from collections import Counter
from datetime import datetime, timedelta

from flask import Blueprint, current_app, request
from sqlalchemy import func, or_

from app.extensions import db
from app.ml.naive_bayes_classifier import NaiveBayesSpamClassifier
from app.models import ContentPost, DetectionConfig, SpamPredictionLog, SpamTrainingSample, User
from app.utils.auth import admin_required, current_user
from app.utils.response import fail, ok


admin_bp = Blueprint("admin", __name__)


def _day_key(day_value) -> str:
    if hasattr(day_value, "isoformat"):
        return day_value.isoformat()
    return str(day_value)


def _tokenize(text: str) -> list[str]:
    content = (text or "").strip()
    if len(content) <= 1:
        return []
    tokens = []
    for i in range(len(content) - 1):
        token = content[i : i + 2]
        if token.strip():
            tokens.append(token)
    return tokens


def _get_or_create_config() -> DetectionConfig:
    cfg = DetectionConfig.query.order_by(DetectionConfig.id.asc()).first()
    if cfg:
        return cfg

    cfg = DetectionConfig(spam_threshold=0.75)
    db.session.add(cfg)
    db.session.commit()
    return cfg


def _serialize_post(item: ContentPost) -> dict:
    row = item.to_dict()
    row["username"] = item.author.username if item.author else ""
    row["nickname"] = item.author.nickname if item.author else ""
    row["recipient_username"] = item.recipient.username if item.recipient else ""
    row["recipient_nickname"] = item.recipient.nickname if item.recipient else ""
    row["reviewer_username"] = item.reviewer.username if item.reviewer else ""
    return row


def _upsert_manual_sample(text: str, label: str, admin_id: int | None) -> None:
    existed = SpamTrainingSample.query.filter_by(text=text, label=label).first()
    if existed:
        existed.is_active = True
        existed.source = existed.source or "manual_review"
        return

    row = SpamTrainingSample(
        text=text,
        label=label,
        source="manual_review",
        created_by=admin_id,
        is_active=True,
    )
    db.session.add(row)


@admin_bp.get("/stats")
@admin_required
def stats():
    user_count = User.query.count()
    sample_count = SpamTrainingSample.query.count()
    predict_count = SpamPredictionLog.query.count()
    post_count = ContentPost.query.count()
    blocked_count = ContentPost.query.filter_by(status="blocked").count()
    published_count = ContentPost.query.filter_by(status="published").count()
    pending_appeal_count = ContentPost.query.filter_by(appeal_status="pending").count()

    now = datetime.utcnow()
    week_ago = now - timedelta(days=6)

    trend_rows = (
        db.session.query(func.date(ContentPost.created_at), func.count(ContentPost.id))
        .filter(ContentPost.created_at >= week_ago)
        .group_by(func.date(ContentPost.created_at))
        .all()
    )

    blocked_7d_count = ContentPost.query.filter(ContentPost.created_at >= week_ago, ContentPost.status == "blocked").count() or 0
    total_7d_count = ContentPost.query.filter(ContentPost.created_at >= week_ago).count() or 0

    day_map = {_day_key(day): int(count or 0) for day, count in trend_rows}
    trend = []
    today = now.date()
    for offset in range(6, -1, -1):
        day = today - timedelta(days=offset)
        key = day.isoformat()
        trend.append({"date": key, "label": day.strftime("%m-%d"), "post_count": day_map.get(key, 0)})

    source_rows = (
        db.session.query(SpamTrainingSample.source, func.count(SpamTrainingSample.id))
        .group_by(SpamTrainingSample.source)
        .order_by(func.count(SpamTrainingSample.id).desc())
        .all()
    )
    source_dist = [{"name": (name or "unknown"), "value": int(value or 0)} for name, value in source_rows]

    blocked_logs = (
        ContentPost.query.filter(ContentPost.created_at >= week_ago, ContentPost.status == "blocked")
        .order_by(ContentPost.id.desc())
        .limit(1000)
        .all()
    )

    token_counter = Counter()
    for row in blocked_logs:
        token_counter.update(_tokenize(row.text))
    top_keywords = [{"token": token, "count": count} for token, count in token_counter.most_common(12)]

    cfg = _get_or_create_config()

    clf = NaiveBayesSpamClassifier(current_app.config["NB_MODEL_PATH"])
    clf.load()

