Skip to content

PLAN.MD

Latest commit

 

History

History
85 lines (72 loc) · 3.21 KB

PLAN.MD

File metadata and controls

85 lines (72 loc) · 3.21 KB

Plan: Split Long AI Responses

Overview

Split AI responses exceeding MaxAIResponseperSay characters into multiple chunks, sent with delays to prevent overwhelming the client while keeping AI characters patient.

Files to Modify

1. openlife/settings/ServerSettings.hx

Add two new static settings:

public static var MaxAIResponseperSay:Int = 100;
public static var AIWaitTimePer100Chars:Float = 10; // seconds per 100 characters

2. openlife/auto/AiBase.hx

Add a new instance method to set waiting time to a minimum:

public function setWaitingTimeMin(minTime:Float):Void {
    if (this.waitingTime < minTime) {
        this.waitingTime = minTime;
    }
}

Note: Remove IncreaseWaitingTimeForAll() - not needed.

3. openlife/server/AiHandler.hx

Modify respondToPlayerAsync() to pass fromPlayer into the thread so we can access the specific AI. Then in sendResponseInChunks():

  1. Get the specific AI via fromPlayer.connection.serverAi.ai
  2. After receiving full response, check if length > MaxAIResponseperSay and contains split separators (.!?*)
  3. If splitting needed:
    • Find the best split point (separator closest to but not exceeding MaxAIResponseperSay)
    • If no separator found within limit, use first separator after limit
    • Call onSuccess with first chunk immediately
    • Calculate wait time: (totalChars / 100) * AIWaitTimePer100Chars seconds
    • Call aiBase.setWaitingTimeMin(totalWaitTime) on the specific AI
    • Sys.sleep(waitTime) in thread
    • Repeat for remaining chunks (max 10 splits total)
  4. If no splitting needed, call onSuccess(response) immediately as before

4. Revised AiHandler Implementation

var aiBase = fromPlayer.connection.serverAi.ai;
if (aiBase != null) {
    aiBase.setWaitingTimeMin(totalWaitTime);
}

Implementation Details

Splitting Algorithm

1. If response.length <= MaxAIResponseperSay OR no separator exists:
   - Call onSuccess(response) immediately
   
2. Otherwise:
   - Find all separator positions (.!?*)
   - Find separator closest to MaxAIResponseperSay from the left
   - If none found from left within limit, use closest from right (after limit)
   - Split at that position
   
3. For each chunk:
   - Call onSuccess(chunk)
   - If not last chunk:
     - Calculate waitTime = (originalResponse.length / 100) * AIWaitTimePer100Chars
     - Call aiBase.setWaitingTimeMin(waitTime)
     - Sys.sleep(waitTime)

Timing Calculation (per user clarification)

  • Example: 500 character response with AIWaitTimePer100Chars = 10
  • Total wait time = (500 / 100) * 10 = 50 seconds
  • Only the specific AI's waitingTime is set to 50 seconds if currently lower
  • Other AIs are not affected

Edge Cases

  • No separator within limit: use first separator after limit
  • Response with no separators: send as single chunk (no splitting)
  • Max 10 splits: stop after 10 chunks even if response remains
  • fromPlayer.connection.serverAi could be null if AI is not fully initialized: add null check

Verification

  • Compile with haxe server.hxml
  • Test with long AI responses (>100 chars) and verify splitting
  • Test that AI characters wait (not running away) during response chunks
  • Test that short responses (<100 chars) work unchanged