import os, time, html
import requests
import gradio as gr

# ---------- Config (set in HF Space → Settings → Secrets) ----------
# If you don't set NGROK_URL in Secrets, we'll use your current tunnel as a fallback.
NGROK_URL = os.environ.get("NGROK_URL", "https://bd8de44dca7e.ngrok-free.app/answer").strip()
BACKEND_TOKEN = os.environ.get("BACKEND_TOKEN", "").strip()     # optional: Bearer token
REQUEST_TIMEOUT = float(os.environ.get("REQUEST_TIMEOUT", "30")) # seconds

HEADERS = {"Content-Type": "application/json"}
if BACKEND_TOKEN:
    HEADERS["Authorization"] = f"Bearer {BACKEND_TOKEN}"

def _resolve_backend_url(raw: str) -> str:
    url = (raw or "").strip().rstrip("/")
    if not url:
        return ""
    if not url.endswith("/answer"):
        url = f"{url}/answer"
    return url

def _endpoints():
    """Return (answer_url, command_url) from NGROK_URL, tolerant if it already has /answer."""
    answer_url = _resolve_backend_url(NGROK_URL)
    base = answer_url.rsplit("/", 1)[0] if answer_url.endswith("/answer") else answer_url
    command_url = f"{base}/command"
    return answer_url, command_url

def _normalize_matches(matches):
    out = []
    if isinstance(matches, list):
        for m in matches:
            if isinstance(m, (list, tuple)) and len(m) >= 2:
                q = str(m[0]); d = float(m[1])
            elif isinstance(m, dict):
                q = str(m.get("question") or m.get("q") or m.get("text") or "")
                d = float(m.get("distance", m.get("score", 0.0)))
            else:
                q = str(m); d = 0.0
            out.append((q, d))
    return out

def chat_handler(message, history, threshold):
    answer_url, command_url = _endpoints()
    if not answer_url:
        return "⚠️ يجب ضبط المتغير السري `NGROK_URL`."

    try:
        # Slash command → send to /command
        if message.strip().startswith("/"):
            r = requests.post(command_url, json={"cmd": message}, headers=HEADERS, timeout=REQUEST_TIMEOUT)
            r.raise_for_status()
            data = r.json() if r.content else {}
            return data.get("output", "—")

        # Normal question → send to /answer
        payload = {"question": message, "threshold": threshold}
        t0 = time.time()
        r = requests.post(answer_url, json=payload, headers=HEADERS, timeout=REQUEST_TIMEOUT)
        r.raise_for_status()
        data = r.json() if r.content else {}
        answer = data.get("answer") or data.get("message") or "لا توجد إجابة."
      # matches = _normalize_matches(data.get("matches", []))
        latency = time.time() - t0

     #  lines = "\n".join([f"{i+1}. \"{q}\"  (distance: {d:.4f})" for i, (q, d) in enumerate(matches[:5])])
        # ... after we parse `answer` and compute `latency`
        md = (
            f"**الإجابة**\n\n{answer}\n\n"
            f"<sub>زمن الاستجابة: {latency:.2f}s</sub>"
        )
        return md

      
    except Exception as e:
        return f"تعذر الاتصال بالخادم: {e}"

demo = gr.ChatInterface(
    fn=chat_handler,
    title="نظام الأسئلة والإجابات لجهاز الخدمة المدنية",
    description="اكتب سؤالك أو استخدم أوامر مثل: `/datasets`, `/use <name>`, `/train`, `/threshold 0.55`, `/probs : <سؤال>`",
    additional_inputs=[gr.Slider(minimum=0.0, maximum=2.0, step=0.1, value=0.6, label="عتبة المسافة")],
)

demo.queue(default_concurrency_limit=8, max_size=32)

if __name__ == "__main__":
    demo.launch(server_name="0.0.0.0", server_port=int(os.environ.get("PORT", "7860")))