feat(settings): support @cocoindex.settings and make init() optional

docs/docs/core/settings.mdx

@@ -10,49 +10,82 @@ import TabItem from '@theme/TabItem';
 
 Certain settings need to be provided for CocoIndex to work, e.g. database connections, app namespace, etc.
 
-## Launch CocoIndex
+## Configure CocoIndex Settings
 
-You have two ways to launch CocoIndex:
+Note that in general, you have two ways to launch CocoIndex:
 
+*   Call CocoIndex APIs from your own Python application or library.
 *   Use [Cocoindex CLI](cli). It's handy for most routine indexing building and management tasks.
-    It will load settings from environment variables, either already set in your environment, or specified in `.env` file.
-    See [CLI](cli#environment-variables) for more details.
 
-*   Call CocoIndex functionality from your own Python application or library.
-    It's needed when you want to leverage CocoIndex support for query, or have your custom logic to trigger indexing, etc.
 
-    <Tabs>
-    <TabItem value="python" label="Python" default>
+CocoIndex exposes process-level settings specified by `cocoindex.Settings` dataclass.
+Settings can be configured in three different ways.
+In the following sections, the later ones will override the earlier ones.
 
-    You need to explicitly call `cocoindex.init()` before doing anything with CocoIndex, and settings will be loaded during the call.
+### Environment Variables
 
-    *   If it's called without any argument, it will load settings from environment variables.
-        Only existing environment variables already set in your environment will be used.
-        If you want to load environment variables from a specific `.env` file, consider call `load_dotenv()` provided by the [`python-dotenv`](https://github.com/theskumar/python-dotenv) package.
+The simplest approach is to set corresponding environment variables.
+See [List of Environment Variables](#list-of-environment-variables) for specific environment variables.
 
-        ```py
-        from dotenv import load_dotenv
-        import cocoindex
+:::tip
 
-        load_dotenv()
-        cocoindex.init()
-        ```
+You can consider place a `.env` file in your directory.
+The [CLI](cli#environment-variables) will load environment variables from the `.env` file (see [CLI](cli#environment-variables) for more details).
+From your own main module, you can also load environment variables with a package like [`python-dotenv`](https://github.com/theskumar/python-dotenv).
 
-    *   It takes an optional `cocoindex.Settings` dataclass object as argument, so you can also construct settings explicitly and pass to it:
+:::
+
+### Setting Function
+
+A more flexible approach is to provide a setting function that returns a `cocoindex.Settings` dataclass object.
+The setting function can have any name, and needs to be decorated with the `@cocoindex.settings` decorator, for example:
+
+```py
+@cocoindex.settings
+def cocoindex_settings() -> cocoindex.Settings:
+    return cocoindex.Settings(
+        database=cocoindex.DatabaseConnectionSpec(
+            url="postgres://cocoindex:cocoindex@localhost/cocoindex"
+        )
+    )
+```
+
+This setting function will be called once when CocoIndex is initialized.
+Once the settings function is provided, environment variables will be ignored.
 
-        ```py
-        import cocoindex
+### `cocoindex.init()` function
 
-        cocoindex.init(
-            cocoindex.Settings(
-                database=cocoindex.DatabaseConnectionSpec(
-                    url="postgres://cocoindex:cocoindex@localhost/cocoindex"
-                )
-            )
+You can also call `cocoindex.init()` with a `cocoindex.Settings` dataclass object as argument, for example:
+
+```py
+cocoindex.init(
+    cocoindex.Settings(
+        database=cocoindex.DatabaseConnectionSpec(
+            url="postgres://cocoindex:cocoindex@localhost/cocoindex"
         )
-        ```
-    </TabItem>
-    </Tabs>
+    )
+)
+```
+
+For example, you can call it in the main function of your application.
+Once the `cocoindex.init()` is called with a `cocoindex.Settings` dataclass object as argument, the `@cocoindex.settings` function and environment variables will be ignored.
+
+This is more flexible, as you can more easily construct `cocoindex.Settings` based on other stuffs you loaded earlier.
+But be careful that if you call `cocoindex.init()` only under the path of main (e.g. within `if __name__ == "__main__":` guard), it won't be executed when you're using CocoIndex CLI, as it won't execute your main logic.
+
+:::info
+
+`cocoindex.init()` is optional:
+
+-   You can call `cocoindex.init()` with a `cocoindex.Settings` dataclass object as argument, or without any argument.
+    When without argument, the settings will be loaded from the `@cocoindex.settings` function or environment variables.
+
+-   You don't have to explicitly call `cocoindex.init()`.
+    CocoIndex will be automatically initialized when needed, e.g. when any method of any flow is called the first time.
+    But calling `cocoindex.init()` explicitly (usually at startup time, e.g. in the main function of your application) has the benefit of making sure CocoIndex library is initialized and any potential exceptions are raised earlier before proceeding with the application.
+    If you need this clarity, you can call it explicitly even if you don't want to provide settings by the `cocoindex.init()` call.
+
+:::
 
