"""SSE streaming chat — respostas token-a-token como assistente."""

from __future__ import annotations

import json
from collections.abc import Iterator
from typing import Any

from app import auth
from app.agents import llm_client, store
from app.agents import messages as agent_messages
from app.agents.chat_context_builder import build_ops_context, build_suggested_actions


def _sse(payload: dict[str, Any]) -> str:
    return f"data: {json.dumps(payload, ensure_ascii=False)}\n\n"


def _thread_history_messages(conn, thread_id: int, limit: int = 16) -> list[dict]:
    msgs = agent_messages.thread_messages(conn, thread_id)[-limit:]
    out: list[dict] = []
    for m in msgs:
        if m["from_type"] == "human":
            out.append({"role": "user", "content": m["body"]})
        elif m["from_type"] == "agent":
            out.append({"role": "assistant", "content": m["body"]})
    return out


def chat_stream_generator(
    *,
    username: str,
    user_role: str,
    question: str,
    target_agent: str,
    agent_name: str,
    agent_role: str,
    include_findings: bool,
    thread_id: int | None = None,
) -> Iterator[str]:
    conn = auth.db()
    try:
        kb = store.search_kb(conn, question)
        kb_snippets = [k["snippet"] for k in kb]
        ops_context = build_ops_context(
            conn,
            question,
            target_agent,
            include_findings=include_findings,
            kb_snippets=kb_snippets,
        )
        suggested_actions = build_suggested_actions(conn, question, target_agent)

        tid = thread_id
        history_msgs: list[dict] = []
        if tid:
            row = conn.execute("SELECT * FROM agent_threads WHERE id=?", (tid,)).fetchone()
            if not row:
                yield _sse({"type": "error", "message": "thread not found"})
                return
            history_msgs = _thread_history_messages(conn, tid)
        else:
            tid = agent_messages.create_thread(
                conn,
                subject=f"Chat: {question[:60]}",
                primary_agent=target_agent,
                severity="info",
            )

        agent_messages.post_message(
            conn,
            thread_id=tid,
            from_type="human",
            from_id=username,
            to_type="agent",
            to_id=target_agent,
            body=question,
        )
        conn.commit()

        messages = llm_client.build_chat_messages(
            question=question,
            kb_snippets=kb_snippets,
            ops_context=ops_context,
            user_role=user_role,
            target_agent=target_agent,
            agent_name=agent_name,
            agent_role=agent_role,
            history_messages=history_msgs,
        )

        yield _sse({"type": "start", "thread_id": tid, "agent": target_agent, "actions": suggested_actions})

        token_iter, model_label = llm_client.stream_llm_chat(messages)
        parts: list[str] = []
        for tok in token_iter:
            parts.append(tok)
            yield _sse({"type": "token", "text": tok})

        answer = "".join(parts).strip()
        if not answer:
            answer = (
                "Não consegui gerar resposta agora. Verifique Groq/Ollama no health do Agentic Ops."
            )
            model_label = "error"
            yield _sse({"type": "token", "text": answer})

        agent_messages.post_message(
            conn,
            thread_id=tid,
            from_type="agent",
            from_id=target_agent,
            to_type="human",
            to_id=username,
            body=answer,
            context={"model": model_label, "kb_hits": len(kb), "streamed": True, "actions": suggested_actions},
        )
        store.log_event(
            conn,
            event_type="chat.stream" if thread_id else "chat.query",
            message=question[:120],
            agent_id=target_agent,
            payload={"user": username, "model": model_label, "thread_id": tid},
        )
        conn.commit()
        yield _sse({
            "type": "done",
            "answer": answer,
            "model": model_label,
            "thread_id": tid,
            "kb_hits": len(kb),
            "actions": suggested_actions,
        })
    except Exception as exc:
        yield _sse({"type": "error", "message": str(exc)[:300]})
    finally:
        conn.close()