    return ok(
        {
            "user_count": user_count,
            "sample_count": sample_count,
            "predict_count": predict_count,
            "post_count": post_count,
            "blocked_count": blocked_count,
            "published_count": published_count,
            "pending_appeal_count": pending_appeal_count,
            "blocked_ratio_7d": round(blocked_7d_count / total_7d_count, 4) if total_7d_count else 0,
            "total_7d": total_7d_count,
            "trend_7d": trend,
            "source_distribution": source_dist,
            "top_keywords": top_keywords,
            "model_info": clf.model_info(),
            "threshold": cfg.to_dict(),
        }
    )


@admin_bp.get("/stats/report")
@admin_required
def generate_report():
    """生成运营报告：垃圾信息变化、风险词排名、误判率趋势"""
    now = datetime.utcnow()
    week_ago = now - timedelta(days=13)  # 近14天

    # 1. 垃圾信息数量变化（近14天）
    blocked_trend_rows = (
        db.session.query(func.date(ContentPost.created_at), func.count(ContentPost.id))
        .filter(ContentPost.created_at >= week_ago, ContentPost.status == "blocked")
        .group_by(func.date(ContentPost.created_at))
        .all()
    )
    published_trend_rows = (
        db.session.query(func.date(ContentPost.created_at), func.count(ContentPost.id))
        .filter(ContentPost.created_at >= week_ago, ContentPost.status == "published")
        .group_by(func.date(ContentPost.created_at))
        .all()
    )

    blocked_map = {_day_key(day): int(count or 0) for day, count in blocked_trend_rows}
    published_map = {_day_key(day): int(count or 0) for day, count in published_trend_rows}

    spam_trend = []
    today = now.date()
    for offset in range(13, -1, -1):
        day = today - timedelta(days=offset)
        key = day.isoformat()
        spam_trend.append({
            "date": key,
            "label": day.strftime("%m-%d"),
            "blocked": blocked_map.get(key, 0),
            "published": published_map.get(key, 0),
            "total": blocked_map.get(key, 0) + published_map.get(key, 0)
        })

    # 2. 高频风险词排名（近14天）
    blocked_logs = (
        ContentPost.query.filter(ContentPost.created_at >= week_ago, ContentPost.status == "blocked")
        .order_by(ContentPost.id.desc())
        .limit(1000)
        .all()
    )
    token_counter = Counter()
    for row in blocked_logs:
        token_counter.update(_tokenize(row.text))
    top_keywords = [{"token": token, "count": count} for token, count in token_counter.most_common(20)]

    # 3. 误判率趋势（近14天，基于人工复核）
    review_trend_rows = (
        db.session.query(func.date(ContentPost.manual_review_at), func.count(ContentPost.id))
        .filter(ContentPost.manual_review_at >= week_ago, ContentPost.manual_review_status != "none")
        .group_by(func.date(ContentPost.manual_review_at))
        .all()
    )
    approved_trend_rows = (
        db.session.query(func.date(ContentPost.manual_review_at), func.count(ContentPost.id))
        .filter(ContentPost.manual_review_at >= week_ago, ContentPost.manual_review_status == "approved_ham")
        .group_by(func.date(ContentPost.manual_review_at))
        .all()
    )

    review_map = {_day_key(day): int(count or 0) for day, count in review_trend_rows}
    approved_map = {_day_key(day): int(count or 0) for day, count in approved_trend_rows}

    misjudge_trend = []
    for offset in range(13, -1, -1):
        day = today - timedelta(days=offset)
        key = day.isoformat()
        reviewed = review_map.get(key, 0)
        approved = approved_map.get(key, 0)
        misjudge_rate = round(approved / reviewed, 4) if reviewed > 0 else 0
        misjudge_trend.append({
            "date": key,
            "label": day.strftime("%m-%d"),
            "reviewed": reviewed,
            "approved": approved,
            "misjudge_rate": misjudge_rate,
            "misjudge_rate_text": f"{misjudge_rate * 100:.1f}%"
        })

