Godot: Add a GDScript Trait system.

[willnationsdev]

(Edit:
To minimize further XY problem issues:
The problem addressed here is that Godot's Node-Scene system / scripting languages don't yet support creating re-usable, grouped implementations that are 1) specific to the root node's features and 2) that can be swapped around and/or combined. Scripts with static methods or sub-nodes with scripts could be used for the latter bit, and for many cases this works. However, Godot generally prefers you to keep the logic for your scene's overall behavior stored in the root node while it uses data computed by the child nodes or delegates significantly-deviated sub-tasks to them, e.g. a KinematicBody2D doesn't manage animation so it delegates that to an AnimationPlayer.

Having a thinner root node that uses "component" child nodes to drive its behavior is a weak system in comparison. Having a largely empty root node that just delegates all of its behavior to child nodes runs counter to this paradigm. The child nodes become behavioral extensions of the root node rather than self-sufficent objects that accomplish a task in their own right. It's very clunky, and the design could be simplified/improved by allowing us to consolidate all of the logic in the root node, but also allow us to break up the logic into different, composable chunks.

I suppose the topic of this Issue is more about how to deal with the above problem than it is specifically about GDScript, yet I believe GDScript Traits would be the simplest and most straightforward approach to solving the issue.
)

For the uninformed, traits are essentially a way of blending two classes together into one (pretty much a copy/paste mechanism), only, rather than literally copy/pasting the text of the files, all you are doing is using a keyword statement to link the two files. (Edit: the trick is that while a script can only inherit one class, it can include multiple traits)

I'm imagining something where any GDScript file can be used as a trait for another GDScript file, so long as the traited type extends a class inherited by the merged-into script, i.e. a Sprite-extending GDScript can't use a Resource GDScript as a trait, but it can use a Node2D GDScript. I would imagine a syntax similar to this:

# move_right_trait.gd
extends Node2D
class_name MoveRightTrait # not necessary, but just for clarity
func move_right():
    position.x += 1

# my_sprite.gd
extends Sprite
is MoveRightTrait # maybe add a 'use' or 'trait' keyword for this instead?
is "res://move_right_trait.gd" # alternative if class_name isn't used
func _physics_process():
    move_right() # MoveRightTrait's content has been merged into this script
    if MoveRightTrait in self:
        print("I have a MoveRightTrait")

I can see two ways of doing this:

1. Pre-parse the script via RegEx for "^trait \reload process). We'd have to not support trait nesting or continuously re-examine the generated source code after every iteration to see if more trait insertions have been made.
2. Parse the script normally, but teach the parser to recognize the keyword, load the referred to script, parse THAT script, and then append its ClassNode's content to the generated ClassNode of the current script (effectively taking the parsed results of one script and adding it to the parsed results of the other script). This would automatically support nesting of traited types.

On the other hand, people might want the trait GDScript to have a name but might NOT want that GDScript's class_name to show up in the CreateDialog (cause it's not meant to be created on its own). In this case, it may actually NOT be a good idea to make any script support it; only those that are specially marked (perhaps by writing 'trait' at the top of the file?). Anyway, stuff to think about.

Thoughts?

Edit: After some pondering, I believe option 2 would be much better since 1) we'd know WHICH script a script segment came from (for better error reporting) and 2) we'd be able to identify errors as they happen since the included scripts must be parsed in sequence, rather than just parsing everything at the end. This would cut down on the processing time it adds to the parsing process.

[MrJustreborn]

What's the benefit instead of: extends "res://move_right_trait.gd"

[willnationsdev]

@MrJustreborn Because you can have multiple traits in a class, but you can only inherit one script.

[aaronfranke]

If I understand correctly, this is basically what C# calls "interfaces", but with non-abstract methods? It might be better to call the feature interfaces instead of traits to be familiar to programmers.

[willnationsdev]

@aaronfranke Traits, basically the same thing as Mixins, have a completely different use-case from interfaces precisely because they include implementations of the methods. If an interface gave a default implementation, then it wouldn't really be an interface anymore.

Traits/Mixins are present in PHP, Ruby, D, Rust, Haxe, Scala, and many other languages (as detailed in the linked Wikis), so they should already be widely familiar with people who have a broad repertoire of programming language familiarity.

If we were to implement interfaces (which I'm not opposed to either, especially with optional static typing coming), it would effectively just be a way of specifying function signatures and then requiring that the relevant GDScript scripts implement those function signatures, with traits included (if those existed by that point).

[avencherus]

Maybe a keyword like includes?

extends Node2D
includes TraitClass

Though other names like trait, mixin, has, etc. are surely fine too.

I do like the idea too of having some option to exclude class_name type from the add menu. It can get very cluttered with small types that don't function on their own as nodes.

It may even be just a feature topic of its own.

[LikeLakers2]

(Accidentally deleted my comment, woops! Also, obligatory "why don't you just allow multiple scripts, unity does it")

How will this work in VisualScript, if at all?

Also, could it be beneficial to include an inspector interface for traits, if traits were implemented? I imagine some use cases for traits may include uses cases where there are only traits and no script (at least, no script besides one that includes the trait files). Though, upon thinking about it more, I wonder if the effort put into making such an interface might even be worth it, compared to just making a script that includes the trait files.

[willnationsdev]

@LikeLakers2

How will this work in VisualScript, if at all?

If it is done the way I suggested, it wouldn't happen for VisualScript at all. GDScript only. Any trait system implemented for VisualScript would be designed completely differently because VisualScript isn't a parsed language. Doesn't preclude the possibility at all though (would just need to be implemented differently). Plus, maybe we should consider getting VisualScript inheritance-support first? lol

Also, could it be beneficial to include an inspector interface for traits, if traits were implemented?

There wouldn't be much of a point. The traits simply impart details onto the GDScript, handing it the properties, constants, signals, and methods defined by the trait.

I imagine some use cases for traits may include uses cases where there are only traits and no script (at least, no script besides one that includes the trait files).

Traits, as they are represented in other languages, are never usable in isolation, but must be included in another script in order to be usable.

I wonder if the effort put into making such an interface might even be worth it

Creating an Inspector interface in some way wouldn't really make much sense for GDScript alone. Adding or removing a trait would involve directly editing the source code of the Script resource's source_code property, i.e. it isn't a property on the Script itself. Therefore, either...

1. the editor would have to be taught how to specifically handle the proper editing of the source code for GDScript files to do this (error-prone), or...
2. all Scripts would have to support traits so that the GDScriptLanguage can provide its own internal process for adding and removing traits (but not all languages DO support traits, so the property wouldn't be meaningful in all cases).

[groud]

What is the need for such feature? Is there anything this allows that you cannot do now? Or does it make some tasks significantly faster to deal with?

I'd rather keep GDscript a simple language than adding complex features almost never used.

[willnationsdev]

It solves the Child-Nodes-As-Script-Dependencies problem that this guy had an issue with, but doesn't come with any of the same kind of baggage that MultiScript had because it is constrained to a single language. The GDScript module can isolate logic on how the traits relate to each other and the main script whereas resolving the differences between different languages would be a lot more complicated.

[Zireael07]

With the absence of multiple imports/multiple inheritance, child-nodes-as-script-dependencies are the only way to avoid repeating the code A LOT, and this would definitely solve the problem in a nice way.

[willnationsdev]

@groud @Zireael07 I mean, the more radical, cross-language approach would be to 1) completely redesign Object to use a ScriptStack to coalesce stacked scripts into a single script representation, 2) re-introduce MultiScript and create editor support that automatically converts the addition of scripts into multiscripts (or just outright making all scripts multiscript for simplicity's sake, in which case MultiScript's implementation would essentially be our ScriptStack), or 3) implement a sort of cross-language trait system for the Object type that can merge in Reference-extending scripts as traits, incorporating their content just like a typical script. All of those options are way more invasive to the engine though. This keeps everything simpler.