 ## List of Settings
 

python/cocoindex/__init__.py

@@ -6,15 +6,20 @@
 
 from . import targets as storages  # Deprecated: Use targets instead
 
-from .auth_registry import AuthEntryReference, add_auth_entry, add_transient_auth_entry
+from .auth_registry import (
+    AuthEntryReference,
+    add_auth_entry,
+    add_transient_auth_entry,
+    ref_auth_entry,
+)
 from .flow import FlowBuilder, DataScope, DataSlice, Flow, transform_flow
 from .flow import flow_def
 from .flow import EvaluateAndDumpOptions, GeneratedField
 from .flow import FlowLiveUpdater, FlowLiveUpdaterOptions, FlowUpdaterStatusUpdates
 from .flow import open_flow
 from .flow import add_flow_def, remove_flow  # DEPRECATED
 from .flow import update_all_flows_async, setup_all_flows, drop_all_flows
-from .lib import init, start_server, stop
+from .lib import settings, init, start_server, stop
 from .llm import LlmSpec, LlmApiType
 from .index import VectorSimilarityMetric, VectorIndexDef, IndexOptions
 from .setting import DatabaseConnectionSpec, Settings, ServerSettings
@@ -65,6 +70,7 @@
     "setup_all_flows",
     "drop_all_flows",
     # Lib
+    "settings",
     "init",
     "start_server",
     "stop",

python/cocoindex/cli.py

@@ -91,8 +91,6 @@ def _load_user_app(app_target: str) -> None:
 
 
 def _initialize_cocoindex_in_process() -> None:
-    settings = setting.Settings.from_env()
-    lib.init(settings)
     atexit.register(lib.stop)
 
 

python/cocoindex/flow.py

@@ -694,16 +694,11 @@ class Flow:
     """
 
     _name: str
-    _full_name: str
     _lazy_engine_flow: Callable[[], _engine.Flow] | None
 
-    def __init__(
-        self, name: str, full_name: str, engine_flow_creator: Callable[[], _engine.Flow]
-    ):
+    def __init__(self, name: str, engine_flow_creator: Callable[[], _engine.Flow]):
         validate_flow_name(name)
-        validate_full_flow_name(full_name)
         self._name = name
-        self._full_name = full_name
         engine_flow = None
         lock = Lock()
 
@@ -762,7 +757,7 @@ def full_name(self) -> str:
         """
         Get the full name of the flow.
         """
-        return self._full_name
+        return get_flow_full_name(self._name)
 
     def update(self, /, *, reexport_targets: bool = False) -> _engine.IndexUpdateInfo:
         """
@@ -861,9 +856,10 @@ def _create_lazy_flow(
     The flow will be built the first time when it's really needed.
     """
     flow_name = _flow_name_builder.build_name(name, prefix="_flow_")
-    flow_full_name = get_flow_full_name(flow_name)
 
     def _create_engine_flow() -> _engine.Flow:
+        flow_full_name = get_flow_full_name(flow_name)
+        validate_full_flow_name(flow_full_name)
         flow_builder_state = _FlowBuilderState(flow_full_name)
         root_scope = DataScope(
             flow_builder_state, flow_builder_state.engine_flow_builder.root_scope()
@@ -873,7 +869,7 @@ def _create_engine_flow() -> _engine.Flow:
             execution_context.event_loop
         )
 
-    return Flow(flow_name, flow_full_name, _create_engine_flow)
+    return Flow(flow_name, _create_engine_flow)
 
 
 _flows_lock = Lock()

python/cocoindex/lib.py

@@ -2,12 +2,57 @@
 Library level functions and states.
 """
 
+import threading
 import warnings
-from typing import Callable, Any
 
 from . import _engine  # type: ignore
 from . import flow, setting
 from .convert import dump_engine_object
+from .validation import validate_app_namespace_name
+from typing import Any, Callable, overload
+
+
+def prepare_settings(settings: setting.Settings) -> Any:
+    """Prepare the settings for the engine."""
+    if settings.app_namespace:
+        validate_app_namespace_name(settings.app_namespace)
+    return dump_engine_object(settings)
+
+
+_engine.set_settings_fn(lambda: prepare_settings(setting.Settings.from_env()))
+
+
+_prev_settings_fn: Callable[[], setting.Settings] | None = None
+_prev_settings_fn_lock: threading.Lock = threading.Lock()
+
+
+@overload
+def settings(fn: Callable[[], setting.Settings]) -> Callable[[], setting.Settings]: ...
+@overload
+def settings(
+    fn: None,
+) -> Callable[[Callable[[], setting.Settings]], Callable[[], setting.Settings]]: ...
+def settings(fn: Callable[[], setting.Settings] | None = None) -> Any:
+    """
+    Decorate a function that returns a settings.Settings object.
+    It registers the function as a settings provider.
+    """
+
+    def _inner(fn: Callable[[], setting.Settings]) -> Callable[[], setting.Settings]:
+        global _prev_settings_fn  # pylint: disable=global-statement
+        with _prev_settings_fn_lock:
+            if _prev_settings_fn is not None:
+                warnings.warn(
+                    f"Setting a new settings function will override the previous one {_prev_settings_fn}."
+                )
+            _prev_settings_fn = fn
+        _engine.set_settings_fn(lambda: prepare_settings(fn()))
+        return fn
+
+    if fn is not None:
+        return _inner(fn)
+    else:
+        return _inner
 