    # 4. 汇总统计
    total_blocked_14d = sum(blocked_map.values())
    total_published_14d = sum(published_map.values())
    total_reviews_14d = sum(review_map.values())
    total_approved_14d = sum(approved_map.values())
    avg_misjudge_rate = round(total_approved_14d / total_reviews_14d, 4) if total_reviews_14d > 0 else 0

    return ok({
        "report_date": now.isoformat(),
        "period": "近14天",
        "spam_trend": spam_trend,
        "top_keywords": top_keywords,
        "misjudge_trend": misjudge_trend,
        "summary": {
            "total_blocked": total_blocked_14d,
            "total_published": total_published_14d,
            "total_posts": total_blocked_14d + total_published_14d,
            "blocked_ratio": round(total_blocked_14d / (total_blocked_14d + total_published_14d), 4) if (total_blocked_14d + total_published_14d) > 0 else 0,
            "total_reviews": total_reviews_14d,
            "total_approved": total_approved_14d,
            "avg_misjudge_rate": avg_misjudge_rate,
            "avg_misjudge_rate_text": f"{avg_misjudge_rate * 100:.1f}%"
        }
    })


@admin_bp.get("/detection/threshold")
@admin_required
def get_threshold():
    return ok(_get_or_create_config().to_dict())


@admin_bp.put("/detection/threshold")
@admin_required
def set_threshold():
    payload = request.get_json(silent=True) or {}
    try:
        threshold = float(payload.get("spam_threshold"))
    except Exception:
        return fail("spam_threshold 必须是数字", 400)

    if threshold < 0.01 or threshold > 0.99:
        return fail("spam_threshold 必须在 0.01 到 0.99 之间", 400)

    cfg = _get_or_create_config()
    admin = current_user()
    cfg.spam_threshold = threshold
    cfg.updated_by = admin.id if admin else None
    db.session.commit()
    return ok(cfg.to_dict(), "阈值更新成功")


@admin_bp.get("/intercepts")
@admin_required
def list_intercepts():
    keyword = (request.args.get("keyword") or "").strip()
    status = (request.args.get("status") or "blocked").strip().lower()
    review_status = (request.args.get("review_status") or "").strip().lower()
    page = max(int(request.args.get("page", 1) or 1), 1)
    page_size = min(max(int(request.args.get("page_size", 20) or 20), 1), 100)

    query = ContentPost.query
    if keyword:
        query = query.filter(ContentPost.text.like(f"%{keyword}%"))
    if status in {"blocked", "published"}:
        query = query.filter(ContentPost.status == status)
    if review_status in {"none", "pending", "confirmed_spam", "approved_ham"}:
        query = query.filter(ContentPost.manual_review_status == review_status)

    pagination = query.order_by(ContentPost.id.desc()).paginate(page=page, per_page=page_size, error_out=False)
    return ok(
        {
            "items": [_serialize_post(item) for item in pagination.items],
            "total": pagination.total,
            "page": page,
            "page_size": page_size,
        }
    )


@admin_bp.put("/intercepts/<int:post_id>/review")
@admin_required
def review_intercept(post_id: int):
    row = ContentPost.query.get(post_id)
    if not row:
        return fail("记录不存在", 404)

    payload = request.get_json(silent=True) or {}
    decision = (payload.get("decision") or "").strip().lower()
    note = (payload.get("note") or "").strip()
    if decision not in {"spam", "ham"}:
        return fail("decision 必须是 spam 或 ham", 400)

    admin = current_user()
    now = datetime.utcnow()

    row.manual_review_by = admin.id if admin else None
    row.manual_review_note = note
    row.manual_review_at = now

    if decision == "spam":
        row.status = "blocked"
        row.prediction = "spam"
        row.manual_review_status = "confirmed_spam"
        if row.appeal_status == "pending":
            row.appeal_status = "rejected"
            row.appeal_admin_note = note or "人工复核确认为垃圾信息"
            row.appeal_processed_at = now
            row.appeal_processed_by = admin.id if admin else None
        _upsert_manual_sample(row.text, "spam", admin.id if admin else None)
    else:
        row.status = "published"
        row.prediction = "ham"
        row.manual_review_status = "approved_ham"
        if row.appeal_status == "pending":
            row.appeal_status = "approved"
            row.appeal_admin_note = note or "人工复核后解除拦截"
            row.appeal_processed_at = now
            row.appeal_processed_by = admin.id if admin else None
        _upsert_manual_sample(row.text, "ham", admin.id if admin else None)

    db.session.commit()
    return ok(_serialize_post(row), "人工复核完成")


