Change scenes manually¶
Sometimes it helps to have more control over how one swaps scenes around. As mentioned above, a Viewport's child nodes will render to the image it generates. This holds true even for nodes outside of the "current" scene. Autoloads fall into this category, but so do scenes which one instances and adds to the tree at runtime:
To complete the cycle and swap out the new scene with the old one, developers have a choice to make. Many strategies exist for removing a scene from view of the Viewport. The tradeoffs involve balancing operation speed and memory consumption as well as balancing data access and integrity.
We can delete the existing scene. SceneTree.change_scene() and SceneTree.change_scene_to() will delete the current scene immediately. Developers can also delete the main scene though. Assuming the root node's name is "Main", one could do
get_node("/root/Main").free()to delete the whole scene.
Pro: RAM is no longer dragging the dead weight.
Con: Returning to that scene is now more expensive since it must be loaded back into memory again (takes time AND memory). Not a problem if returning soon is unnecessary.
Con: No longer have access to that scene's data. Not a problem if using that data soon is unnecessary.
Note: It can be useful to preserve the data in a soon-to-be-deleted scene by re-attaching one or more of its nodes to a different scene, or even directly to the SceneTree.
Pro: No nodes means no process, physics process, or input handling. The CPU is available to work on the new scene's contents.
Con: Those nodes' processing and input handling no longer operate. Not a problem if using the updated data is unnecessary.
We can hide the existing scene. By changing the visibility or collision detection of the nodes, we can hide the entire node sub-tree from the player's perspective.
Memory still exists.
Pro: One can still access the data if need be.
Pro: There's no need to move any more nodes around to save data.
Con: More data is being kept in memory which will be become a problem on memory-sensitive platforms like web or mobile.
Pro: Data continues to receive processing updates, so the scene will keep updated any data within it that relies on delta time or frame data.
Pro: Nodes are still members of groups (since groups belong to the SceneTree).
Con: The CPU's attention is now divided between both scenes. Too much load could result in low frame rates. One should be sure to test performance as they go to ensure the target platform can support the load they are giving it.
We can remove the existing scene from the tree. Assign a variable to the existing scene's root node. Then use Node.remove_child(Node) to detach the entire scene from the tree.
Memory still exists (similar pros/cons as with hiding it from view).
Processing stops (similar pros/cons as with deleting it completely).
Pro: This variation of "hiding" it is much easier to show/hide. Rather than potentially keeping track of multiple changes to the scene, one must only call the one method add/remove_child pair of methods. It is similar to disabling game objects in other engines.
Con: Unlike with hiding it from view only, the data contained within the scene will become stale if it relies on delta time, input, groups, or other data that is derived from SceneTree access.
There are also cases where one may wish to have many scenes present at the same time. Perhaps one is adding their own singleton at runtime, or preserving a a scene's data between scene changes (adding the scene to the root node).
Perhaps instead they wish to display multiple scenes at the same time using ViewportContainers. This is optimal in cases where the intent is to render different content in different parts of the screen. Minimaps and split-screen multiplayer are good examples.
Each option will have cases where it is best appropriate, so one must examine the effects of each and determine what path best fits their unique situation.