import os
# Hugging Face token uyarısını kapat
os.environ["HF_HUB_DISABLE_IMPLICIT_TOKEN"] = "1"

import gradio as gr
import feedparser
import pandas as pd
import numpy as np
import faiss
import matplotlib.pyplot as plt
import re
import html
import requests
from sentence_transformers import SentenceTransformer
from transformers import AutoTokenizer, AutoModelForSequenceClassification, TextClassificationPipeline, pipeline
import warnings

warnings.filterwarnings('ignore')

# --- Global Değişkenler ---
model = None
sentiment_analyzer = None
translator = None
df = None
index = None
embeddings = None

# --- Modeller ---
CRYPTOBERT_NAME = "ElKulako/cryptobert"
TRANSLATOR_NAME = "Helsinki-NLP/opus-mt-tc-big-en-tr"

# --- Ayarlar ---
NEUTRAL_CONFIDENCE_FLOOR = 0.55

# --- Sinyal Kelimeleri ---
NEGATIVE_CUES = { "crash", "crashes", "dump", "dumps", "plunge", "plunges", "tumble", "tumbles", "falls", "fall", "drop", "drops", "slump", "slumps", "sell-off", "selloff", "panic", "fear", "fears", "concern", "concerns", "pressure", "pressures", "risk", "risks", "risk-off", "lawsuit", "hacked", "hack", "breach", "ban", "banned", "crackdown", "probe", "investigation", "charges", "liquidation", "liquidations", "lag", "lags", "weak", "weaker", "over?", "collapse", "collapses", "recession", "loss", "losses" }
POSITIVE_CUES = { "surge", "surges", "pump", "pumps", "rally", "rallies", "soar", "soars", "breakout", "breaks out", "record", "ath", "all-time high", "wins", "approval", "approved", "etf", "inflows", "adoption", "partnership", "partners", "launch", "launches", "upgrade", "upgrades", "bull", "bullish", "rise", "rises", "beats", "rebound", "rebounds", "gain", "gains", "bullrun" }

# --- Çeviri Haritası (Görünüm İçin) ---
LABEL_MAP = {
    "bullish": "OLUMLU",
    "bearish": "OLUMSUZ",
    "neutral": "NÖTR"
}

# --- TEMİZLEME VE YARDIMCI FONKSİYONLAR ---
def clean_html_tags(text):
    if not text: return ""
    text = html.unescape(text)
    text = re.sub(r'<script.*?>.*?</script>', '', text, flags=re.DOTALL | re.IGNORECASE)
    text = re.sub(r'<style.*?>.*?</style>', '', text, flags=re.DOTALL | re.IGNORECASE)
    text = re.sub(r'', '', text, flags=re.DOTALL)
    text = re.sub(r'<[^>]+>', ' ', text)
    text = re.sub(r'\bstyle="[^"]+"', '', text)
    text = re.sub(r'\bclass="[^"]+"', '', text)
    text = re.sub(r'\s+', ' ', text).strip()
    return text

def initialize_models():
    global model, sentiment_analyzer, translator
    status_msg = []
    try:
        if model is None:
            model = SentenceTransformer('all-MiniLM-L6-v2')
            status_msg.append("✅ Embedding Modeli Hazır")
    except Exception as e: status_msg.append(f"❌ Embedding: {str(e)}")

    try:
        if sentiment_analyzer is None:
            tokenizer = AutoTokenizer.from_pretrained(CRYPTOBERT_NAME, use_fast=True)
            clf_model = AutoModelForSequenceClassification.from_pretrained(CRYPTOBERT_NAME)
            sentiment_analyzer = TextClassificationPipeline(model=clf_model, tokenizer=tokenizer, device=-1, max_length=128, truncation=True, padding="max_length")
            status_msg.append("✅ Sentiment Modeli Hazır")
    except Exception as e: status_msg.append(f"❌ Sentiment: {str(e)}")

    try:
        if translator is None:
            translator = pipeline("translation", model=TRANSLATOR_NAME, device=-1)
            status_msg.append("✅ Çeviri Modeli Hazır")
    except Exception as e: status_msg.append(f"❌ Çeviri: {str(e)}")
    