@admin_bp.get("/appeals")
@admin_required
def list_appeals():
    keyword = (request.args.get("keyword") or "").strip()
    status = (request.args.get("status") or "pending").strip().lower()
    page = max(int(request.args.get("page", 1) or 1), 1)
    page_size = min(max(int(request.args.get("page_size", 20) or 20), 1), 100)

    query = ContentPost.query.filter(ContentPost.appeal_status != "none")
    if keyword:
        query = query.filter(
            or_(
                ContentPost.text.like(f"%{keyword}%"),
                ContentPost.appeal_reason.like(f"%{keyword}%"),
                ContentPost.appeal_admin_note.like(f"%{keyword}%"),
            )
        )
    if status in {"pending", "approved", "rejected"}:
        query = query.filter(ContentPost.appeal_status == status)

    pagination = query.order_by(ContentPost.id.desc()).paginate(page=page, per_page=page_size, error_out=False)
    return ok(
        {
            "items": [_serialize_post(item) for item in pagination.items],
            "total": pagination.total,
            "page": page,
            "page_size": page_size,
        }
    )


@admin_bp.put("/appeals/<int:post_id>/process")
@admin_required
def process_appeal(post_id: int):
    row = ContentPost.query.get(post_id)
    if not row:
        return fail("记录不存在", 404)
    if row.appeal_status != "pending":
        return fail("该申诉不在待处理状态", 400)

    payload = request.get_json(silent=True) or {}
    action = (payload.get("action") or "").strip().lower()
    note = (payload.get("note") or "").strip()
    if action not in {"approve", "reject"}:
        return fail("action 必须是 approve 或 reject", 400)

    admin = current_user()
    now = datetime.utcnow()

    row.appeal_status = "approved" if action == "approve" else "rejected"
    row.appeal_admin_note = note
    row.appeal_processed_at = now
    row.appeal_processed_by = admin.id if admin else None

    row.manual_review_by = admin.id if admin else None
    row.manual_review_note = note
    row.manual_review_at = now

    if action == "approve":
        row.status = "published"
        row.prediction = "ham"
        row.manual_review_status = "approved_ham"
        _upsert_manual_sample(row.text, "ham", admin.id if admin else None)
        # 申诉通过，增加用户信誉分
        if row.author:
            row.author.credit_score = min(200, (row.author.credit_score or 100) + 10)
    else:
        row.status = "blocked"
        row.prediction = "spam"
        row.manual_review_status = "confirmed_spam"
        _upsert_manual_sample(row.text, "spam", admin.id if admin else None)
        # 申诉驳回，减少用户信誉分
        if row.author:
            row.author.credit_score = max(0, (row.author.credit_score or 100) - 5)

    db.session.commit()
    return ok(_serialize_post(row), "申诉处理完成")


@admin_bp.get("/users")
@admin_required
def list_users():
    keyword = (request.args.get("keyword") or "").strip()
    page = max(int(request.args.get("page", 1) or 1), 1)
    page_size = min(max(int(request.args.get("page_size", 20) or 20), 1), 100)

    query = User.query
    if keyword:
        query = query.filter(User.username.like(f"%{keyword}%") | User.nickname.like(f"%{keyword}%"))

    pagination = query.order_by(User.id.desc()).paginate(page=page, per_page=page_size, error_out=False)
    return ok(
        {
            "items": [item.to_dict() for item in pagination.items],
            "total": pagination.total,
            "page": page,
            "page_size": page_size,
        }
    )


@admin_bp.post("/users/import")
@admin_required
def import_users():
    payload = request.get_json(silent=True) or {}
    items = payload.get("items") or []
    if not isinstance(items, list) or not items:
        return fail("items 必须是非空数组", 400)

    created = 0
    updated = 0

    for row in items:
        username = (row.get("username") or "").strip()
        if len(username) < 3:
            continue

        user = User.query.filter_by(username=username).first()
        if not user:
            user = User(
                username=username,
                nickname=(row.get("nickname") or username).strip(),
                company=(row.get("company") or "").strip(),
                title=(row.get("title") or "").strip(),
                phone=(row.get("phone") or "").strip(),
                is_admin=bool(row.get("is_admin", False)),
            )
            user.set_password(row.get("password") or "123456")
            db.session.add(user)
            created += 1
            continue

        user.nickname = (row.get("nickname") or user.nickname).strip()
        user.company = (row.get("company") or user.company).strip()
        user.title = (row.get("title") or user.title).strip()
        user.phone = (row.get("phone") or user.phone).strip()
        if "is_admin" in row:
            user.is_admin = bool(row.get("is_admin"))
        if row.get("password"):
            user.set_password(row["password"])
        updated += 1

    db.session.commit()
    return ok({"created": created, "updated": updated}, "用户导入完成")