[fian46]

i don't think trait is necessary. what we need the most is fast engine release cycle. i mean we need to make the engine more flexible to add new feature as simple just add new dll or so files and the engine auto integrating with it self, like plugins style in most IDE. for example i really desperate need for websocket to work, i dont need to wait it until 3.1 to release. 3.1 too broken right now with so many bugs. it will be great if we have this feature. new class can auto inject to GDScript from random .dll or .so in some path. i don't know how much effort this to do in c++ but i hope this is not too hard 😁

[willnationsdev]

@fian46 Well, if someone had implemented websockets as a downloadable GDNative plugin, then yeah, what you described would be the workflow. Instead, they opted to make it an integrated feature available in the vanilla engine. There's nothing stopping people from creating features that way, so your point is really unrelated to the topic of this Issue.

[fian46]

oops i dont know GDNative is exist 😂😂😂. Trait is awesome but is it easier to make fake trait class and instance it and later call the function like basic script ?

[DriNeo]

If a Godot script is an unnamed class, why not instancing "move_right_trait.gd" in "my_sprite.gd" ?
Sorry for my ignorance if I don't understand the issue.

[Eoin-ONeill-Yokai]

I understand the use of traits in more strongly typed languages such as Rust or (interfaces in) C++, but in a ducked typed language isn't it a bit unnecessary? Simply implementing the same functions should allow you to achieve a uniform interface among your types. I guess I'm a bit unsure what the exact problem is with the way GDScript handles interfaces or how a trait system would even really help.

Couldn't you also use preload("Some-other-behavior.gd") and store the results in a variable to achieve basically the same effect?

[willnationsdev]

@fian46 @DriNeo Well, yes and no. Loading scripts and using script classes take care of that already, but the problem extends beyond that.

@TheYokai

Implementing the same functions should allow you to achieve a uniform interface among your types

The problem is not achieving a uniform interface which you're right, duck-typing solves just fine, but rather organizing (combining/swapping) groups of related implementations efficiently.

---

In 3.1, with script classes, you can define static functions on a Reference script (or any type really) and then use that script as a namespace to globally access those functions in GDScript.

extends Reference
class_name Game
static func print_text(p_text):
    print(p_text)
# can even add inner classes for sub-namespaces

extends Node
func _ready():
    Game.print_text("Hello World!")

However, when it comes to non-static content, especially stuff that uses node-specific functions, it's troublesome to divide out one's logic.

For example, what if I have a KinematicBody2D and I want to have a "Jump" behavior and a "Run" behavior? Each of these behaviors would need access to the input handling and move_and_slide features of the KinematicBody2D. Ideally, I would be able to swap out each behavior's implementation independently and keep all of the code for each behavior in separate scripts.

Currently, all the workflows I'm familiar with for this simply aren't optimal.

1. If you keep all of the implementations in the same script and just swap out which functions are being used...
   
   
   
   1. changing "behaviors" may involve swapping out several functions as a set, so you can't group the implementation changes effectively.
   
   
   2. All of the functions for every behavior (X * Y) are sitting in your single script, so it can get bloated very quickly.
   
   
2. You can just replace the entire script, but then that means you have to create a new script for every combination of behaviors plus any logic that uses those behaviors.
3. If you use child nodes as script dependencies, then that means you'd have these weird "component" Node2D nodes that grab their parent and call the move_and_slide method FOR it which is kind of unnatural, relatively speaking.
   
   
   
   • You have to make an assumption that your parent is going to implement that method or do logic to check that it has the method. And if you do a check, then you can either fail silently and potentially have a silent bug in your game or you can turn it into a tool script unnecessarily just so you can set a configuration warning on the node to visually tell you in the editor that there's a problem.
   
   
   • You also don't get proper code completion for the intended operations of the nodes since they derive from Node2D and the whole point is to drive the behavior of a KinematicBody2D parent.
   
   

Now, I'll admit that option 3 is the most effective workflow currently, and that its problems can largely be resolved by using the handy-dandy static typing for GDScript in 3.1. However, there is a more fundamental issue at play.

Godot's node-scene system generally has the form of users creating nodes or scenes that do a particular job, in their own enclosed system. You might instance those nodes/scenes within another scene and have them compute data that is then used by the parent scene (such is the case with the Area2D and CollisionShape2D relationship).

However, the vanilla engine's usage and general best practices recommendation is to keep the behavior of your scene locked to the root node and/or its script. Hardly ever do you have "behavior component" nodes that actually tell the root what to do (and when it is there, it feels very clunky). The AnimationPlayer/Tween nodes are the only arguable exceptions I can think of, but even their operations are directed by the root (it's effectively delegating the control to them temporarily). (Edit: Even in this case, animating and tweening aren't the job of the KinematicBody2D, so it makes sense that those tasks would be delegated. Movement, however, like running and jumping is its responsibility) It's simpler and more natural to allow a trait implementation to organize the code as it keeps the relationships between the nodes strictly data-up/behavior-down and keeps code more isolated into its own script files.

[OvermindDL1]

Eh, marking yourself as 'implementing an interface/trait' also should fulfill an * is * test though, which is convenient to test functionality of something.

[willnationsdev]

@OvermindDL1 I mean, I gave an example of doing a test like that, but I used in instead since I wanted to distinguish between inheritance and trait-usage.

[willnationsdev]

I guess I kinda walked into an XY problem here, my bad. I had just come from 2 other issues (#23052, #15996) that addressed this topic in one way or another and figured I'd submit a proposal, but I didn't really give all the context.

[QbieShay]

@groud this solution will solve one of the problems raised against #19486.

@willnationsdev great idea, I am looking forward to it!

[AfterRebelion]

From my limited understanting, the thing that this Trait system wants to accomplish is enable something similar to the workflow shown in this video: https://www.youtube.com/watch?v=raQ3iHhE_Kk
(Take into account, i'm talking about the _workflow_ shown, not the feature used)

In the video, it is compared with other kinds of workflows, with their advantages and disadvantages.

At least to my knowledge, this kind of workflow is impossible in GDScript right now, because of how inheritance works.

[willnationsdev]

@AfterRebelion The first few minutes of that video where he isolates the modularity, editability, and debugability of the codebase (and the related details of those attributes) is indeed the pursuit of having this feature.

At least to my knowledge, this kind of workflow is impossible in GDScript right now, because of how inheritance works.

This bit isn't quite true, because Godot actually does this very well in regards to node hierarchies and designing scenes. Scenes are modular by nature, properties can be exported (and even animated) directly from the editor without ever dealing with code, and everything can be tested/debugged in isolation because scenes can executed independently.

The difficulty is when the logic that would typically be outsourced to a child node must be executed by the root node because the logic relies on the root node's inherited features. In these cases, the only way to use composition is to have the children start telling the parent what to do rather than having them mind their own business while the parent uses them.

This isn't a problem in Unity because GameObject doesn't have any real inheritance that users can take advantage of. In Unreal it might be a bit(?) of a problem since they have similar node/component-based internal hierarchies for Actors.

[willnationsdev]

Okay, let's play Devil's Advocate here for a bit (@MysteryGM, you might get a kick out of this). Spent some time thinking about how I might write such a system in Unreal and that's giving me a new perspective on it. Sorry for the folks who were thinking this would be a good idea / were excited about it:

Introducing a trait system adds a layer of complexity to GDScript as a language that may make it more difficult to maintain.

On top of that, traits as a feature make it more difficult to isolate where variables, constants, signals, methods, and even subclasses may actually be coming from. If your Node's script suddenly has 12 different traits, then you wouldn't necessarily know where everything is coming from. If you see a reference to something, then you'd have to look through 12 different files to even know where a thing is located in the codebase.

