#!/usr/bin/env python3
# -*- coding: utf-8 -*-
"""
tests/test_token_report_service.py
LLM Token 日報服務單元測試 (Operation Ollama-First v5.0 — Phase 1 收尾)

測試紀律:
  - 不真連 DB：mock _exec_query 返回固定資料
  - 不真連 Telegram：mock send_telegram_with_result
  - 不真寫 ai_insights：mock _persist_to_ai_insights
  - 7 個告警規則各自獨立觸發測試
  - HTML escape 驗證（caller 名含 < / & 不破版）
  - 訊息字數 ≤ 4096 驗證
"""

from __future__ import annotations

import os
import sys
from datetime import date, datetime, timedelta, timezone
from typing import Any, Dict, List

import pytest

sys.path.insert(0, os.path.dirname(os.path.dirname(os.path.abspath(__file__))))

import services.token_report_service as svc


# ─────────────────────────────────────────────────────────────────────────────
# 共用 fixtures
# ─────────────────────────────────────────────────────────────────────────────

TARGET_DATE = date(2026, 5, 3)


def _make_summary(**overrides) -> Dict[str, Any]:
    base = {
        'total_tokens':      3_142_891,
        'total_calls':       2_847,
        'total_cost_usd':    0.36,
        'avg_duration_ms':   1847.0,
        'success_rate':      98.7,
        'failed_calls':      37,
        'ollama_pct':        64.3,
        'prev_total_tokens': 2_905_000,
        'wow_pct':           8.2,
    }
    base.update(overrides)
    return base


def _make_by_provider(**overrides) -> List[Dict[str, Any]]:
    """7 個 provider 的預設配置，可用 overrides={'gemini': {'pct': 50}} 覆寫"""
    defaults = {
        'gcp_ollama':       {'tokens': 2_021_000, 'pct': 64.3, 'calls': 2103, 'cost_usd': 0.0,  'avg_duration_ms': 1200},
        'ollama_111':       {'tokens':    12_000, 'pct':  0.4, 'calls':   18, 'cost_usd': 0.0,  'avg_duration_ms': 2400},
        'gemini':           {'tokens':   892_000, 'pct': 28.4, 'calls':  589, 'cost_usd': 0.31, 'avg_duration_ms': 2100},
        'claude':           {'tokens':   178_000, 'pct':  5.7, 'calls':   98, 'cost_usd': 0.04, 'avg_duration_ms': 3200},
        'nim':              {'tokens':    28_000, 'pct':  0.9, 'calls':   24, 'cost_usd': 0.0,  'avg_duration_ms': 1800},
        'openrouter':       {'tokens':    12_000, 'pct':  0.4, 'calls':   15, 'cost_usd': 0.01, 'avg_duration_ms': 2900},
        'nim_via_elephant': {'tokens':    27_000, 'pct':  0.9, 'calls':   12, 'cost_usd': 0.0,  'avg_duration_ms': 3100},
    }
    for k, v in (overrides or {}).items():
        defaults.setdefault(k, {}).update(v)
    return [{'provider': k, **v} for k, v in defaults.items()]


def _make_top_callers() -> List[Dict[str, Any]]:
    return [
        {'caller': 'km_embedding_worker', 'provider': 'gcp_ollama',
         'model': 'bge-m3:latest', 'tokens': 892_000, 'calls': 1247, 'delta_pct': 5.0},
        {'caller': 'hermes_analyst', 'provider': 'gcp_ollama',
         'model': 'hermes3:latest', 'tokens': 482_000, 'calls': 72, 'delta_pct': -2.0},
        {'caller': 'code_review_hermes', 'provider': 'claude',
         'model': 'claude-opus-4-7', 'tokens': 158_000, 'calls': 8, 'delta_pct': 42.0},
    ]


