ADR-165: Resolving component numbers for the scripting of the SDK

More details about this document
Latest published version:
https://adr.decentraland.org/adr/ADR-165
Authors:
menduz
Feedback:
GitHub decentraland/adr (pull requests, new issue, open issues)
Edit this documentation:
GitHub View commits View commits on githistory.xyz

Abstract

This RFC describes the non-functional issues to solve about the component numbers of the SDK. It presents synchronization, usability, and ergonomic problems behind the technical requirements for the performance. It extends alternatives as a journey to add context to the decisions. The document is still not conclusive.

Context, Reach & Prioritization

Assuming a deterministic and compiled file (game.js) and a set of well-known ECS components for rendering (Transform, Gltf, Material, MeshRenderer), it is easy to think about synchronizing either two instances of game.js#1 and game.js#2, or broadcasting visual updates back-and-forth with a renderer. In this case, all components between all actors share a stable numeric identifier (component number), allowing ultra-fast synchronization and CRDT semantics, which were hard technical requirements at the design stage.

But there is more to it than just technical requirements. Storing the state in a human-readable way is essential for the product, enabling artists and developers to collaborate and create a polished final experience. A key element of iterating an experience (using either a text or WYSIWYG editor) is the evolution of assets, including the data stored in the entities' components.

As of the moment of writing this essay, components are identified by a user-defined number pointing to a schema. And that works perfectly for the use cases described in the first paragraph and from a technical perspective. But it is very short-sighted for the big non-functional requirements like enabling evolution, collaboration, and reusability/DRY:

And due to the imperative code nature, defining components (and their schemas) makes it difficult to statically analyze the environment to extract information from it to create better tools. It would be highly beneficial to "analyze" a project to assess its components rather than "run" the code to extract its information.

Solution space exploration

The solution space exploration will evaluate three big problems, defining the component number, coding ergonomics, and static reflection.

By defining the component number, the stability and speed of the internal protocols will be the main factor. Among using third-party libraries without component collisions.

With coding ergonomics, how the developers define and use the components will be evaluated.

And lastly, static reflection contemplates the ability of the approach to extract information from the original assets of the project (and its libraries) to enable better creation tools and static representations of the scenes and composites.

Ideal solution

The ideal solution is a nominal collection of components, in which their names are used to identify the schemas.

Characteristics:

Calculating the component number

For the rest of the document, we can assume that the component number will be derived using a hash of the FQN. The caveat is that the image of the function will be limited to int32, and that greatly increases the chances of collisions. This will mostly cover the defining the component number dimension, focusing the domain of the problem to how to define the FQN.

For the sake of simplicity in implementation and distribution of hashes, a regular CRC32 will be used to calculate the component number using a function like this:

// components with fixed numbers, used for the rendering for performance
// reasons
const MAX_RESERVED_COMPONENT_NUMBER = 2048
// NOTE: this function will always return uint32
function computeComponentNumber(fullyQualifiedName: string): number {
  return (unsignedCRC32(fullyQualifiedName) + 2048) & 0xffff_ffff
}

Alternative 1

Using a convention over configuration approach, components are defined in a well-known folder, using a constrained type to prevent imperative code and race conditions.

All the components will be stored as separate files in the root folder "components". The FQN will be derived from the packageName::FileName, and the component name will be only defined by its filename.

Each file exports an export default declareComponent(schema) value to be statically analyzed by simply importing the file with any JS runtime. The tooling also detects other side effects of importing the file to help/prevent the developers from tinting the component files with unrelated code.

The resolution of FQN in the current scene and its dependencies is deterministic by traversing the dependencies, detecting the convention, and then importing each component.

For developers to use the components, the tooling generates a source file materializing them with an added name. The generated code looks like this:

// components.generated.ts
//! THIS FILE IS AUTOGENERATED, DO NOT EDIT
import { engine } from "@dcl/ecs"

// third party components
import RotatorComponentDeclaration from "rotator-library/components/Velocity"

// own components
import VelocityComponentDeclaration from "./components/Velocity"
import AccelerationComponentDeclaration from "./components/Acceleration"

export const Rotator = engine.addComponent(RotatorComponentDeclaration)
export const Velocity = engine.addComponent(VelocityComponentDeclaration)
export const Acceleration = engine.addComponent(AccelerationComponentDeclaration)

This approach has many limitations and problems.

Alternative 2

The idea has evolved from the previous alternative by making all components nominal and referenced by their name/symbol in TypeScript. They are then exported as constants in a single file ./components.ts, which allows the tooling to analyze only one file like before, but with the removal of the entire code generation stage.

// components.ts
export { Velocity } from "./components/Velocity"
export { Acceleration } from "./components/random-file"

Third-party components are imported directly from their libraries in the file that they are going to be used.

import { Rotator } from "rotator-library/components"

In this approach, no automatic convention is available to resolve their number or FQN, and thus the definition of the components themselves would need a manual FQN.

// components/Velocity.ts
export const Velocity = declareComponent("my-scene::Velocity", {
  x: Schemas.Float,
  y: Schemas.Float,
  z: Schemas.Float
})

This approach will also enable the tooling to load and run diverse files to, as statically as possible, know their exposed components and the components used internally to be hidden from the editor or other tools.

A naive component loader would look like this:

export function loadAllComponents(workDir: string) {
  const { dependencies } = readPackageJson(workDir)

  // load dependencies with components
  const componentFilesToLoad = dependencies.filter((dep) => resolveFile(dep, "/components.js"))

  // load current project
  componentFilesToLoad.push(resolveFile(workDir, "/components.js"))

  // reduce results
  const loadedComponents: ComponentDefinition[] = []
  for (const file of loadedComponents) {
    loadedComponents.push(Object.values(require(file)))
  }
  // validate
  validateComponents(loadedComponents)
  return loadedComponents
}

RFC 2119 and RFC 8174

The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "NOT RECOMMENDED", "MAY", and "OPTIONAL" in this document are to be interpreted as described in RFC 2119 and RFC 8174.

License

Copyright and related rights waived via CC0-1.0. Living