This actually then tanks the debuggability of GDScript as a language since any given problem may require you to pick apart on average 2 or 3 different locations in the code base. If any of those locations are hard to find cause they say they are located in one script but are actually located somewhere else - and if the code's readability doesn't clearly state which thing is responsible for the data/logic - then those 2 or 3 steps are being multiplied into an arbitrarily large, highly stressful set of steps.

The increasing size and scope of a project magnifies these negative effects even further and make the use of traits a pretty untenable quality.

---

But what can be done to resolve the problem? We don't want child "component logic" nodes telling scene roots what to do, but we also can't rely on inheritance or changing whole scripts to solve our problem.

Well, what would any non-ECS engine do in this situation? Composition is still the answer, but in this case a full Node is unreasonable when taken to scale / complicates the dynamic of the ownership hierarchy. Instead, one can just define non-Node implementation objects that abstract away the concrete implementation of a behavior, but which are all still owned by the root node. This can be done with a Reference script.

# root.gd
extends KinematicBody2D

export(Script) var jump_impl_script = null setget set_jump_impl_script
var jump_impl
func set_jump_impl_script(p_script):
    jump_impl = p_script.new() if p_script else null

export(Script) var move_impl_script = null setget set_move_impl_script
var move_impl
func set_move_impl_script(p_script):
    move_impl = p_script.new() if p_script else null

func _physics_process():
    # use logic involving these...
    move_impl.move(...)
    jump_impl.jump(...)

If the exports worked in such a way that we could edit them in the Inspector as enums for classes that derive a particular type like we can for new Resource instances, then that'd be cool. The only way to do that now would be to fix Resource script exports and then make your implementation script extend Resource (for no other reason). Although, having them extend Resource would be a good idea if the implementation itself requires parameters that you'd like to be able to define from the Inspector. :-)

Now, what would make THIS easier would be having a snippet system or macro system so that creating those re-used declarative code sections is easier for developers.

Anyway, yeah, I think I kinda identified glaring issues with a trait system and a better approach to solving the problem. Hooray for XY problem Issues! /s

Edit:

So, the above example's workflow would involve setting the implementation script and then using the script's instance at runtime to define the behavior. But what if the implementation itself requires parameters that you'd like to set statically from the Inspector? Here's a version based on a Resource instead.

# root.gd
extends KinematicBody2D

# if you use a Resource script AND had a way of specifying that the assigned Resource 
# must extend that script, then the editor would automatically assign an instance of 
# that resource script to the var. No separate instancing or setter necessary.

export(Resource) var jump_impl = null # set jump duration, max height, tween easing via Inspector
export(Resource) var move_impl = null # similarly customize movement from Inspector

# can then create different Resources as different implementations. Because they are resources,
# one can edit them even outside of a scene!
func _physics_process():
    move_impl.move(...)
    jump_impl.jump(...)

[willnationsdev]

Related: #22660

[willnationsdev]

@AfterRebelion

Take into account, i'm talking about the workflow shown, not the feature used

It's ironic that, after you clarify this and I agree with the optimal workflow, and then disagree with later portions of the comment, I then follow it up by basically saying how the "feature used" in that video is actually the ideal way to tackle this problem in Godot anyway. Haha.

[AfterRebelion]

# root.gd
extends KinematicBody2D

export(Script) var jump_impl_script = null setget set_jump_impl_script
var jump_impl
func set_jump_impl_script(p_script):
jump_impl = p_script.new() if p_script else null
...

Wow, didn't know export was so powerful. Expected it to only be able to interact with primitives, and other data structures.

This makes my earlier comment invalid.
As you said, if some kind of macro is implemented to make its implementation easier, it would be the best way to implement that workflow without the need for MultiScript. Maybe not as versatile as Unity, because you would still need to declare every possible script beforehand, but still a good option.

[willnationsdev]

@AfterRebelion

As you said, if some kind of macro is implemented to make its implementation easier, it would be the best way to implement that workflow without the need for MultiScript.

Well, the Resource-based approach I mentioned in the same comment, combined with some better editor support from #22660, would make the quality comparable to what Unity can do.

Maybe not as versatile as Unity, because you would still need to declare every possible script beforehand, but still a good option.

Well, if they fix the array type hints in 3.2, then you can define an exported array for file paths that must extend a script and effectively add your own. This could even be done via a plugin in 3.1 by using the EditorInspectorPlugin class to add custom content to the Inspector for certain resources or nodes.

I mean, if you wanted a "Unity"-like system, then you really WOULD have sub-nodes that tell the root what to do and you would only need to fetch them by referring to their name, without manually declaring them or adding them from the root node's script. The Resource method is generally a lot more effective and maintains a cleaner codebase.

[willnationsdev]

Because the trait system would put an inordinate strain on the usability of GDScript code, based on the reasons I outlined, I'm going to go ahead and close this issue. The pitfalls of its addition far outweigh any comparatively meager benefits we might receive from such a system, and those same benefits can be implemented in different ways with more clarity and usability.

[vnen]

Well, I missed this discussion while I was out. Haven't read everything yet, but I had the idea of adding traits to GDScript to solve the code reuse problem, which I find way more elegant and clear than the faux multiple inheritance of attaching a script to a derived type. Though what I would do was making specific trait files, not allowing any class file to be a trait. I don't think it would be too bad.

But I'm open to suggestions to solve the main issue, which is reusing code in multiple node types.

[willnationsdev]

@vnen the solution I came up with which is the last item is to outsource re-usable sections to Resource scripts.

• They can still be exposed and their properties edited in the Inspector, just like if they were member variables on a node.

• They preserve a clear trail of where data and logic is coming from, something that could easily be compromised if many traits are included into one script.

• They don't add any undue complexity to GDScript as a language. For example, if they existed, we'd have to resolve how shared qualities (properties, constants, signals) would be merged into the main script (or, if not merged, force users to deal with dependency conflicts).

• Resource scripts are better since they can be assigned and changed from the Inspector. Designers, writers, etc. will be able to change implementation objects directly from the editor.

[vnen]

@willnationsdev I see (though the name "resource scripts" sounds odd, since all scripts are resources). The main issue with this solution is that it doesn't solve something that people expect with the inheritance approach: adding exported variables and signals to the root node of the scene (in particular when instancing the scene elsewhere). You can still edit the exported variables from the subresources, but it becomes less practical (you can't tell at a glance which properties you can edit).

The other issue is that it requires boilerplate code to be repeated a lot. You also have to make sure the resource scripts are actually set before calling the functions.

The advantage is that we don't need to anything, it's already available. I guess this should be documented and people using the current "attach script to derived node" method could comment on the solution to see how viable it is.

[willnationsdev]

@vnen

but it becomes less practical (you can't tell at a glance which properties you can edit).

Can you elaborate on what you mean here? I don't see how there could be any substantial lack of clarity as to which properties are accessible, especially if something like #22660 were merged.

1. I instance the scene, want to know how I can edit it, and so look at that scene's root node's script.
2. Inside the script, I see...
   
   
   
   • export(MoveImpl) var move_impl = FourWayMoveImpl.new()
   
   
   • use FourWayMoveTrait
   
   
3. Assuming we have a means of click-tracing the identifier (which should really be a 3.1.1 feature!) to open the script, we then end up opening the associated script and can view its properties.

Seems like the same number of steps to me, unless I'm missing something.

Furthermore, which properties are editable is actually even more clear with resources I would say since, if an implementation has properties specific to it, then the fact that the data is related to the implementation is more clear if you have to preface accessing it with the implementation instance, i.e. move_impl.<property>.

The other issue is that it requires boilerplate code to be repeated a lot. You also have to make sure the resource scripts are actually set before calling the functions.

This is true, however, I still think the benefits of it being an export with an initialization outweigh the cost of the added verbage:

("High-Level Teammate", HLT, like designers, writers, artists, etc.)