 
 def init(settings: setting.Settings | None = None) -> None:
@@ -16,9 +61,7 @@ def init(settings: setting.Settings | None = None) -> None:
 
     If the settings are not provided, they are loaded from the environment variables.
     """
-    settings = settings or setting.Settings.from_env()
-    _engine.init(dump_engine_object(settings))
-    setting.set_app_namespace(settings.app_namespace)
+    _engine.init(prepare_settings(settings) if settings is not None else None)
 
 
 def start_server(settings: setting.ServerSettings) -> None:

python/cocoindex/setting.py

@@ -6,16 +6,15 @@
 
 from typing import Callable, Self, Any, overload
 from dataclasses import dataclass
-from .validation import validate_app_namespace_name
-
-_app_namespace: str = ""
+from . import _engine  # type: ignore
 
 
 def get_app_namespace(*, trailing_delimiter: str | None = None) -> str:
     """Get the application namespace. Append the `trailing_delimiter` if not empty."""
-    if _app_namespace == "" or trailing_delimiter is None:
-        return _app_namespace
-    return f"{_app_namespace}{trailing_delimiter}"
+    app_namespace: str = _engine.get_app_namespace()
+    if app_namespace == "" or trailing_delimiter is None:
+        return app_namespace
+    return f"{app_namespace}{trailing_delimiter}"
 
 
 def split_app_namespace(full_name: str, delimiter: str) -> tuple[str, str]:
@@ -26,14 +25,6 @@ def split_app_namespace(full_name: str, delimiter: str) -> tuple[str, str]:
     return (parts[0], parts[1])
 
 
-def set_app_namespace(app_namespace: str) -> None:
-    """Set the application namespace."""
-    if app_namespace:
-        validate_app_namespace_name(app_namespace)
-    global _app_namespace  # pylint: disable=global-statement
-    _app_namespace = app_namespace
-
-
 @dataclass
 class DatabaseConnectionSpec:
     """

src/builder/analyzer.rs

@@ -995,7 +995,7 @@ pub async fn analyze_flow(
     impl Future<Output = Result<ExecutionPlan>> + Send + use<>,
 )> {
     let analyzer_ctx = AnalyzerContext {
-        lib_ctx: get_lib_context()?,
+        lib_ctx: get_lib_context().await?,
         flow_ctx,
     };
     let root_data_scope = Arc::new(Mutex::new(DataScopeBuilder::new()));
@@ -1109,7 +1109,7 @@ pub async fn analyze_transient_flow<'a>(
 )> {
     let mut root_data_scope = DataScopeBuilder::new();
     let analyzer_ctx = AnalyzerContext {
-        lib_ctx: get_lib_context()?,
+        lib_ctx: get_lib_context().await?,
         flow_ctx,
     };
     let mut input_fields = vec![];

src/builder/flow_builder.rs

@@ -247,8 +247,12 @@ pub struct FlowBuilder {
 #[pymethods]
 impl FlowBuilder {
     #[new]
-    pub fn new(name: &str) -> PyResult<Self> {
-        let lib_context = get_lib_context().into_py_result()?;
+    pub fn new(py: Python<'_>, name: &str) -> PyResult<Self> {
+        let lib_context = py
+            .allow_threads(|| -> anyhow::Result<Arc<LibContext>> {
+                get_runtime().block_on(get_lib_context())
+            })
+            .into_py_result()?;
         let root_op_scope = OpScope::new(
             spec::ROOT_SCOPE_NAME.to_string(),
             None,

src/execution/db_tracking_setup.rs

@@ -252,7 +252,7 @@ impl ResourceSetupChange for TrackingTableSetupChange {
 
 impl TrackingTableSetupChange {
     pub async fn apply_change(&self) -> Result<()> {
-        let lib_context = get_lib_context()?;
+        let lib_context = get_lib_context().await?;
         let pool = lib_context.require_builtin_db_pool()?;
         if let Some(desired) = &self.desired_state {
             for lagacy_name in self.legacy_tracking_table_names.iter() {