    return " | ".join(status_msg)

def _normalize_label(label: str) -> str:
    l = (label or "").strip().lower()
    if l in {"label_2"} or "bull" in l: return "bullish"
    if l in {"label_0"} or "bear" in l: return "bearish"
    return "neutral"

def _cue_score(text: str) -> int:
    t = (text or "").lower()
    pos = sum(1 for w in POSITIVE_CUES if w in t)
    neg = sum(1 for w in NEGATIVE_CUES if w in t)
    if re.search(r"is .* over\??", t): neg += 1
    return pos - neg

def _apply_post_rules(label: str, score: float, text: str) -> tuple[str, float]:
    cue = _cue_score(text)
    if label == "neutral":
        if cue <= -2: return "bearish", max(0.60, score)
        if cue >= 2: return "bullish", max(0.60, score)
    if label == "neutral" and score < NEUTRAL_CONFIDENCE_FLOOR:
        if cue > 0: return "bullish", min(0.60, score + 0.10)
        if cue < 0: return "bearish", min(0.60, score + 0.10)
        return label, score
    if label == "bullish" and cue < -1 and score < 0.70: return "neutral", score * 0.9
    if label == "bearish" and cue > 1 and score < 0.70: return "neutral", score * 0.9
    return label, score

def fetch_feed_data(url):
    headers = { "User-Agent": "Mozilla/5.0 (Windows NT 10.0; Win64; x64) Chrome/120.0.0.0 Safari/537.36", "Accept": "application/rss+xml, text/xml" }
    try:
        response = requests.get(url, headers=headers, timeout=10)
        return feedparser.parse(response.content) if response.status_code == 200 else None
    except: return None

# --- HTML FORMATLAMA FONKSİYONU ---
def format_news_as_html(dataframe):
    if dataframe is None or len(dataframe) == 0:
        return "<div style='padding:20px; text-align:center; color: #666;'>Haber yok. Lütfen 'Verileri Yenile' butonuna basın.</div>"
    
    html_content = "<div class='news-feed-container'>"
    
    for _, row in dataframe.iterrows():
        sentiment = row['sentiment_label']
        score = row['sentiment_score']
        
        # Skor Formatı: 0.98 -> 98
        confidence_percent = int(score * 100)
        
        # Türkçe Etiket
        tr_label = LABEL_MAP.get(sentiment, "NÖTR")
        
        # Renk sınıfları
        color_class = "neutral-card"
        icon = "➖"
        if sentiment == "bullish": 
            color_class = "bullish-card"
            icon = "🚀"
        elif sentiment == "bearish": 
            color_class = "bearish-card"
            icon = "🔻"
            
        html_content += f"""
        <div class='news-card {color_class}'>
            <div class='card-header'>
                <span class='badge {sentiment}'>{icon} {tr_label} (%{confidence_percent})</span>
                <span class='date'>{row['published'][:16]}</span>
            </div>
            <h3><a href="{row['link']}" target="_blank">{row['title']}</a></h3>
            <p>{row['summary'][:160]}...</p>
        </div>
        """
    html_content += "</div>"
    return html_content

def fetch_news_wrapper():
    global df, index, embeddings, sentiment_analyzer, model
    
    if sentiment_analyzer is None:
        return "⚠️ Önce Modelleri Yükleyin!", "", gr.update(choices=[]), None
        
    RSS_URLS = [
        "https://cointelegraph.com/rss",
        "https://cryptonews.com/news/feed",
        "https://www.coindesk.com/arc/outboundfeeds/rss/",
        "https://tr.investing.com/rss/302.rss"
    ]
    
    all_entries = []
    
    for url in RSS_URLS:
        feed = fetch_feed_data(url)
        if feed and feed.entries:
            for entry in feed.entries[:20]:
                clean_title = clean_html_tags(entry.get("title", ""))
                if clean_title:
                    all_entries.append({
                        "title": clean_title,
                        "link": entry.get("link", ""),
                        "summary": clean_html_tags(entry.get("summary", "")),
                        "published": entry.get("published", "")
                    })
    
    if not all_entries:
        return "❌ Haber alınamadı.", "", gr.update(choices=[]), None
        
    df = pd.DataFrame(all_entries).drop_duplicates(subset="title").reset_index(drop=True)
    