• One can assign the values directly from the Inspector, rather than having to open up the script, locate the correct line to change, and then change it (already mentioned, but it leads into...).

• One can specify that the exported content has a base type requirement. The Inspector can then provide an enumerated list of allowable implementations automatically. HLTs can then safely assign only derivations of that type. This helps insulate them from the alternative of needing to know the ramifications of all the different trait scripts floating around. We would also have to modify the autocompletion in GDScript to support looking up named and unnamed trait files in response to seeing the use keyword.

• One can serialize a configuration of an implementation as a *.tres file. HLTs can then drag-and-drop them from the FileSystem dock or even create their own right in the Inspector. If one wished to do the same with traits, they'd have to create a derived trait that provides a custom constructor to override the default one. Then they'd use that trait instead as a "preconfiguration" via an imperatively coded constructor.

  1. Weaker because its imperative rather than declarative.
  2. Weaker because the constructor must be explicitly defined in script.
  3. If the trait is unnamed, then the user will need to know where the trait is located in order to properly use it instead of the default, base trait. If the trait is named, then it clogs up the global namespace unnecessarily.
  4. If they change the script to say use FourWayMoveTrait instead of use MoveTrait, there is no longer any lasting indication that the script is even compatible with the base MoveTrait. It enables HLT confusion as to whether the FourWayMoveTrait can even change to a different MoveTrait without breaking things.
  5. If an HLT were creating a new trait implementation in this way, they wouldn't necessarily know all of the properties that can be/need to be set from the base trait. This is a non-Issue with resources created in the Inspector.

• One can even have multiple Resources of the same type (if there were a reason for that). A trait would not support this, but would instead trigger parse conflicts.

• One can change configurations and/or their individual values without ever leaving the 2D/3D Viewport. This is much more comfortable for HLTs (I know plenty who outright get annoyed when they have to look at code at all).

With all that said, I agree that the boilerplate is annoying. To deal with that though, I'd rather add a macro or snippet system to simplify it. A snippet system would be good idea since it could conceivably support any language that can be edited in the ScriptEditor, rather than just GDScript alone.

[vnen]

About:

but it becomes less practical (you can't tell at a glance which properties you can edit).

I was talking about the inspector, so I mean the "HLT" here. People who won't look into the code. With traits we could add new exported properties to the script. With resources scripts you can only export variables to the resource itself, so they won't show in the inspector unless you edit the subresource.

I understand your argument but it goes beyond the original problem: avoid repeating code (without inheritance). Traits are for programmers. In many cases there is only one implementation you want to reuse, and you don't want to expose that in the inspector. Of course you can still use the resource scripts without exporting just by assigning directly in the code, but still doesn't solve the problem of reusing the exported variables and signals from the common implementation, which is one of the main reasons people are trying to use inheritance.

That is, people are currently trying to use inheritance from a generic script not only for functions, but also for exported properties (which are usually related to those functions) and signals (which then can be connected with the UI).

This is the hard problem to solve. There are some ways to do it with GDScript alone, but again they require boilerplate code to be copied around.

[avencherus]

I can only imagine the back and forth that would have to take place to get external scripts to behave as if they were written directly in the main script itself.

Would be a nice thing to have if heaven and earth don't have to be moved to achieve it. X)

[willnationsdev]

@vnen I see what you're saying now. Well, since it seems like there may yet be some life in this Issue, I'll reopen it when I next get a chance.

[AfterRebelion]

Any news on reopening this? Now that GodotCon 2019 is anounced, and the Godot Sprint is a thing, maybe this is worth talking there.

[willnationsdev]

@AfterRebelion I had just forgotten to come back and re-open it. Thanks for reminding me. XD

[xDGameStudios]

@willnationsdev I liked what I read regarding the EditorInspectorPlugin! quick question, this means I can create a custom inspector for a data type... for example... add a button to the inspector.
(been quite some time since I wanted to do this, to have a way to trigger events for debugging purposes with a button in the inspector I could make the button press execute some script)

[willnationsdev]

@xDGameStudios Yup, that and a lot more is all possible. Any custom Control can be added to the Inspector, either at the top, at the bottom, above a property, or below a category.

[xDGameStudios]

@willnationsdev I don't know if I may contact you by private message!! But I would like to know more about the EditorInspectorPlugin (like some sample code).. something in the lines of a custom resource type (for example) MyResource that has a property export "name" and an inspector button that prints the "name" variable if I press it (in editor or during debugging)! Documentation is lacking big time in this matter... I would write the documentation myself if I knew how to use this! :D thanks

[avencherus]

I would also like to know more about it. X)

[asheraryam]

So is this the same as an autoload script with subclasses containing static functions?

e.g. your case would become Traits.MoveRightTrait.move_right()

or even simpler, having different autoload scripts for different trait classes:

Movement.move_right()

[willnationsdev]

No, traits are a language-specific feature equivalent to copy-pasting the source code of one script into another script (kinda like a file-merge). So if I have a trait with move_right(), and I then declare that my second script uses that trait, then it too can use move_right(), even if it isn't static and even if it accesses properties elsewhere in the class. It would just result in a parse error if the property didn't exist in the second script.

[Zireael07]

I have discovered that I either have duplicate code (smaller functions, e.g. make an arc) or superfluous nodes because I can't have multiple inheritance.

[razcore-art]

This would be awesome, I find myself having to make a script with the exact same functionality just be used on different node types cause different extends so basically just for one line of code. By the way if anyone knows how to make that with the current system let me know.

If I have functionality for a script with extends Node, is there a way to attach the same behavior to another node type without having to duplicate the source file and replacing with appropriate extend?

[Zireael07]

Any progress on that? I keep having to duplicate code or having to add nodes, as I said before. I know it won't be done in 3.1, but maybe aim for 3.2?

[willnationsdev]

Oh, I haven't been working on this at all. In fact, I'm further along on my extension method implementation than on this. I need to talk with vnen about both of them though since I'd like to work out the best syntax/implementation details with him.

Edit: if someone else wants to take a crack at implementing traits, they are welcome to jump into it. Just need to coordinate with the devs.

[Zireael07]

"extension method implementation"?

[willnationsdev]

@Zireael07 #15586
It allows people to write scripts which can add new "built in" functions for engine classes. My interpretation of the syntax would be something like this:

static Array func sum(p_self: Array):
    if not len(p_self):
        return 0
    var value = p_self[0]
    for i in range(1, len(p_self)):
        value += p_self[i]
    return value

Then, somewhere else, I'd be able to just do:

var arr = [1, 2, 3]
print(arr.sum()) # prints 6

It would secretly make a call to the constantly-loaded extension script and call into the static function of the same name that is bound to the Array class, passing in the instance as the first parameter.

[jabcross]

I've had some talks with @vnen and @jahd2602 about this. One thing that comes to mind is Jai's solution to polymorphism: importing a property's namespace.

This way, you could do something like the following:

class A:
    var a_prop: String = "Hello"
    func foo():
        print("A's a_prop: ", a_prop)
    func bar():
        print("A's bar()")

class B:
    using var a: A = A.new()
    var a_prop: String = "World" # Overriding A's a_prop

    func bar():  # Overriding A's bar()
        print("B's bar()")

func main():
    var b: B = B.new()
    b.foo() # output: "A's a_prop: World"
    b.bar() # output: "B's bar()"

The point being is that the using keyword imports the namespace of a property, such that b.foo() is really only syntactic sugar for b.a.foo().

And then, making sure that b is A == true and that B can be used in typed situations that accept A as well.

This also has the benefit that things don't have to be declared as traits, this would work for anything that doesn't have common quality names.