def _make_trends() -> Dict[str, Any]:
    return {
        'today_total_tokens':  3_142_000,
        'today_gemini_tokens':   892_000,
        'today_ollama_tokens': 2_033_000,
        'today_claude_tokens':   178_000,
        'today_avg_duration':   1847.0,
        'today_error_rate':       1.3,
        'today_gcp_hit_pct':     99.6,
        '7d_avg_total':       2_905_000,
        '7d_avg_gemini':        948_000,
        '7d_avg_ollama':      1_712_000,
        '7d_avg_claude':        165_000,
        '7d_avg_duration':     1920.0,
        '7d_error_rate':          1.8,
        '7d_total_tokens':   18_832_000,
        '7d_total_cost':         11.84,
        '7d_gcp_hit_pct_7d':     98.9,
        '7d_gcp_hit_pct':        98.9,
    }


def _make_budgets(**overrides) -> Dict[str, Any]:
    base = {
        'daily_spent':    0.36,
        'weekly_spent':   1.92,
        'monthly_spent':  5.84,
        'daily_budget':   1.00,
        'weekly_budget':  5.00,
        'monthly_budget': 20.00,
    }
    base.update(overrides)
    return base


def _make_cache_stats(**overrides) -> Dict[str, Any]:
    base = {
        'claude': {'total': 98, 'hits': 62, 'pct': 63.3},
        'gemini': {'total':  0, 'hits':  0, 'pct':  0.0},
    }
    base.update(overrides)
    return base


# ─────────────────────────────────────────────────────────────────────────────
# 1. 報表組裝測試 — generate_daily_report 路徑
# ─────────────────────────────────────────────────────────────────────────────

class TestReportFormat:
    """測 _format_report 主要章節都出現 & 字數合理。"""

    def test_format_report_contains_all_six_sections(self):
        """6 個段落標題都應出現。"""
        out = svc._format_report(
            target_date=TARGET_DATE,
            summary=_make_summary(),
            by_provider=_make_by_provider(),
            top_callers=_make_top_callers(),
            costs=[{'provider': 'gemini', 'model': 'gemini-2.5-flash', 'cost_usd': 0.26, 'calls': 50}],
            trends=_make_trends(),
            budgets=_make_budgets(),
            cache_stats=_make_cache_stats(),
            alerts=[],
            insights=[{'icon': '✅', 'text': 'Ollama-First 達標'}],
        )
        assert '【1】今日總覽' in out
        assert '【2】供應商分布' in out
        assert '【3】呼叫點 TOP' in out
        assert '【4】成本分析' in out
        assert '【5】趨勢與洞察' in out
        assert '【6】告警與建議' in out

    def test_format_report_under_telegram_limit(self):
        """完整報表（含 10 個 caller / 12 個成本項 / 多個告警）不應超過 4096 字元。"""
        big_callers = _make_top_callers() * 4  # 12 筆
        big_costs = [{'provider': 'p', 'model': f'model-{i}', 'cost_usd': 0.01, 'calls': 1}
                     for i in range(12)]
        big_alerts = [
            {'level': 'P1', 'icon': '🔴', 'title': 'X' * 80, 'suggestion': 'Y' * 80}
            for _ in range(5)
        ]
        out = svc._format_report(
            target_date=TARGET_DATE,
            summary=_make_summary(),
            by_provider=_make_by_provider(),
            top_callers=big_callers[:10],
            costs=big_costs,
            trends=_make_trends(),
            budgets=_make_budgets(),
            cache_stats=_make_cache_stats(),
            alerts=big_alerts,
            insights=[],
        )
        # send_daily_report 端會做 4000 字截斷（HTML 安全），單元測試先確認原始長度可控
        assert len(out) < 6000, f"原始報表 {len(out)} 字元，可能需縮減欄位寬度"

    def test_format_report_html_escape_caller_name(self):
        """caller 名含 <script> 不應原樣輸出（防 HTML 注入）。"""
        nasty_callers = [{
            'caller':   'evil<script>',
            'provider': 'gcp_ollama',
            'model':    'a&b<c>',
            'tokens':   100,
            'calls':    1,
            'delta_pct': None,
        }]
        out = svc._format_report(
            target_date=TARGET_DATE,
            summary=_make_summary(),
            by_provider=_make_by_provider(),
            top_callers=nasty_callers,
            costs=[],
            trends=_make_trends(),
            budgets=_make_budgets(),
            cache_stats=_make_cache_stats(),
            alerts=[],
            insights=[],
        )
        assert '<script>' not in out, "caller 含 <script> 必須被 escape"
        assert '&lt;script&gt;' in out
        assert '&amp;' in out