    # Sentiment Analizi
    def analyze_row(row):
        text = f"{row['title']}. {row['summary']}"[:1000]
        try:
            out = sentiment_analyzer(text)[0]
            lbl, scr = _apply_post_rules(_normalize_label(out.get("label")), float(out.get("score")), text)
            return lbl, scr
        except: return "neutral", 0.0

    df["sentiment_label"], df["sentiment_score"] = zip(*df.apply(analyze_row, axis=1))
    
    # Faiss
    corpus = df['title'].tolist()
    embeddings = model.encode(corpus, show_progress_bar=False)
    index = faiss.IndexFlatL2(embeddings.shape[1])
    index.add(embeddings.astype('float32'))
    
    # UI Çıktıları
    status_text = f"✅ {len(df)} Haber Analiz Edildi"
    html_feed = format_news_as_html(df)
    choices = [(f"{i}. {t[:40]}...", i) for i, t in enumerate(df["title"])]
    
    # Grafik oluştur (Türkçe Etiketli)
    sentiment_counts = df["sentiment_label"].value_counts()
    
    fig, ax = plt.subplots(figsize=(6, 3))
    colors = {'bullish': '#4CAF50', 'bearish': '#F44336', 'neutral': '#FFC107'}
    
    # Etiketleri Türkçeye çevir
    tr_labels = [LABEL_MAP.get(x, x) for x in sentiment_counts.index]
    bar_colors = [colors.get(x, '#333') for x in sentiment_counts.index]
    
    ax.bar(tr_labels, sentiment_counts.values, color=bar_colors)
    ax.set_title("Piyasa Duygu Durumu")
    plt.tight_layout()
    
    return status_text, html_feed, gr.update(choices=choices, value=None), fig

# --- Çeviri ve Arama Fonksiyonları ---
def perform_translation(news_index):
    global df
    if df is None or news_index is None: return "Lütfen bir haber seçin.", "..."
    try:
        idx = int(news_index)
        row = df.iloc[idx]
        title_tr = translate_text_en_to_tr(row['title'])
        summary_tr = translate_text_en_to_tr(row['summary'])
        return title_tr, summary_tr
    except Exception as e: return f"Hata: {e}", "..."

def translate_text_en_to_tr(text):
    global translator
    if not text: return ""
    try: return translator(text[:512])[0]['translation_text']
    except: return "Çeviri hatası"

def search_news(query):
    global df, index, model
    if df is None: return "Veri yok."
    try:
        q_vec = model.encode([query])
        D, I = index.search(q_vec.astype('float32'), 5)
        results = df.iloc[I[0]]
        return format_news_as_html(results)
    except: return "Arama hatası."

def analyze_coin(coin_name):
    global df
    if df is None: return None, "Veri yok."
    filtered = df[df["title"].str.contains(coin_name, case=False, na=False)]
    if len(filtered) == 0: return None, f"{coin_name} hakkında haber yok."
    
    counts = filtered["sentiment_label"].value_counts()
    
    # Türkçe etiketler ve renkler
    labels_tr = [LABEL_MAP.get(x, x) for x in counts.index]
    colors = ['#4CAF50' if x=='bullish' else '#F44336' if x=='bearish' else '#FFC107' for x in counts.index]
    
    fig, ax = plt.subplots(figsize=(5, 5))
    ax.pie(counts.values, labels=labels_tr, autopct='%1.1f%%', colors=colors)
    ax.set_title(f"{coin_name.upper()} Analizi")
    return fig, format_news_as_html(filtered)

# --- CSS STİLLERİ ---
custom_css = """
/* Genel Kart Yapısı */
.news-card {
    background-color: #ffffff; 
    border-radius: 10px;
    padding: 15px;
    margin-bottom: 15px;
    border: 1px solid #e0e0e0;
    box-shadow: 0 2px 5px rgba(0,0,0,0.05);
}