[jabcross]

One problem is that this doesn't mesh well with the current inheritance system. If both A and B are Node2D and we make a function in A: func baz(): print(self.position), which position will be printed when we call b.baz()?
One solution could be having the caller determine self. Calling b.foo() would call foo() with b as self and b.a.foo() would call foo() with a as self.

If we had freestanding methods such as Python (where x.f(y) is sugar for f(x,y)), this could be really easy to implement.

[jabcross]

Another, unrelated idea:

Focus only on freestanding functions, JavaScript style.

If we adopt the x.f(y) == f(x,y) convention for static functions, we could very easily have the following:

class Jumper:
    static func jump(_self: KinematicBody2D):
        # jump implementation

class Runner:
    static func run(_self: KinematicBody2D, direction: Vector2):
        # run implementation

class Character:
    extends KinematicBody2D
    func run = Runner.run       # Example syntax
    func jump = Jumper.jump

func main():
    var character = Character.new()
    character.jump()
    character.run(Vector2(1,0))

This would have minimal impact on the class system, since it really only affects methods. If we really wanted this to be flexible, though, we could go full JavaScript and just allow function definitions to be assignable, callable values.

[avencherus]

@jabcross Sounds good, I do like the concept of some sort of optional namespacing, and the function idea is interesting.

Regarding the namespace one, I wonder why not simply using A, the other declarative things seem extraneous.

Curious too how it would need to be resolved with the multiple inheritance. I suppose option A is force both scripts to inherit the same type, so they're both just extended on top of the same class, without any special merging.

Option B, maybe additional GDScript keywords to specify a trait class, and what class you'd like to have hints from. It would be the same idea, but just only extra steps to appear more explicit.

At the top of A.gd:

extends Trait as Node2D
is Trait as Node2D
is Trait extends B
extends B as Trait

[willnationsdev]

Ohhh, I really like the concept of namespace importing. It solves not only the trait issue, but also potentially the "extension methods" concept for adding content to engine types.

class_name ArrayExt
static func sum(_self: Array) -> int:
    var sum: int = 0
    for a_value in _self:
        sum += a_value
    return sum

using ArrayExt
func _ready():
    var a = [1, 2, 3]
    print(a.sum())

@jabcross If we then also added lambas and/or allowed objects to implement a call operator (and had a callable type for compatible values), then we could start adding a more functional-oriented approach to GDScript code (which I think would be a great idea). Granted, walking more into @vnen's #18698 territory at that point, but...

[vnen]

We have to consider that GDScript is still a dynamic language and some of those suggestions require a runtime check to correctly dispatch the calls, adding a performance penalty (even to people who doesn't use the features, since it needs to check all function calls and maybe property lookup as well). Which is also why I'm not sure if adding extensions is a good idea (especially since they are essentially syntax sugar).

I prefer the pure trait system, where traits are not classes but a thing of their own. This way they can be fully resolved in compile time, and provide error messages in cases of name conflicts. I believe this would solve the issue without additional runtime overhead.

[willnationsdev]

@vnen Ahhh, didn't realize that about the runtime cost. And if that applies to any extension method implementation, then I guess that wouldn't be ideal either.

If we did a pure trait system then, were you thinking just making trait TraitName in place of extends paired with a using TraitName under extends in other scripts? And would you be implementing this yourself, or would it be delegated?

[vnen]

If we did a pure trait system then, were you thinking just making trait TraitName in place of extends paired with a using TraitName under extends in other scripts?

That's my idea. I believe it's simple enough and cover pretty much all use cases for code reuse. I would even allow the class using the trait to override the trait methods (if it's possible to do on compile-time). Traits could also extend other traits.

And would you be implementing this yourself, or would it be delegated?

I wouldn't mind to give the task to someone else who's up to it. I'm short on time anyway. But we should agree on design beforehand. I'm quite flexible with the details but it should 1) not incur in runtime checks (believe that GDScript lends itself well to stuff that's impossible to figure out on compilation), 2) be relatively simple, and 3) not add too much to compilation time.

[avencherus]

@vnen Liking these ideas. Was wondering how you imagine a trait would be able to do things like auto-completion for the classes that would include it, or would that not be possible?

[vnen]

Was wondering how you imagine a trait would be able to do things like auto-completion for the classes that would include it, or would that not be possible?

A trait would be essentially an "import" in my view. It should be trivial to show completion, assuming that completion for members work.

[willnationsdev]

@vnen I would imagine that you would essentially parse into a ClassNode with a trait flag set on it. And then if you do a using statement, it would attempt merge all of the properties/methods/signals/constants/subclasses into the current script.

1. If a method collides, then the current script's implementation would override the base method, just as if it were overriding an inherited method.
   
   
   
   • But what to do if a base class already has the "merged" method?
   
   
2. If a property, signal, and constant overlaps, check if it's the same type / signal signature. If there is no mismatch, then just consider it a "merging" of the property/signal/constant. Otherwise, notify the user of a type/signature conflict (a parse error).

Bad idea? Or should we just outright forbid non-method conflicts? What about subclasses? I have a feeling we should make those be conflicts.

[jahd2602]

@willnationsdev Sound like the "diamond problem" (aka "deadly diamond of death"), a well documented ambiguity with different solutions already applied in various popular programming languages.

Which reminds me:
@vnen are traits going to be able to extend other traits?

[willnationsdev]

@jahd2602 He already suggested that as a possibility

Traits could also extend other traits.

[willnationsdev]

@jahd2602 Based on the Perl/Python solutions, it seems like they basically form a "stack" of layers containing the content for each class so that conflicts from the last-used trait rest on top of, and overwrite, the other versions. That sounds like a pretty good solution for this scenario. Unless you or @vnen have alternative thoughts. Thanks for the linked overview of solutions jahd.

[willnationsdev]

A few questions.

First: In what ways should we support the using statement?

I'm thinking that the using statement should require some constant-value GDScript.

using preload("res://my_trait.gd") # a preloaded expression
using ScriptClass.MyTrait # a const resource
using Autoload.MyTrait # a const resource
using MyTrait # a regular script class

I'm thinking all of the above.

Second: What should we say the allowed syntax for defining a trait and/or its name should be?

Someone may not necessarily want to use a script class for their trait, so I don't think we should enforce a trait TraitName requirement. I'm thinking that trait must be on a line.

So, if it's a tool script, it of course should have tool at the top. Then, if it's a trait, it must define whether it is a trait on the following line. Optionally allow someone to state the script class name following the trait declaration on the same line and if they do, don't allow them to also use class_name. If they omit the trait name, then class_name <name> is fine. Then, when extending another type, we could insert the extends after the trait declaration and/or on a line of its own after the trait declaration. So, I would consider each of these valid:

# Global name from trait keyword.
trait MyTrait extends BaseTrait

# Global name from class_name keyword, but is still a trait and also happens to be a tool script.
tool
trait
extends BaseTrait
class_name MyTrait

# A trait with no global name associated with it. Does not extend anything.
trait

Third: should we, for autocompletion purposes and/or declaration of intent/requirements, allow a trait to define a base type that it must extend?

We've already discussed that traits should support inheriting other traits. But should we allow a TraitA to extend Node, allow the TraitA script to get Node autocompletion, but then also trigger a parse error if we do a using TraitA statement when the current class doesn't extend Node or any of its derived types?

Fourth: Rather than having traits extend other traits, can't we instead simply keep the extends statement reserved for class extensions, allow a trait to not need this statement at all, but rather than extending a base trait, allow traits to simply have their own using statements which sub-import those traits?

# base_trait.gd
trait
func my_method():
    print("Hello")

# derived_trait.gd
trait
using preload("base_trait.gd")
func my_method():
   print("World") # overrides previous method, will only print "World".

Of course, the benefit here would be that you'd be able to batch multiple traits under a single trait name by using multiple using statements, similar to C++ include files that include several other classes.

