188d3c6", size = 12770962, upload-time = "2026-03-28T17:55:38.627Z" },
+    { url = "https://files.pythonhosted.org/packages/49/b7/228298112a69d0b74e6e93041bffcf1fc96d03cf252be94a354f277d4789/zensical-0.0.30-cp310-abi3-musllinux_1_2_i686.whl", hash = "sha256:ce33d1002438838a35fa43358a1f43d74f874586596d3d116999d3756cded00e", size = 12919256, upload-time = "2026-03-28T17:55:41.413Z" },
+    { url = "https://files.pythonhosted.org/packages/de/c7/5b4ea036f7f7d84abf907f7f7a3e8420b054c89279c5273ca248d3bc9f48/zensical-0.0.30-cp310-abi3-musllinux_1_2_x86_64.whl", hash = "sha256:029dad561568f4ae3056dde16a81012efd92c426d4eb7101f960f448c1168196", size = 12869760, upload-time = "2026-03-28T17:55:44.474Z" },
+    { url = "https://files.pythonhosted.org/packages/36/b4/77bef2132e43108db718ae014a5961fc511e88fc446c11f1c3483def429e/zensical-0.0.30-cp310-abi3-win32.whl", hash = "sha256:0105672850f053c326fba9fdd95adf60e9f90308f8cc1c08e3a00e15a8d5e90f", size = 11905658, upload-time = "2026-03-28T17:55:47.416Z" },
+    { url = "https://files.pythonhosted.org/packages/a1/59/23b6c7ff062e2b299cc60e333095e853f9d38d1b5abe743c7b94c4ac432c/zensical-0.0.30-cp310-abi3-win_amd64.whl", hash = "sha256:b879dbf4c69d3ea41694bae33e1b948847e635dcbcd6ec8c522920833379dd48", size = 12101867, upload-time = "2026-03-28T17:55:50.083Z" },
 ]

zensical.toml

@@ -140,6 +140,7 @@ Examples = [
 
 [[project.nav]]
 Migration = [
+    {"v3.0" = "migration/v3.md"},
     {"v2.0" = "migration/v2.md"},
 ]