/* Yazı renklerini zorla koyu yap */
.news-card h3 a {
    text-decoration: none;
    color: #222222 !important; 
    font-weight: 700;
}
.news-card p {
    color: #444444 !important; 
    font-size: 0.95em;
    line-height: 1.5;
}
.card-header {
    display: flex;
    justify-content: space-between;
    margin-bottom: 8px;
}
.date {
    font-size: 0.8em;
    color: #888888 !important;
}

/* Renk Çizgileri */
.bullish-card { border-left: 6px solid #4CAF50; }
.bearish-card { border-left: 6px solid #F44336; }
.neutral-card { border-left: 6px solid #FFC107; }

/* Badge Tasarımı */
.badge {
    padding: 3px 8px;
    border-radius: 4px;
    font-size: 0.75em;
    font-weight: bold;
    color: white;
}
.bullish { background-color: #4CAF50; }
.bearish { background-color: #F44336; }
.neutral { background-color: #FFC107; color: #333; }

.news-feed-container {
    max-height: 800px;
    overflow-y: auto;
    padding-right: 10px;
}
"""

# --- UI TASARIMI ---
with gr.Blocks(theme=gr.themes.Soft(primary_hue="blue"), css=custom_css, title="Crypto News AI") as app:
    
    # Üst Bar (Header)
    with gr.Row(elem_id="header"):
        with gr.Column(scale=3):
            gr.Markdown("# ⚡ AI Crypto Sentiment Dashboard")
        with gr.Column(scale=1):
            init_btn = gr.Button("🚀 1. Modelleri Başlat", variant="primary", size="sm")
            load_status = gr.Textbox(show_label=False, placeholder="Model Durumu...", lines=1)
            
    gr.Markdown("---")
    
    # Ana İçerik
    with gr.Row():
        # --- SOL SÜTUN (Kontrol & Analiz) ---
        with gr.Column(scale=1, min_width=300):
            
            with gr.Group():
                gr.Markdown("### 📡 Veri Kontrol")
                refresh_btn = gr.Button("🔄 2. Verileri Yenile", variant="secondary")
                data_status = gr.Textbox(show_label=False, placeholder="Veri bekleniyor...", lines=1)
                overview_plot = gr.Plot(label="Piyasa Özeti")
            
            gr.Markdown("### 🔍 Hızlı Filtre")
            with gr.Tabs():
                with gr.Tab("Coin Ara"):
                    coin_input = gr.Textbox(placeholder="BTC, ETH, SOL...", show_label=False)
                    coin_btn = gr.Button("Analiz Et")
                    coin_plot = gr.Plot(show_label=False)
                
                with gr.Tab("Haber Ara"):
                    search_input = gr.Textbox(placeholder="Konu girin...", show_label=False)
                    search_btn = gr.Button("Ara")
            
            gr.Markdown("### 🇹🇷 Çeviri Aracı")
            with gr.Group():
                news_selector = gr.Dropdown(label="Haber Seçin", choices=[], type="value", interactive=True)
                translate_btn = gr.Button("Türkçeye Çevir")
                tr_title_out = gr.Textbox(label="Başlık (TR)", lines=2)
                tr_summary_out = gr.Textbox(label="Özet (TR)", lines=4)

        # --- SAĞ SÜTUN (Haber Akışı) ---
        with gr.Column(scale=2):
            gr.Markdown("### 📰 Canlı Haber Akışı")
            news_feed_html = gr.HTML(label="Haberler", value="<div style='padding:20px; color:#666;'>Veriler yüklenince burada görünecek...</div>")

    # --- ETKİLEŞİMLER (EVENTS) ---
    init_btn.click(initialize_models, outputs=load_status)
    
    refresh_btn.click(
        fetch_news_wrapper, 
        outputs=[data_status, news_feed_html, news_selector, overview_plot]
    )
    
    coin_btn.click(analyze_coin, inputs=coin_input, outputs=[coin_plot, news_feed_html])
    search_btn.click(search_news, inputs=search_input, outputs=news_feed_html)
    
    translate_btn.click(
        perform_translation, 
        inputs=news_selector, 
        outputs=[tr_title_out, tr_summary_out]
    )

if __name__ == "__main__":
    app.launch()