Fifth: if we have a trait, and it has a using or an extends for a method, and then implements its own, what do we do when it calls, within that function .<method_name> to execute the base implementation? Do we assume that these calls always execute within the class inheritance context, and the trait hierarchy has no influence here?

cc @vnen

[vnen]

First: In what ways should we support the using statement?

I'm thinking that the using statement should require some constant-value GDScript.

using preload("res://my_trait.gd") # a preloaded expression
using ScriptClass.MyTrait # a const resource
using Autoload.MyTrait # a const resource
using MyTrait # a regular script class

I'm okay with all of those. But for path I use a string directly: using "res://my_trait.gd"

Second: What should we say the allowed syntax for defining a trait and/or its name should be?

Someone may not necessarily want to use a script class for their trait, so I don't think we should enforce a trait TraitName requirement. I'm thinking that trait must be on a line.

So, if it's a tool script, it of course should have tool at the top. Then, if it's a trait, it must define whether it is a trait on the following line. Optionally allow someone to state the script class name following the trait declaration on the same line and if they do, don't allow them to also use class_name. If they omit the trait name, then class_name <name> is fine. Then, when extending another type, we could insert the extends after the trait declaration and/or on a line of its own after the trait declaration. So, I would consider each of these valid:

# Global name from trait keyword.
trait MyTrait extends BaseTrait

# Global name from class_name keyword, but is still a trait and also happens to be a tool script.
tool
trait
extends BaseTrait
class_name MyTrait

# A trait with no global name associated with it. Does not extend anything.
trait

tool on a trait should not make any difference, since they are not executed directly.

I agree that a trait not necessarily has a global name. I would use trait in a similar way to tool. It must be the first thing in the script file (except for comments). The keyword should optionally be followed by the trait name. I wouldn't use class_name for them, since they are not classes.

Third: should we, for autocompletion purposes and/or declaration of intent/requirements, allow a trait to define a base type that it must extend?

I honestly don't like adding features in the language for the sake of the editor. That's where annotations would come in handy.

Now if we want to make traits only applied to a certain type (and its derivatives), then it's okay. In fact, I think this is better for the sake of the static checks: it allows a trait to use stuff from a class while the compile can actually check if they are being used with the correct types and whatnot.

Fourth: Rather than having traits extend other traits, can't we instead simply keep the extends statement reserved for class extensions, allow a trait to not need this statement at all, but rather than extending a base trait, allow traits to simply have their own using statements which sub-import _those_ traits?

Well, that's mostly a matter of semantics. When I mentioned that traits can extend another, I didn't really mean to use the extends keyword. The main difference is that with extends you can only extend one, with using you can embed many other traits into one. I'm okay with using, as long as there are no cycles, it's not a problem.

Fifth: if we have a trait, and it has a using or an extends for a method, and then implements its own, what do we do when it calls, within that function .<method_name> to execute the base implementation? Do we assume that these calls always execute within the class inheritance context, and the trait hierarchy has no influence here?

That's a tricky question. I would assume that a trait has no business dealing with class inheritance. So the dot notation should call the method on the parent trait, if there's one, otherwise throw an error. A trait should not be aware of the classes they are in.

OTOH, a trait is almost like an "include", so it would be applied verbatim to the class, therefore calling the parent implementation. But honestly I'd simply forbid the dot notation if the method isn't found in the parent trait.

[jabcross]

What about a trait that requires the class to have one or more other traits? For instance, a trait DoubleJumper that requires both trait Jumper, a trait Upgradable and a class that inherits KinematicBody2D.

Rust, for instance, allows you to use type signatures like those. Something like KinematicBody2D: Jumper, Upgradable. But since we're using : to annotate type, we could just use KinematicBody2D & Jumper & Upgradable or something.

[jabcross]

There's also the polymorphism issue. What if the implementation of the trait is different for each class, but it exposes the same interface?

For instance, we want a method kill() in the trait Jumper, which is used by both Enemy and Player. We want different implementations for each case, while still keeping both compatible with the same Jumper type signature. How to do this?

[willnationsdev]

For the polymorphism, you would just create a separate trait that includes the trait with kill() and then implements its own specific version of the method. Using traits that override previously included traits' methods is how you would handle it.

[willnationsdev]

Also, I don't think there were any plans (yet) to make a type hint have trait requirements. Is that something we would want to do?

[jabcross]

create a separate trait

Wouldn't that generate a bunch of one-off trait files? If we could do nested trait declarations (similar to the class keyword), that could be more convenient. We could also just override the methods directly in the class that's using the trait.

I'd really appreciate a strong type signature system (maybe with boolean composition and optionals/non-nulls). Traits would fit right in.

[mnn]

I am no sure if it was discussed, but I think it should be possible to invoke trait specific version of a function. For example:

trait A
func m():
  print("A")

trait B
func m():
  print("B")

class C
using A
using B

func c():
  A.m()
  B.m()
  m()

which prints: A, B, B.

---

Also I am not entirely sure about the "no runtime cost". How would be handled dynamically loaded scripts (not available during export) with classes which are using traits which were defined before export? Am I misunderstanding something? Or that case is not considered "runtime"?

[vnen]

I am no sure if it was discussed, but I think it should be possible to invoke trait specific version of a function.

I was already considering this, but I'm not sure about allowing a class to use conflicting traits (that is, traits that define method with the same name). The order of using statements should make no difference.

Also I am not entirely sure about the "no runtime cost". How would be handled dynamically loaded scripts (not available during export) with classes which are using traits which were defined before export? Am I misunderstanding something? Or that case is not considered "runtime"?

It's not about exporting. It will definitely impact loading time, since compilation happens at loading (though I don't think it'll be very significant), but it shouldn't impact when the script is being executed. Ideally scripts should be compiled on export, but that's another discussion.

[busy8]

Hello everyone.

I am new to Godot and have been getting used to it over the past few days. As I tried to figure out the best practices to use to make reusable components I had decided on a pattern. I would always make the root Node of a sub-scene, that is intended to be instanced in another scene, export all the properties that I intend to be set from the outside. As much as possible, I wanted to make knowledge of the internal structure of the instanced branch unnecessary to the rest of the scene.

To make this work the root node has to "export" properties and then copy the values to the appropriate child in _ready. So, for example, imagine a Bomb node with a child Timer. The root Bomb node in the sub-scene would export "detonation_time" and then it would do $Timer.wait_time = detonation_time in _ready. This allows us to set it nicely in Godot's UI whenever we instance it without having to make children editable and drill down to the Timer.

However
1) It's a very mechanical transformation so it seems like something similar could be supported by the system
2) It probably adds a slight inefficiency over setting the appropriate value directly in the child node.

Before I go on, this might seem tangential to what is being discussed because it doesn't involve allowing a sort of "private" inheritance (in C++ parlance). However, I actually like Godot's system of building behaviors by composing Scene elements instead of more inheritance-like engineering. Those "written-in" relationships are unchanging and static. OTOH, the scene structure is dynamic, you can even change it at run time. Game logic is so very subject to change during development that I think Godot's design fits the use case very well.

It is true that child nodes are used as behavioral extensions of root nodes but that does not cause them to lack self-sufficiency, IMO. A timer is perfectly self-contained and predictable in behavior regardless of what it is used to time. Whether you use a spoon to drink soup or eat ice cream, it suitably performs its function even though it's acting as an extension of your hand. I view root nodes as maestros that coordinate the behaviors of child nodes so they don't have to know directly about each other and are THEREFORE able to remain self-contained. Parent/root nodes are desk-bound managers that delegate responsibilities but don't do much direct work. Since they're thin, it's easy to create a new one for slightly different behavior.

