Every Object in Godot implements a
_notification method. Its purpose is to
allow the Object to respond to a variety of engine-level callbacks that may
relate to it. For example, if the engine tells a
CanvasItem to "draw", it will call
Some of these notifications, like draw, are useful to override in scripts. So much so that Godot exposes many of them with dedicated functions:
What users might not realize is that notifications exist for types other than Node alone:
Object::NOTIFICATION_POSTINITIALIZE: a callback that triggers during object initialization. Not accessible to scripts.
Object::NOTIFICATION_PREDELETE: a callback that triggers before the engine deletes an Object, i.e. a 'destructor'.
MainLoop::NOTIFICATION_WM_MOUSE_ENTER: a callback that triggers when the mouse enters the window in the operating system that displays the game content.
And many of the callbacks that do exist in Nodes don't have any dedicated methods, but are still quite useful.
Node::NOTIFICATION_PARENTED: a callback that triggers anytime one adds a child node to another node.
Node::NOTIFICATION_UNPARENTED: a callback that triggers anytime one removes a child node from another node.
Popup::NOTIFICATION_POST_POPUP: a callback that triggers after a Popup node completes any
popup*method. Note the difference from its
about_to_showsignal which triggers before its appearance.
One can access all these custom notifications from the universal
Methods in the documentation labeled as "virtual" are also intended to be overridden by scripts.
A classic example is the
_init method in Object. While it has no
NOTIFICATION_* equivalent, the engine still calls the method. Most languages
(except C#) rely on it as a constructor.
So, in which situation should one use each of these notifications or virtual functions?
_process vs. _physics_process vs. *_input¶
_process when one needs a framerate-dependent deltatime between
frames. If code that updates object data needs to update as often as
possible, this is the right place. Recurring logic checks and data caching
often execute here, but it comes down to the frequency at which one needs
the evaluations to update. If they don't need to execute every frame, then
implementing a Timer-yield-timeout loop is another option.
_physics_process when one needs a framerate-independent deltatime
between frames. If code needs consistent updates over time, regardless
of how fast or slow time advances, this is the right place.
Recurring kinematic and object transform operations should execute here.
While it is possible, to achieve the best performance, one should avoid
making input checks during these callbacks.
_physics_process will trigger at every opportunity (they do not "rest" by
default). In contrast,
*_input callbacks will trigger only on frames in
which the engine has actually detected the input.
One can check for input actions within the input callbacks just the same. If one wants to use delta time, one can fetch it from the related deltatime methods as needed.
_init vs. initialization vs. export¶
If the script initializes its own node subtree, without a scene,
that code should execute here. Other property or SceneTree-independent
initializations should also run here. This triggers before
_enter_tree, but after a script creates and initializes its properties.
Scripts have three types of property assignments that can occur during instantiation:
When instantiating a scene, property values will set up according to the following sequence:
Initial value assignment: instantiation will assign either the initialization value or the init assignment value. Init assignments take priority over initialization values.
Exported value assignment: If instancing from a scene rather than a script, Godot will assign the exported value to replace the initial value defined in the script.
As a result, instantiating a script versus a scene will affect both the initialization and the number of times the engine calls the setter.
_ready vs. _enter_tree vs. NOTIFICATION_PARENTED¶
When instantiating a scene connected to the first executed scene, Godot will
instantiate nodes down the tree (making
_init calls) and build the tree
going downwards from the root. This causes
_enter_tree calls to cascade
down the tree. Once the tree is complete, leaf nodes call
_ready. A node
will call this method once all child nodes have finished calling theirs. This
then causes a reverse cascade going up back to the tree's root.
When instantiating a script or a standalone scene, nodes are not
added to the SceneTree upon creation, so no
trigger. Instead, only the
_init and later
_ready calls occur.
If one needs to trigger behavior that occurs as nodes parent to another, regardless of whether it occurs as part of the main/active scene or not, one can use the PARENTED notification. For example, here is a snippet that connects a node's method to a custom signal on the parent node without failing. Useful on data-centric nodes that one might create at runtime.