    def test_failure_report_html_safe(self):
        """DB 失敗時的 fallback 訊息不應洩漏 stack trace 且 HTML 安全。"""
        out = svc._format_failure_report(TARGET_DATE, 'DB error: <a href="x">x</a>')
        assert '日報生成失敗' in out
        assert '&lt;a href' in out  # < 已被 escape


# ─────────────────────────────────────────────────────────────────────────────
# 2. 告警規則測試 — _detect_alerts 7 條規則
# ─────────────────────────────────────────────────────────────────────────────

class TestAlertRules:
    """每條告警規則一個獨立測試，確保都會觸發。"""

    def test_rule1_caller_token_spike(self):
        """R1: 單一 caller 暴增 ≥ +40% (factor=1.4)"""
        callers = [{'caller': 'code_review_hermes', 'provider': 'claude',
                    'model': 'claude-opus-4-7', 'tokens': 158_000,
                    'calls': 8, 'delta_pct': 42.0}]
        alerts = svc._detect_alerts(_make_summary(), _make_by_provider(),
                                    callers, _make_trends(),
                                    _make_budgets(), _make_cache_stats())
        assert any('暴增' in a['title'] and a['level'] == 'P2' for a in alerts), \
            f"R1 未觸發；alerts={alerts}"

    def test_rule2_gemini_share_too_high(self):
        """R2: Gemini 占比 > 35% → 「Ollama-First 失守」"""
        prov = _make_by_provider()
        for p in prov:
            if p['provider'] == 'gemini':
                p['pct'] = 50.0
        alerts = svc._detect_alerts(_make_summary(), prov, [], _make_trends(),
                                    _make_budgets(), _make_cache_stats())
        assert any('Gemini 占比' in a['title'] for a in alerts), \
            f"R2 未觸發；alerts={alerts}"

    def test_rule3_error_rate_critical(self):
        """R3: 全域失敗率 > 5% → P1"""
        summary = _make_summary(failed_calls=300, total_calls=2000)  # 15%
        alerts = svc._detect_alerts(summary, _make_by_provider(), [],
                                    _make_trends(), _make_budgets(), _make_cache_stats())
        p1 = [a for a in alerts if a['level'] == 'P1' and '失敗率' in a['title']]
        assert p1, f"R3 未觸發；alerts={alerts}"

    def test_rule4_budget_overrun(self):
        """R4: 月成本達 80% 預算 → P1"""
        budgets = _make_budgets(monthly_spent=18.0, monthly_budget=20.0)  # 90%
        alerts = svc._detect_alerts(_make_summary(), _make_by_provider(), [],
                                    _make_trends(), budgets, _make_cache_stats())
        assert any('月成本' in a['title'] and a['level'] == 'P1' for a in alerts), \
            f"R4 未觸發；alerts={alerts}"

    def test_rule5_gcp_hit_low(self):
        """R5: GCP Ollama 命中率 < 90% → P2 (但需有 Ollama 流量)"""
        trends = _make_trends()
        trends['today_gcp_hit_pct'] = 70.0
        alerts = svc._detect_alerts(_make_summary(), _make_by_provider(), [],
                                    trends, _make_budgets(), _make_cache_stats())
        assert any('GCP Ollama 命中率' in a['title'] for a in alerts), \
            f"R5 未觸發；alerts={alerts}"

    def test_rule6_claude_cache_low(self):
        """R6: Claude cache 命中率 < 40% (≥10 calls 才檢查) → INFO"""
        cache = _make_cache_stats(claude={'total': 100, 'hits': 20, 'pct': 20.0})
        alerts = svc._detect_alerts(_make_summary(), _make_by_provider(), [],
                                    _make_trends(), _make_budgets(), cache)
        assert any('Claude prompt cache' in a['title'] for a in alerts), \
            f"R6 未觸發；alerts={alerts}"