However, I think root nodes should also act as the primary INTERFACE to the functionality of the entire instanced branch. All properties that can be tweaked in the instance should be "settable" in the root node of the branch even if the ultimate owner of the property is some child node. Unless I'm missing something, this has to be manually arranged in the current version of Godot. It would be nice if this could be automated somehow to combine the benefits of a dynamic system with easier scripting.

One thing I'm thinking about is a system of "dynamic inheritance", if you will, available to subclasses of Node. There would be two sources of properties/methods in such a script, those of the script it extends and those "bubbled up" from children within the scene structure. So my example with the Bomb would become something like export lifted var $Timer.wait_time [= value?] as detonation_time within the bomb.gd script's member variables section. The system would essentially generate $Timer.wait_time = detonation_time in the _ready callback and generate the getter/setter that will allow $Bomb.detonation_time = 5 from the parent of Bomb node to result in $Timer.wait_time = 5 being set.

In the OP's example with MoveRightTrait we'd have the node to which mysprite.gd is attached have MoveRightTrait as a child node. Then in mysprite.gd we'd have something like lifted func $MoveRightTrait.move_right [as move_right] (perhaps 'as' could be optional when the name will be the same). Now calling move_right on a script object created from mysprite.gd would automatically delegate to the appropriate child node. Perhaps signals could be bubbled so they can be attached to a child node from the root? Perhaps whole nodes could be bubbled with just lifted $MoveRightTrait [as MvR] without func, signal or var. In this case all methods and properties on MoveRightTrait would be accessible from mysprite directly as mysprite.move_right or through mysprite.MvR.move_right if the 'as MvR' is used.

That's one idea of how to simplify creation of an INTERFACE to a scene structure in the root of an instanced branch, increasing their "black box" characteristic and getting scripting convenience along with the power of Godot's dynamic scene system. Of course, there would be many side details to consider. For example, unlike base classes child nodes can be removed at run time. How should the bubbled/lifted functions and properties behave if called/accessed in that error case? If a node with the right NodePath is added back do the lifted properties start working again ? [YES, IMO] Also it would be an error to use 'lifted' in classes not derived from Node since there would never be children to bubble/lift from in that case. In addition, name clashes are possible with duplicated "as {name}" or "lifted $Timer1 lifted $Timer2" where nodes have properties/methods with the same name. The script interpreter would ideally detect such logical problems.

I feel like this would give us a lot of what we want even though it's really just syntactic sugar that saves us from having to write forwarding functions and initializations. Also, because it's fundamentally simple conceptually, it shouldn't be that hard to implement or explain.

Anyway, if you got this far thanks for reading!

[busy8]

I used "lifted" everywhere but that's just illustrative.
Something like using var $Timer.wait_time as detonation_time or using $Timer is obviously just as good. In any case you get to conveniently pseudo-inherit from child nodes, creating a coherent single point of access to the desired functionality in the root of the branch to be instanced. The requirement on the reusable pieces of functionality is that they extend Node or a subclass of it so that they can be added as children to the larger component.

Another way of looking at it is that the "extends" keyword on a script that inherits from a Node gives you your "is-a" relationship while using the "using" or "lifted" keyword on a script to "bubble up" a descendant node's members gives you something akin to "implements" [hey, possible keyword] that exists in languages with single inheritance but multiple "interfaces" (eg. Java). In unrestricted multiple inheritance (like c++) base classes form a [static, written-in] tree. By analogy I'm kind of proposing layering convenient syntax and boilerplate elimination over Godot's existing Node trees.

If it is ever determined that this is something worth exploring, there are aspects of the design space to consider:
1) Should we allow only immediate children in a "using". IOW using $Timer but not using $Bomb/Timer'? This would be simpler but would force us to write boilerplate in some cases. I say that a full NodePath ROOTED in the Node to which the script is attached should be legal [but NO references to parents/siblings allowed]. 2) Should there be an option thatfind_node'sthe "using"-ed node instead of following a written in NodePath? For exampleusing "Timer"with a string for the pattern would be slower but the forwarding architecture would continue to work if a referenced node's position in the sub-tree changes at run time. This could be used selectively for child nodes that we expect to move around beneath the root. Of course syntax would have to be worked out especially when using a particular member (eg.using var "Timer".wait_time as detonation_timeis icky). 3) Should there be a way query for certain functionality [equivalent to asking if an interface is implemented or a child node is present]? Perhaps "using" entire nodes with aliases should allow testing the alias to be a query. Sousing MoveRightTrait as DirectionalMoverin a script would result innode.DirectionalMoverreturning the child MoveRightTrait. This is logical becausenode.DirectionalMover.move_right()calls the method on the child MoveRightTrait. Other nodes without that statement would return null. So the statementif node.DirectionalMover:` would become a test for the functionality by convention.
4) The State Pattern should be implementable by replacing a "using"-ed node with another that has variant behavior but the same interface [duck typing] and same NodePath as referenced in the "using" statement. With the way the scene tree works this would work out almost for free. However, the system would have to track signals connected through a parent and restore connections in the replaced child.

[mnn]

I have been working with GDScript for some time now and I have to agree, some kind of trait/mixin and proxy/delegation feature is direly needed. It is quite annoying having to setup all this boilerplate just to connect properties or expose methods of children in the root of the scene.

Or adding levels of the tree only to simulate components (it gets quite cumbersome rather quickly, because then you break all node paths with each new component). Maybe there is a better way, something like meta/multi script allowing multiple scripts on one node? If you have idiomatic solution, please share...

Throwing C++ (GDNative) into mix makes things even worse, because _ready and _init behave differently there (read: initialization with default values half-works or doesn't work at all).

[lmerriam]

This is the main thing I have to work around in GDScript. I often need to share functionality across nodes without structuring my entire inheritance structure around it -- for example, my player and shopkeepers have an inventory, my player+items+enemies have stats, my player and enemies have equipped items, etc.

Currently I implement these shared 'components' as classes or nodes loaded into the 'entities' that require them, but it's messy (adds lots of searches for nodes, makes duck-typing nearly impossible, etc) and alternative approaches have their own drawbacks so I haven't found a better way. Traits/mixins would absolutely save my life.

What it comes down to is being able to share code across objects without using inheritance, which I think is both necessary and not possible to do cleanly in Godot right now.

[raymoo]

The way I understand Rust traits (https://doc.rust-lang.org/1.8.0/book/traits.html), is that they are like Haskell typeclasses, where you require some parameterized functions to be defined for the type you are adding a trait to, and then you get to use some generic functions defined over any types that implement a trait. Are Rust traits something different from what's proposed here?

[AfterRebelion]

https://github.com/godotengine/godot-proposals/

[Zireael07]

This one will probably be migrated wholesale, as it's had extensive discussion here.

[Xrayez]

This one will probably be migrated wholesale, as it's had extensive discussion here.

_I find the "moving" of proposals pointless imo, they are better closed and asked to re-open in godot-proposals if people express interest, and let other proposals actually implemented if needed. Anyways..._

I stumbled upon this issue a year ago, but only now I begin to understand the potential usefulness of the trait system.

Sharing my current workflow in hopes to inspire someone to understand the problem better (and perhaps suggest a better alternative other than implementing traits system).

1. Create a tool to generate component templates for each of used node type in the project:

@willnationsdev https://github.com/godotengine/godot/issues/23101#issuecomment-431468744

Now, what would make THIS easier would be having a snippet system or macro system so that creating those re-used declarative code sections is easier for developers.

