Ideas for further investigation, specification and implementation (WIP)

[Bia10]

This is summary of some talk I had with @Kaioru on discord.
Putting it up here so its up for further specification and discussion.

Edelstein

Cleanup the entire solution using modern C# syntax features where possible

Primary ctors, object initializers, switch expressions, pattern matching expressions

This must be done manually no batching without breaking possible.

Performance and memory optimizations

Investigate where modern features of .NET could improve performance or reduce allocation.

Better use of Span, ReadOnlySpan, ISpanFormattable

Try to find places where stack allocation would suffice for object type conversion.
Find operations with byte buffers which could be done on stack only.

Investigate use of DotNext

DotNext offers some very nice set of API wich may be of use for us.

Investigate possibility of running Edelstein channels on RAFT cluster.
Investigate possibility of using Atomics API for field object id stuff.

Plugins

• Investigate possibility of Hot-Reload
• Investigate possibility of registering external plugins
• Investigate possibility of source gening the plugins
• Investigate possibility of runtime patching/injecting

SnapshotPlugin

Ideally this could be external plugin however that presumes some progress on the before mentioned points :{
Creates interface for writing/reading snapshots which are stored in structured CSV.
A marvelous work Sep is utilized to parse/read the actual csv.

SnapshotReaderWriter - somewhat done
SnapshotContext - done
Stats - done
Skills - works but there's possibility of mismatch between application of skillsSnapshot
Inventory - only handles equipment so far
CharacterBuild - works with exceptions mentioned before

AdminPanel

This largely depends on development of SukiUI as many needed controls were already mentioned in possible future development of SukiUI.

Setup

Setup wizard

We don't need any complicated workflow system, a Cake.Frosting async pipeline should be entirely sufficient for this.

Well use Octokit to interface with github.

Autoinstall from source

Download repo
Unzip to install dir
Run dotnet tools
Download data/scripts
Unzip into server root dir

Autoinstall from archive

Download binary
Unzip to install dir
Download data/scripts
Unzip into server root dir

Server

Launch/Monitoring

Launch via process and its info, window hidden, output redirected, use args if any.
Observe the process properties
Filter output into separate logs

Configuration

Start file watcher.
Load configuration into property grid.
Use ReactiveValidation on the input -> figure out bounds for server settings.
On change write file back to storage.

AI

Interfaces:
OpenAI - OpenAI-DotNet
Anthropic - Claudia

Generation

Since snapshots have predefined structure it should be easy to use AI LLM to generate valid snapshots which could be initialized.
This leads to interesting possibility of making a Text -> Server states convertor.

Role play

AI is prompt engineered to play a game which us where context of being NPC is given and uses reduced protocol for utilizing the NPC and receiving model updates.

RL/RLHF

Learn from prediction of input to attain basic goals/rewards.

Reward function:

Positive:

• gain exp, level, stats, skills
• explore map, new screen
• obtain, finish quests
• trade items for mesos

Negative:

• lose exp, level, stats, skills
• no new visual scene for some time

Data

Scripts

Load scripts in AvaloniaEdit control to edit them and save.
Reflect the exposed LuaApi as autoCompletion choices.

WZ/NX game archives

Pack/Unpack.
Read/Write.
Parse/Display.

Database

Manage and monitor PostgreSQL installation/service.
View the db tables.
Populate db with data from GUI.
Write and execute SQL queries.