    def test_rule6_claude_cache_low_skipped_when_few_calls(self):
        """R6 邊界：< 10 calls 時不應觸發告警（樣本不足）"""
        cache = _make_cache_stats(claude={'total': 5, 'hits': 0, 'pct': 0.0})
        alerts = svc._detect_alerts(_make_summary(), _make_by_provider(), [],
                                    _make_trends(), _make_budgets(), cache)
        cache_alerts = [a for a in alerts if 'Claude prompt cache' in a['title']]
        assert not cache_alerts, "樣本不足時不應告警"

    def test_no_alerts_when_healthy(self):
        """健康狀態下應無 P1/P2 告警。"""
        alerts = svc._detect_alerts(_make_summary(), _make_by_provider(),
                                    _make_top_callers()[:2],  # 不含 +42% spike
                                    _make_trends(), _make_budgets(),
                                    _make_cache_stats())
        critical = [a for a in alerts if a['level'] in ('P1', 'P2')]
        assert not critical, f"健康狀態不應有 P1/P2 告警；得到：{critical}"


# ─────────────────────────────────────────────────────────────────────────────
# 3. 智能建議測試 — _generate_insights
# ─────────────────────────────────────────────────────────────────────────────

class TestInsights:

    def test_ollama_first_target_met(self):
        """Ollama 占比 ≥ 60% → 應含「達標」建議。"""
        insights = svc._generate_insights(TARGET_DATE,
                                          _make_summary(ollama_pct=64.3),
                                          _make_by_provider())
        assert any('達標' in i['text'] for i in insights)

    def test_ollama_first_target_missed(self):
        """Ollama 占比 < 60% → 應含「未達」建議。"""
        insights = svc._generate_insights(TARGET_DATE,
                                          _make_summary(ollama_pct=45.0),
                                          _make_by_provider())
        assert any('未達' in i['text'] for i in insights)

    def test_nim_low_usage_suggestion(self):
        """NIM 用量 < 100K 時應建議下線 NIM。"""
        prov = _make_by_provider()
        for p in prov:
            if p['provider'] in ('nim', 'nim_via_elephant'):
                p['tokens'] = 5000
        insights = svc._generate_insights(TARGET_DATE, _make_summary(), prov)
        assert any('NIM 用量' in i['text'] for i in insights)


# ─────────────────────────────────────────────────────────────────────────────
# 4. SQL 查詢測試 — mock _exec_query 驗證 SQL 結構正確
# ─────────────────────────────────────────────────────────────────────────────

class TestQueriesViaMock:
    """mock _exec_query 確認查詢函數呼叫順序與參數正確。"""

    def test_query_summary_calls_two_windows(self, monkeypatch):
        """_query_summary 應分別查今日 + 昨日（共 2 次 SQL）。"""
        captured: List[Dict] = []

        def fake_exec(sql, params):
            captured.append({'sql_head': sql.strip().split('\n')[0],
                             'params': dict(params)})
            # 第 1 次回今日資料；第 2 次回昨日資料
            if 'COUNT(*)' in sql:
                return [{'total_tokens': 100_000, 'total_calls': 50,
                         'total_cost_usd': 0.5, 'avg_duration_ms': 1500,
                         'ok_calls': 49, 'ollama_tokens': 70_000}]
            return [{'prev_total_tokens': 90_000}]

        monkeypatch.setattr(svc, '_exec_query', fake_exec)
        result = svc._query_summary(TARGET_DATE)

        assert len(captured) == 2
        # 第二次查詢的 end 應等於第一次的 start（昨日窗）
        assert captured[1]['params']['end'] == captured[0]['params']['start']

