Logic preferences

Ever wondered whether one should approach problem X with strategy Y or Z? This article covers a variety of topics related to these dilemmas.

Adding nodes and changing properties: which first?

When initializing nodes from a script at runtime, you may need to change properties such as the node's name or position. A common dilemma is, when should you change those values?

It is the best practice to change values on a node before adding it to the scene tree. Some property's setters have code to update other corresponding values, and that code can be slow! For most cases, this code has no impact on your game's performance, but in heavy use cases such as procedural generation, it can bring your game to a crawl.

For these reasons, it is always a best practice to set the initial values of a node before adding it to the scene tree.

Loading vs. preloading

In GDScript, there exists the global preload method. It loads resources as early as possible to front-load the "loading" operations and avoid loading resources while in the middle of performance-sensitive code.

Its counterpart, the load method, loads a resource only when it reaches the load statement. That is, it will load a resource in-place which can cause slowdowns when it occurs in the middle of sensitive processes. The load() function is also an alias for ResourceLoader.load(path) which is accessible to all scripting languages.

So, when exactly does preloading occur versus loading, and when should one use either? Let's see an example:

# my_buildings.gd
extends Node

# Note how constant scripts/scenes have a different naming scheme than
# their property variants.

# This value is a constant, so it spawns when the Script object loads.
# The script is preloading the value. The advantage here is that the editor
# can offer autocompletion since it must be a static path.
const BuildingScn = preload("res://building.tscn")

# 1. The script preloads the value, so it will load as a dependency
#    of the 'my_buildings.gd' script file. But, because this is a
#    property rather than a constant, the object won't copy the preloaded
#    PackedScene resource into the property until the script instantiates
#    with .new().
#
# 2. The preloaded value is inaccessible from the Script object alone. As
#    such, preloading the value here actually does not benefit anyone.
#
# 3. Because the user exports the value, if this script stored on
#    a node in a scene file, the scene instantiation code will overwrite the
#    preloaded initial value anyway (wasting it). It's usually better to
#    provide null, empty, or otherwise invalid default values for exports.
#
# 4. It is when one instantiates this script on its own with .new() that
#    one will load "office.tscn" rather than the exported value.
@export var a_building : PackedScene = preload("office.tscn")

# Uh oh! This results in an error!
# One must assign constant values to constants. Because `load` performs a
# runtime lookup by its very nature, one cannot use it to initialize a
# constant.
const OfficeScn = load("res://office.tscn")

# Successfully loads and only when one instantiates the script! Yay!
var office_scn = load("res://office.tscn")

Preloading allows the script to handle all the loading the moment one loads the script. Preloading is useful, but there are also times when one doesn't wish for it. To distinguish these situations, there are a few things one can consider:

1. If one cannot determine when the script might load, then preloading a resource, especially a scene or script, could result in further loads one does not expect. This could lead to unintentional, variable-length load times on top of the original script's load operations.

2. If something else could replace the value (like a scene's exported initialization), then preloading the value has no meaning. This point isn't a significant factor if one intends to always create the script on its own.

3. If one wishes only to 'import' another class resource (script or scene), then using a preloaded constant is often the best course of action. However, in exceptional cases, one may wish not to do this:

   1. If the 'imported' class is liable to change, then it should be a property instead, initialized either using an @export or a load() (and perhaps not even initialized until later).

   2. If the script requires a great many dependencies, and one does not wish to consume so much memory, then one may wish to, load and unload various dependencies at runtime as circumstances change. If one preloads resources into constants, then the only way to unload these resources would be to unload the entire script. If they are instead loaded properties, then one can set them to null and remove all references to the resource entirely (which, as a RefCounted-extending type, will cause the resources to delete themselves from memory).

Large levels: static vs. dynamic

If one is creating a large level, which circumstances are most appropriate? Should they create the level as one static space? Or should they load the level in pieces and shift the world's content as needed?

Well, the simple answer is, "when the performance requires it." The dilemma associated with the two options is one of the age-old programming choices: does one optimize memory over speed, or vice versa?

The naive answer is to use a static level that loads everything at once. But, depending on the project, this could consume a large amount of memory. Wasting users' RAM leads to programs running slow or outright crashing from everything else the computer tries to do at the same time.

No matter what, one should break larger scenes into smaller ones (to aid in reusability of assets). Developers can then design a node that manages the creation/loading and deletion/unloading of resources and nodes in real-time. Games with large and varied environments or procedurally generated elements often implement these strategies to avoid wasting memory.

On the flip side, coding a dynamic system is more complex, i.e. uses more programmed logic, which results in opportunities for errors and bugs. If one isn't careful, they can develop a system that bloats the technical debt of the application.

As such, the best options would be...

1. To use a static level for smaller games.

2. If one has the time/resources on a medium/large game, create a library or plugin that can code the management of nodes and resources. If refined over time, so as to improve usability and stability, then it could evolve into a reliable tool across projects.

3. Code the dynamic logic for a medium/large game because one has the coding skills, but not the time or resources to refine the code (game's gotta get done). Could potentially refactor later to outsource the code into a plugin.

For an example of the various ways one can swap scenes around at runtime, please see the Change scenes manually documentation.