An incremental binary state serializer with delta encoding for games.
Although it was born to be used on Colyseus, this library can be used as standalone.
Defining Schema
As Colyseus is written in TypeScript, the schema is defined as type annotations inside the state class. Additional server logic may be added to that class, but client-side generated (not implemented) files will consider only the schema itself.
import { Schema, type, ArraySchema, MapSchema } from '@colyseus/schema';
export class Player extends Schema {
@type("string")
name: string;
@type("number")
x: number;
@type("number")
y: number;
}
export class State extends Schema {
@type('string')
fieldString: string;
@type('number')
fieldNumber: number;
@type(Player)
player: Player;
@type([ Player ])
arrayOfPlayers: ArraySchema<Player>;
@type({ map: Player })
mapOfPlayers: MapSchema<Player>;
}
See example.
Supported types
Primitive Types
Type | Description | Limitation |
---|
string | utf8 strings | maximum byte size of 4294967295 |
number | auto-detects int or float type. (extra byte on output) | 0 to 18446744073709551615 |
boolean | true or false | 0 or 1 |
int8 | signed 8-bit integer | -128 to 127 |
uint8 | unsigned 8-bit integer | 0 to 255 |
int16 | signed 16-bit integer | -32768 to 32767 |
uint16 | unsigned 16-bit integer | 0 to 65535 |
int32 | signed 32-bit integer | -2147483648 to 2147483647 |
uint32 | unsigned 32-bit integer | 0 to 4294967295 |
int64 | signed 64-bit integer | -9223372036854775808 to 9223372036854775807 |
uint64 | unsigned 64-bit integer | 0 to 18446744073709551615 |
float32 | single-precision floating-point number | -3.40282347e+38 to 3.40282347e+38 |
float64 | double-precision floating-point number | -1.7976931348623157e+308 to 1.7976931348623157e+308 |
Declaration:
Primitive types (string
, number
, boolean
, etc)
@type("string")
name: string;
@type("int32")
name: number;
Custom Schema
type
@type(Player)
player: Player;
Array of custom Schema
type
@type([ Player ])
arrayOfPlayers: ArraySchema<Player>;
Array of a primitive type
You can't mix types inside arrays.
@type([ "number" ])
arrayOfNumbers: ArraySchema<number>;
@type([ "string" ])
arrayOfStrings: ArraySchema<string>;
Map of custom Schema
type
@type({ map: Player })
mapOfPlayers: MapSchema<Player>;
Map of a primitive type
You can't mix types inside maps.
@type({ map: "number" })
mapOfNumbers: MapSchema<number>;
@type({ map: "string" })
mapOfStrings: MapSchema<string>;
Backwards/forwards compability
Backwards/fowards compatibility is possible by declaring new fields at the
end of existing structures, and earlier declarations to not be removed, but
be marked @deprecated()
when needed.
This is particularly useful for native-compiled targets, such as C#, C++,
Haxe, etc - where the client-side can potentially not have the most
up-to-date version of the schema definitions.
Reflection
The Schema definitions can encode itself through Reflection
. You can have the
definition implementation in the server-side, and just send the encoded
reflection to the client-side, for example:
import { Schema, type, Reflection } from "@colyseus/schema";
class MyState extends Schema {
@type("string")
currentTurn: string;
}
const encodedStateSchema = Reflection.encode(new MyState());
const myState = Reflection.decode(encodedStateSchema);
Data filters
On the example below, considering we're making a card game, we are filtering the cards to be available only for the owner of the cards, or if the card has been flagged as "revealed"
.
import { Schema, type, filter } from "@colyseus/schema";
export class State extends Schema {
@filterChildren(function(client: any, key: string, value: Card, root: State) {
return (value.ownerId === client.sessionId) || value.revealed;
})
@type({ map: Card })
cards = new MapSchema<Card>();
}
Limitations and best practices
- Each
Schema
structure can hold up to 64
fields. If you need more fields, use nested structures. NaN
or null
numbers are encoded as 0
null
strings are encoded as ""
Infinity
numbers are encoded as Number.MAX_SAFE_INTEGER
- Multi-dimensional arrays are not supported.
- Items inside Arrays and Maps must be all instance of the same type.
@colyseus/schema
encodes only field values in the specified order.
- Both encoder (server) and decoder (client) must have same schema definition.
- The order of the fields must be the same.
- Avoid manipulating indexes of an array. This result in at least
2
extra bytes for each index change. Example: If you have an array of 20 items, and remove the first item (through shift()
) this means 38
extra bytes to be serialized.
Generating client-side schema files (for strictly typed languages)
If you're using JavaScript or LUA, there's no need to bother about this.
Interpreted programming languages are able to re-build the Schema locally through the use of Reflection
.
You can generate the client-side schema files based on the TypeScript schema definitions automatically.
# C#/Unity
schema-codegen ./schemas/State.ts --output ./unity-project/ --csharp
# C/C++
schema-codegen ./schemas/State.ts --output ./cpp-project/ --cpp
# Haxe
schema-codegen ./schemas/State.ts --output ./haxe-project/ --haxe
Benchmarks:
Scenario | @colyseus/schema | msgpack + fossil-delta |
---|
Initial state size (100 entities) | 2671 | 3283 |
Updating x/y of 1 entity after initial state | 9 | 26 |
Updating x/y of 50 entities after initial state | 342 | 684 |
Updating x/y of 100 entities after initial state | 668 | 1529 |
Decoder implementations
Decoders for each target language are located at /decoders/
. They have no third party dependencies.
Why
Initial thoghts/assumptions, for Colyseus:
- little to no bottleneck for detecting state changes.
- have a schema definition on both server and client
- better experience on staticaly-typed languages (C#, C++)
- mutations should be cheap.
Practical Colyseus issues this should solve:
- Avoid decoding large objects that haven't been patched
- Allow to send different patches for each client
- Better developer experience on statically-typed languages
Inspiration:
License
MIT