        assert result['total_tokens'] == 100_000
        assert result['ollama_pct'] == pytest.approx(70.0, rel=0.01)
        assert result['success_rate'] == pytest.approx(98.0, rel=0.01)
        assert result['failed_calls'] == 1
        assert result['wow_pct'] == pytest.approx(11.11, rel=0.01)

    def test_query_by_provider_returns_all_eight_providers(self, monkeypatch):
        """即使只有 1 個 provider 有資料，也要回傳 8 個 provider（0 占位）。
        critic-A11 B4 修補：補 ollama_secondary 後從 7 → 8 個（三主機架構一致性）。
        """
        def fake_exec(sql, params):
            return [{'provider': 'gcp_ollama', 'tokens': 1000, 'calls': 5,
                     'cost_usd': 0.0, 'avg_duration_ms': 1000}]
        monkeypatch.setattr(svc, '_exec_query', fake_exec)
        result = svc._query_by_provider(TARGET_DATE)
        assert len(result) == 8
        gcp = next(r for r in result if r['provider'] == 'gcp_ollama')
        assert gcp['tokens'] == 1000
        secondary = next(r for r in result if r['provider'] == 'ollama_secondary')
        assert secondary['tokens'] == 0  # 沒資料應給 0
        gemini = next(r for r in result if r['provider'] == 'gemini')
        assert gemini['tokens'] == 0  # 沒資料應給 0

    def test_query_top_callers_orders_by_tokens(self, monkeypatch):
        def fake_exec(sql, params):
            return [
                {'caller': 'a', 'provider': 'gcp_ollama', 'top_model': 'm1',
                 'tokens': 500, 'calls': 5, 'avg_tokens_7d': 400},
                {'caller': 'b', 'provider': 'gemini', 'top_model': 'm2',
                 'tokens': 200, 'calls': 2, 'avg_tokens_7d': 0},
            ]
        monkeypatch.setattr(svc, '_exec_query', fake_exec)
        result = svc._query_top_callers(TARGET_DATE, limit=10)
        assert len(result) == 2
        assert result[0]['caller'] == 'a'
        # delta = (500-400)/400 = 25%
        assert result[0]['delta_pct'] == pytest.approx(25.0, rel=0.01)
        # baseline=0 → delta_pct=None（避免除 0）
        assert result[1]['delta_pct'] is None

    def test_query_cost_breakdown_filters_zero_cost(self, monkeypatch):
        """Ollama 等成本 0 的 model 不應出現在拆解中。"""
        captured = []
        def fake_exec(sql, params):
            captured.append(sql)
            return []
        monkeypatch.setattr(svc, '_exec_query', fake_exec)
        svc._query_cost_breakdown(TARGET_DATE)
        assert 'cost_usd > 0' in captured[0]


# ─────────────────────────────────────────────────────────────────────────────
# 5. send_daily_report 整合 — mock 整條鏈
# ─────────────────────────────────────────────────────────────────────────────

class TestSendDailyReport:

    def test_send_happy_path(self, monkeypatch):
        """整條鏈走通：generate → send → persist 都被呼叫。"""
        monkeypatch.setattr(svc, 'generate_daily_report', lambda d: '<b>OK</b>')

        sent_calls = []
        def fake_send(text, **kwargs):
            sent_calls.append({'text': text, 'kwargs': kwargs})
            return {'ok': True, 'sent': 1, 'failed': 0, 'chat_ids': [-1], 'errors': []}

        # mock telegram_templates.send_telegram_with_result
        import services.telegram_templates as tg
        monkeypatch.setattr(tg, 'send_telegram_with_result', fake_send)

        persist_calls = []
        monkeypatch.setattr(svc, '_persist_to_ai_insights',
                            lambda d, c, r: persist_calls.append((d, c, r)))

        result = svc.send_daily_report(TARGET_DATE)

        assert result['ok'] is True
        assert result['sent'] == 1
        assert len(sent_calls) == 1
        assert sent_calls[0]['kwargs'].get('parse_mode') == 'HTML'
        assert len(persist_calls) == 1
        assert persist_calls[0][0] == TARGET_DATE