@admin_bp.put("/users/<int:user_id>")
@admin_required
def update_user(user_id: int):
    user = User.query.get(user_id)
    if not user:
        return fail("用户不存在", 404)

    payload = request.get_json(silent=True) or {}
    if "nickname" in payload:
        user.nickname = (payload.get("nickname") or user.nickname).strip()
    if "company" in payload:
        user.company = (payload.get("company") or "").strip()
    if "title" in payload:
        user.title = (payload.get("title") or "").strip()
    if "phone" in payload:
        user.phone = (payload.get("phone") or "").strip()
    if "is_admin" in payload:
        user.is_admin = bool(payload.get("is_admin"))
    if "credit_score" in payload:
        try:
            credit = int(payload.get("credit_score", 100))
            user.credit_score = max(0, min(200, credit))
        except Exception:
            pass
    if payload.get("password"):
        if len(payload["password"]) < 6:
            return fail("密码至少6位", 400)
        user.set_password(payload["password"])

    db.session.commit()
    return ok(user.to_dict(), "用户更新成功")


@admin_bp.delete("/users/<int:user_id>")
@admin_required
def delete_user(user_id: int):
    user = User.query.get(user_id)
    if not user:
        return fail("用户不存在", 404)

    if user.is_admin and User.query.filter_by(is_admin=True).count() <= 1:
        return fail("至少保留一个管理员账号", 400)

    db.session.delete(user)
    db.session.commit()
    return ok({}, "用户已删除")


@admin_bp.put("/users/<int:user_id>/credit")
@admin_required
def update_user_credit(user_id: int):
    """手动调整用户信誉分"""
    user = User.query.get(user_id)
    if not user:
        return fail("用户不存在", 404)

    payload = request.get_json(silent=True) or {}
    try:
        credit = int(payload.get("credit_score", user.credit_score or 100))
        credit = max(0, min(200, credit))
    except Exception:
        return fail("信誉分必须是0-200之间的整数", 400)

    user.credit_score = credit
    db.session.commit()
    return ok(user.to_dict(), "信誉分已更新")


@admin_bp.post("/users/recalculate-credit")
@admin_required
def recalculate_all_credit():
    """根据用户发布历史和申诉通过率重新计算信誉分"""
    users = User.query.filter_by(is_admin=False).all()
    updated_count = 0

    for user in users:
        posts = ContentPost.query.filter_by(user_id=user.id).all()
        if not posts:
            continue

        # 计算发布成功率
        published_count = sum(1 for p in posts if p.status == "published")
        blocked_count = sum(1 for p in posts if p.status == "blocked")
        total_count = len(posts)

        if total_count == 0:
            continue

        publish_ratio = published_count / total_count

        # 计算申诉通过率
        appeals = [p for p in posts if p.appeal_status != "none"]
        approved_appeals = sum(1 for p in appeals if p.appeal_status == "approved")
        appeal_ratio = approved_appeals / len(appeals) if appeals else 0

        # 基础信誉分：发布成功率贡献
        base_score = 100
        if publish_ratio >= 0.9:
            base_score += 30  # 90%以上发布成功，+30
        elif publish_ratio >= 0.7:
            base_score += 15  # 70%以上，+15
        elif publish_ratio < 0.5:
            base_score -= 20  # 低于50%，-20

        # 申诉通过率贡献
        if appeal_ratio >= 0.8 and len(appeals) >= 3:
            base_score += 20  # 80%以上申诉通过且有3次以上申诉，+20
        elif appeal_ratio >= 0.5 and len(appeals) >= 2:
            base_score += 10

        # 限制范围
        user.credit_score = max(0, min(200, base_score))
        updated_count += 1

    db.session.commit()
    return ok({"updated_count": updated_count}, "信誉分批量重算完成")