Walking in your footsteps... 😅

tool
extends EditorScript

const TYPES = [
    'Node',
    'Node2D',
]
const TYPES_PATH = 'types'
const TYPE_BASENAME_TEMPLATE = 'component_%s.gd'

const TEMPLATE = \
"""class_name Component{TYPE} extends {TYPE}

signal host_assigned(node)

export(bool) var enabled = true

export(NodePath) var host_path
var host

func _ready():
    ComponentCommon.init(self, host_path)"""

func _run():
    _update_scripts()


func _update_scripts():

    var base_dir = get_script().resource_path.get_base_dir()
    var dest = base_dir.plus_file(TYPES_PATH)

    for type in TYPES:
        var filename = TYPE_BASENAME_TEMPLATE % [type.to_lower()]
        var code = TEMPLATE.format({"TYPE" : type})
        var path = dest.plus_file(filename)

        print_debug("Writing component code for: " + path)

        var file = File.new()
        file.open(path, File.WRITE)
        file.store_line(code)
        file.close()

2. Create a static method to be reused for initializing components to host (root for instance):

class_name ComponentCommon

static func init(p_component, p_host_path = NodePath()):

    assert(p_component is Node)

    # Try to assign
    if not p_host_path.is_empty():
        p_component.host = p_component.get_node(p_host_path)

    elif is_instance_valid(p_component.owner):
        p_component.host = p_component.owner

    elif is_instance_valid(p_component.get_parent()):
        p_component.host = p_component.get_parent()

    # Check
    if not is_instance_valid(p_component.host):
        push_warning(p_component.name.capitalize() + ": couldn't find a host, disabling.")
        p_component.enabled = false
    else:
        p_component.emit_signal('host_assigned')

This is how a component (trait) looks like once generated with the first script:

class_name ComponentNode2D extends Node2D

signal host_assigned(node)

export(bool) var enabled = true

export(NodePath) var host_path
var host

func _ready():
    ComponentCommon.init(self, host_path)

(Optional) 3. Extend the component (trait)

@vnen https://github.com/godotengine/godot/issues/23101#issuecomment-471816901

That's my idea. I believe it's simple enough and cover pretty much all use cases for code reuse. I would even allow the class using the trait to override the trait methods (if it's possible to do on compile-time). Traits could also extend other traits.

Walking in your footsteps... 😅

class_name ComponentMotion2D extends ComponentNode2D

const MAX_SPEED = 100.0

var linear_velocity = Vector2()
var collision

export(Script) var impl
...

Actually, the exported Scripts are used in these components to drive the behavior of specific host/root node types per component. Here, ComponentMotion2D will have mainly two scripts:

• motion_kinematic_body_2d.gd
• motion_rigid_body_2d.gd

So the children still drive the host/root behavior here. The host terminology comes from me using state machines, and here's where traits wouldn't be perfect fit perhaps, because the states are better organized as nodes imo.

The components themselves are "hardwired" into the root by making them onready members, effectively decreasing the boilerplate code (with the expense of actually having to reference them as object.motion)

extends KinematicBody2D

onready var motion = $motion # ComponentMotion2D

[TheColorRed]

Not sure if this would help solve the problem, but C# has a thing called extension methods that extend the functionality of a class type.

Basically the function must be static, and the first parameter is required and must be self. It would look like this as a definition:

extension.gd

# any script that uses this method must be an instance of `Node2D`
static func distance(self source: Node2D, target: Node2D):
    return source.global_position.distance_to(target.global_position)

# any script that uses this method must be an instance of `Rigidbody2D`
# a `Sprite` instance cannot use this method
static func distance(self source: Rigidbody2D, target: Node2D):
    return source.global_position.distance_to(target.global_position)

Then when you want to use the distance method, you then just do this:

player.gd

func _ready() -> void:
    print(self.distance($Enemy))
    print($BulletPoint.distance($Enemy))

[willnationsdev]

I am familiar with it, but that does not help solve the problem. Hehe, thanks though.

[vnen]

@TheColorRed extension methods were already proposed, but I don't think they are feasible in a dynamic language. I also don't think they solve the root problem that initially started this discussion.

---

On another note, I'll probably open many of the proposals for GDScript as GIPs (this included, if @willnationsdev don't mind).

I still believe that traits makes the most sense to share code horizontally in an OO language (without multiple inheritance, but I don't want to go that way).

[mnn]

I don't think they are feasible in a dynamic language

Is GDS dynamic though? Extension methods could be limited to only typed instances and would work exactly same as in other languages - just a syntactic sugar during compilation replacing method call with static method (function) call. Honestly, I would prefer pimps (ext. methods) before prototypes ala JS or other dynamic ways of attaching methods to classes or even just instances.

[jabcross]

Whatever we decide to do, I hope we don't decide to name it "pimps".

[vnen]

Is GDS dynamic though?

There are plenty of definitions of "dynamic" in this context. To be clear: the variable type may not be known at compilation time, so the method extension check needs to be done at runtime (which will hurt performance one way or another).

Extension methods could be limited to only typed instances

If we start doing this we might as well make GDScript typed only. But that's a whole other discussion that I don't to get here.

Point is: things shouldn't start or stop working because the user added types to a script. It's almost always confusing when it happens.

Again, I don't think it solves the issue anyway. We are trying to attach the same code to multiple types, while an extension method will add it only to one type.

Honestly, I would prefer pimps (ext. methods) before prototypes ala JS or other dynamic ways of attaching methods to classes or even just instances.

Nobody proposed (yet) attaching methods dynamically (at runtime) and I don't want that either. Traits would be applied statically at compilation time.

[jabcross]

I originally made a comment about Haxe and its mixin macro library, but then I realized most users won't use a third-party language anyway.

[supagu]

I've recently run into a need for this.

I have some objects the user can interact with but cannot share the same parent, but they need similar groups of API's

for example I have some classes that cant inherit from the same parent, but use a similar set of API's:
Warehouse: Finances, Deletion, MouseInteraction + others
Vehicle: Finances, Deletion, MouseInteraction + others
VehicleTerminal: Finances, Deletion, MouseInteraction + others

For the Finances I've used composition, as this requires the least boiler plate code as get_finances_component() is a sufficient API as it doesn't really care about the game object's at all.

The others:
MouseInteraction and Delection I have just had to copy and paste as it needs to know about the game objects, some composition doesn't work here unless I did some weird delegation:

Warehouse:
  func delete():
      get_delete_component().delete(self);

but that doesn't really allow me to override how delete works where if it was an inherited class I could have the ability to re-write some of the deletion code if needed.

[Xrayez]

MouseInteraction and Delection I have just had to copy and paste as it needs to know about the game objects, some composition doesn't work here unless I did some weird delegation

I access components via onready nodes currently. I'm doing something similar:

# character.gd

var input = $input # input component

func _set(property, value):
    if property == "focused": # override
        input.enabled = value
    return true

So this:

character.input.enabled = true

becomes this:

character.focused = true

[MathiasBaumgartinger]

As @Calinou kindly pointed out my issue https://github.com/godotengine/godot-proposals/issues/758 is closely related to this. What do you think of the proposal to be able to add a trait to a group? This could drastically the need for scripts and other overhead.

[avencherus]

Just would be great to have a way to inject shareable code into classes, and if they have exported values have these appear in the inspector, and have the methods and properties available and detected by code completion.

[clayjohn]

Feature and improvement proposals for the Godot Engine are now being discussed and reviewed in a dedicated Godot Improvement Proposals (GIP) (godotengine/godot-proposals) issue tracker. The GIP tracker has a detailed issue template designed so that proposals include all the relevant information to start a productive discussion and help the community assess the validity of the proposal for the engine.

The main (godotengine/godot) tracker is now solely dedicated to bug reports and Pull Requests, enabling contributors to have a better focus on bug fixing work. Therefore, we are now closing all older feature proposals on the main issue tracker.

If you are interested in this feature proposal, please open a new proposal on the GIP tracker following the given issue template (after checking that it doesn't exist already). Be sure to reference this closed issue if it includes any relevant discussion (which you are also encouraged to summarize in the new proposal). Thanks in advance!

Note: this is a popular proposal, if someone moves it over to Godot Proposals, please try to summarize the discussion as well.