    def test_send_truncates_oversized_message(self, monkeypatch):
        """訊息 > 4000 應自動截斷並加省略尾。"""
        big = 'X' * 5000
        monkeypatch.setattr(svc, 'generate_daily_report', lambda d: big)

        captured_text = []
        def fake_send(text, **kwargs):
            captured_text.append(text)
            return {'ok': True, 'sent': 1, 'failed': 0, 'chat_ids': [], 'errors': []}

        import services.telegram_templates as tg
        monkeypatch.setattr(tg, 'send_telegram_with_result', fake_send)
        monkeypatch.setattr(svc, '_persist_to_ai_insights', lambda *a, **k: None)

        svc.send_daily_report(TARGET_DATE)

        assert len(captured_text) == 1
        assert len(captured_text[0]) <= svc._TELEGRAM_MAX_CHARS
        assert '截斷' in captured_text[0]

    def test_send_resilient_to_telegram_failure(self, monkeypatch):
        """Telegram 送失敗時 send_daily_report 仍應回 dict（不爆）。"""
        monkeypatch.setattr(svc, 'generate_daily_report', lambda d: 'msg')

        def boom(text, **kwargs):
            raise RuntimeError("network down")

        import services.telegram_templates as tg
        monkeypatch.setattr(tg, 'send_telegram_with_result', boom)
        monkeypatch.setattr(svc, '_persist_to_ai_insights', lambda *a, **k: None)

        result = svc.send_daily_report(TARGET_DATE)
        assert result['ok'] is False
        assert any('telegram' in e for e in result['errors'])

    def test_generate_returns_failure_msg_when_db_dies(self, monkeypatch):
        """DB 例外時 generate_daily_report 應回 fallback 字串而不是丟 exception。"""
        def boom(*a, **kw):
            raise RuntimeError("DB connection refused")
        monkeypatch.setattr(svc, '_query_summary', boom)

        out = svc.generate_daily_report(TARGET_DATE)
        assert '日報生成失敗' in out
        assert '<code>' in out  # fallback 訊息含 escape 過的錯誤


# ─────────────────────────────────────────────────────────────────────────────
# 6. telegram_templates.daily_token_report 包裝測試
# ─────────────────────────────────────────────────────────────────────────────

class TestTelegramTemplate:

    def test_daily_token_report_appends_footer(self):
        from services.telegram_templates import daily_token_report
        out = daily_token_report("body", footer_url="http://x/y")
        assert 'body' in out
        assert 'http://x/y' in out

    def test_daily_token_report_truncates_to_4096(self):
        from services.telegram_templates import daily_token_report
        big = 'A' * 5000
        out = daily_token_report(big)
        assert len(out) <= 4096
        assert '截斷' in out

    def test_daily_token_report_escapes_footer_url(self):
        """footer_url 含特殊字元應被 escape。"""
        from services.telegram_templates import daily_token_report
        out = daily_token_report("body", footer_url="http://x/?a=1&b=<2>")
        assert '<2>' not in out  # 應 escape
        assert '&amp;' in out or '&lt;2&gt;' in out


# ─────────────────────────────────────────────────────────────────────────────
# 7. 格式化工具測試
# ─────────────────────────────────────────────────────────────────────────────

class TestFormatHelpers:

    def test_fmt_kb(self):
        assert svc._fmt_kb(0) == '0'
        assert svc._fmt_kb(500) == '500'
        assert svc._fmt_kb(1500) == '2K'  # round
        assert svc._fmt_kb(2_021_000) == '2.0M'

    def test_esc_handles_none(self):
        assert svc._esc(None) == ''
        assert svc._esc('<a>') == '&lt;a&gt;'
        assert svc._esc('a&b') == 'a&amp;b'

    def test_budget_line_zero_budget(self):
        line = svc._budget_line("📅 本日", 0.5, 0.0)
        assert '未設定預算' in line

    def test_trend_line_handles_zero_baseline(self):
        line = svc._trend_line("X", 100.0, 0.0)
        assert '—' in line  # 無基準應顯示「—」