import { withUser, parseBody } from "@/lib/api";
import { LIMITS } from "@/lib/rate-limit";
import { chatRequestSchema } from "@/lib/chat/schemas";
import { prepareTurn, runTurn, toPublicMessage } from "@/lib/chat/service";
import { sseResponse } from "@/lib/chat/sse";

export const dynamic = "force-dynamic";
export const maxDuration = 900;

/**
 * POST /api/chat — streams a model turn as Server-Sent Events.
 * Events: meta → (text-delta | reasoning-delta | tool-* | citation | server-tool)* → done | error
 *
 * `meta` carries `ephemeral` (temporary chat: conversationId is the "ephemeral" sentinel and nothing
 * is persisted except the usage record) and `requestId` (shown in the error details sheet).
 */
export const POST = withUser(
  async (ctx) => {
    const input = await parseBody(ctx.req, chatRequestSchema, 2_500_000);
    const turn = await prepareTurn({ userId: ctx.user.id, requestId: ctx.requestId, ip: ctx.ip }, input);
    return sseResponse(
      async (emit, signal) => {
        // Abort provider work when the client disconnects.
        const onAbort = () => controller.abort();
        const controller = new AbortController();
        signal.addEventListener("abort", onAbort, { once: true });
        ctx.req.signal?.addEventListener("abort", onAbort, { once: true });
        emit({
          type: "meta",
          conversationId: turn.conversation.id,
          isNewConversation: turn.isNewConversation,
          userMessage: turn.userMessage ? toPublicMessage(turn.userMessage) : null,
          assistantMessageId: turn.assistantMessage.id,
          modelKey: turn.model.key,
          clientId: input.clientId,
          ephemeral: turn.ephemeral,
          requestId: ctx.requestId,
        });
        const outcome = await runTurn({ userId: ctx.user.id, requestId: ctx.requestId, ip: ctx.ip }, turn, emit, controller.signal);
        if (outcome.error && !outcome.message.content) emit({ type: "error", error: outcome.error });
        emit({ type: "done", message: toPublicMessage(outcome.message), usage: outcome.usage, costUsd: outcome.costUsd, latencyMs: outcome.latencyMs, ttftMs: outcome.ttftMs, status: outcome.status, title: outcome.title, error: outcome.error });
      },
      { requestId: ctx.requestId },
    );
  },
  { limit: { ...LIMITS.chat, key: "chat" } },
);