gott/Services/SseHelper.cs

using System.Diagnostics;
using System.Text.Json;
using System.Threading;
using System.Threading.Tasks;
using Gott.Models;
using Microsoft.AspNetCore.Http;

namespace Gott.Services;

/// <summary>
/// Shared SSE-Streaming-Utilities für AnthropicService und OpenAiService.
/// </summary>
public static class SseHelper
{
    /// <summary>
    /// Schreibt ein SSE-Event (wirft bei CancellationToken).
    /// </summary>
    public static async Task WriteSseEvent(HttpResponse response, string eventType, object data, CancellationToken ct = default)
    {
        ct.ThrowIfCancellationRequested();
        var json = JsonSerializer.Serialize(data);
        await response.WriteAsync($"event: {eventType}\ndata: {json}\n\n", ct);
        await response.Body.FlushAsync(ct);
    }

    /// <summary>
    /// Versucht ein SSE-Event zu schreiben, ignoriert Fehler (Client disconnected).
    /// Optionaler <paramref name="gate"/>: serialisiert die Byte-Writes, wenn mehrere
    /// Producer (z.B. parallele Panel-Member) auf DIESELBE HttpResponse schreiben –
    /// sonst vermischen sich die SSE-Frames auf Byte-Ebene.
    /// </summary>
    public static async Task TryWriteSse(HttpResponse response, string eventType, object data, CancellationToken ct, SemaphoreSlim gate = null)
    {
        if (ct.IsCancellationRequested) return;
        bool acquired = false;
        try
        {
            if (gate != null)
            {
                await gate.WaitAsync(ct);
                acquired = true;
            }
            await WriteSseEvent(response, eventType, data, ct);
        }
        catch
        {
            // Client bereits disconnected / abgebrochen – ignorieren
        }
        finally
        {
            if (acquired) gate.Release();
        }
    }

    /// <summary>
    /// Finalisiert eine Chain in der Datenbank (Token-Counts, Stopgrund, Dauer).
    /// </summary>
    public static async Task FinalizeChain(DatabaseService db, ChainRecord chain, TokenUsage usage, int iterations, string stopReason, Stopwatch sw, string streamErrorInfo = null)
    {
        chain.InputTokens = usage.InputTokens;
        chain.OutputTokens = usage.OutputTokens;
        chain.CacheCreationTokens = usage.CacheCreationTokens;
        chain.CacheReadTokens = usage.CacheReadTokens;
        chain.Iterations = iterations;
        chain.StopReason = stopReason;
        chain.StreamErrorInfo = streamErrorInfo;
        chain.DurationMs = sw.ElapsedMilliseconds;
        await db.UpdateChain(chain);
    }

    /// <summary>
    /// Kürzt Tool-Results auf max. Zeichenzahl.
    /// </summary>
    public const int MaxToolResultChars = 20_000;

    public static string TruncateToolResult(string result)
    {
        return result.Length > MaxToolResultChars
            ? result[..MaxToolResultChars] + $"\n\n[... gekürzt, {result.Length - MaxToolResultChars} weitere Zeichen ...]"
            : result;
    }
}

/// <summary>
/// Token-Zähler für eine Chat-Session (Input, Output, Cache).
/// </summary>
public class TokenUsage
{
    public long InputTokens { get; set; }
    public long OutputTokens { get; set; }
    public long CacheCreationTokens { get; set; }
    public long CacheReadTokens { get; set; }

    /// <summary>
    /// Addiert die Werte von <paramref name="other"/> auf. NICHT thread-safe – der einzige
    /// Aufrufer (Panel-Engine) summiert die Member-Usage erst NACH der Task.WhenAll-Barrier,
    /// also single-threaded. Jeder Member befüllt seine EIGENE TokenUsage parallel; geteilt
    /// wird nichts → kein Lock/Interlocked nötig.
    /// </summary>
    public void Add(TokenUsage other)
    {
        if (other == null) return;
        InputTokens += other.InputTokens;
        OutputTokens += other.OutputTokens;
        CacheCreationTokens += other.CacheCreationTokens;
        CacheReadTokens += other.CacheReadTokens;
    }
}

/// <summary>
/// Ergebnis einer Tool-Ausführung.
/// </summary>
public record ToolResult(string Id, string Result, bool IsError);