#!/bin/bash
set -euo pipefail

: '
LoCoMo small 全量评测脚本（locomo10_small.json 全部 samples）

用法:
  ./run_eval_small.sh --import-mode ov      # OV 直接导入 → QA → Judge → Stat
  ./run_eval_small.sh --import-mode claw    # OpenClaw 会话导入 → QA → Judge → Stat
  ./run_eval_small.sh --import-mode both    # OV + OpenClaw 并行导入 → QA → Judge → Stat
  ./run_eval_small.sh --import-mode skip    # 跳过导入 → QA → Judge → Stat

可选参数:
  --run-id NAME          运行标识（默认: 时间戳），用于结果目录命名
  --force-ingest         强制重新导入（忽略已有记录）
  --qa-parallel N        QA 并发数（默认: 10）
  --judge-parallel N     Judge 并发数（默认: 10）
  --gateway-url URL      OpenClaw Gateway 地址（默认: http://127.0.0.1:19789）
  --gateway-token TOKEN  Gateway 认证 Token
  --ov-url URL           OpenViking 服务地址（默认: http://127.0.0.1:2934）
  --skip-judge           跳过 Judge 评分
  --skip-stat            跳过统计
'

SCRIPT_DIR="$(cd "$(dirname "$0")" && pwd)"
DATA_FILE="$SCRIPT_DIR/../data/locomo10_small.json"
SAMPLE_ARG=""
DATASET_TAG="small"

IMPORT_MODE=""
RUN_ID="$(date +%Y%m%d_%H%M%S)"
FORCE_INGEST=""
QA_PARALLEL=10
JUDGE_PARALLEL=10
GATEWAY_URL="http://127.0.0.1:19789"
GATEWAY_TOKEN="${OPENCLAW_GATEWAY_TOKEN:-}"
OV_URL="http://127.0.0.1:2934"
SKIP_JUDGE=false
SKIP_STAT=false

while [[ $# -gt 0 ]]; do
    case $1 in
        --import-mode)
            IMPORT_MODE="$2"; shift 2 ;;
        --run-id)
            RUN_ID="$2"; shift 2 ;;
        --force-ingest)
            FORCE_INGEST="--force-ingest"; shift ;;
        --qa-parallel)
            QA_PARALLEL="$2"; shift 2 ;;
        --judge-parallel)
            JUDGE_PARALLEL="$2"; shift 2 ;;
        --gateway-url)
            GATEWAY_URL="$2"; shift 2 ;;
        --gateway-token)
            GATEWAY_TOKEN="$2"; shift 2 ;;
        --ov-url)
            OV_URL="$2"; shift 2 ;;
        --skip-judge)
            SKIP_JUDGE=true; shift ;;
        --skip-stat)
            SKIP_STAT=true; shift ;;
        *)
            echo "未知参数: $1"; exit 1 ;;
    esac
done

if [[ -z "$IMPORT_MODE" ]]; then
    echo "错误: 必须指定 --import-mode (ov|claw|both|skip)"
    echo "用法: $0 --import-mode ov|claw|both|skip [选项]"
    exit 1
fi

RESULT_DIR="$SCRIPT_DIR/result/${DATASET_TAG}_${RUN_ID}"
mkdir -p "$RESULT_DIR"
OUTPUT_CSV="$RESULT_DIR/qa_results.csv"

echo "============================================"
echo "  LoCoMo 评测: $DATASET_TAG (全部 samples)"
echo "  数据集: locomo10_small.json"
echo "  导入模式: $IMPORT_MODE"
echo "  运行标识: $RUN_ID"
echo "  结果目录: $RESULT_DIR"
echo "============================================"

# Step 1: 导入
case $IMPORT_MODE in
    ov)
        echo "[1/4] OV 直接导入 (全部 samples)..."
        python3 "$SCRIPT_DIR/import_to_ov.py" \
            --input "$DATA_FILE" \
            --openviking-url "$OV_URL" \
            --no-user-agent-id \
            --success-csv "$RESULT_DIR/import_success.csv" \
            --error-log "$RESULT_DIR/import_errors.log" \
            $FORCE_INGEST \
            2>&1 | tee "$RESULT_DIR/import_ov.log"
        echo "导入完成，等待 60 秒让 OV 处理..."
        sleep 60
        ;;
    claw)
        echo "[1/4] OpenClaw 会话导入 (全部 samples)..."
        python3 "$SCRIPT_DIR/eval.py" ingest "$DATA_FILE" \
            --base-url "$GATEWAY_URL" \
            --token "$GATEWAY_TOKEN" \
            --memory-mode openviking \
            $FORCE_INGEST \
            2>&1 | tee "$RESULT_DIR/import_claw.log"
        echo "导入完成，等待 120 秒让记忆稳定..."
        sleep 120
        ;;
    both)
        echo "[1/4] OV + OpenClaw 并行导入 (全部 samples)..."
        python3 "$SCRIPT_DIR/import_to_ov.py" \
            --input "$DATA_FILE" \
            --openviking-url "$OV_URL" \
            --no-user-agent-id \
            --success-csv "$RESULT_DIR/import_success.csv" \
            --error-log "$RESULT_DIR/import_errors.log" \
            $FORCE_INGEST \
            > "$RESULT_DIR/import_ov.log" 2>&1 &
        PID_OV=$!

        python3 "$SCRIPT_DIR/eval.py" ingest "$DATA_FILE" \
            --base-url "$GATEWAY_URL" \
            --token "$GATEWAY_TOKEN" \
            --memory-mode openviking \
            $FORCE_INGEST \
            > "$RESULT_DIR/import_claw.log" 2>&1 &
        PID_CLAW=$!

        echo "等待 OV 导入 (PID=$PID_OV) 和 OpenClaw 导入 (PID=$PID_CLAW) 完成..."
        wait $PID_OV $PID_CLAW
        echo "导入完成，等待 120 秒让记忆稳定..."
        sleep 120
        ;;
    skip)
        echo "[1/4] 跳过导入"
        ;;
    *)
        echo "错误: 无效的 import-mode '$IMPORT_MODE'，可选: ov|claw|both|skip"
        exit 1
        ;;
esac

# Step 2: QA 评测
echo "[2/4] 运行 QA 评估 (并发=$QA_PARALLEL)..."
python3 "$SCRIPT_DIR/eval.py" qa "$DATA_FILE" \
    --base-url "$GATEWAY_URL" \
    --token "$GATEWAY_TOKEN" \
    --ov-api-url "$OV_URL" \
    --parallel "$QA_PARALLEL" \
    --output "${OUTPUT_CSV%.csv}" \
    2>&1 | tee "$RESULT_DIR/qa.log"

# Step 3: Judge 评分
if [[ "$SKIP_JUDGE" == false ]]; then
    echo "[3/4] Judge 评分 (并发=$JUDGE_PARALLEL)..."
    python3 "$SCRIPT_DIR/judge.py" \
        --input "$OUTPUT_CSV" \
        --parallel "$JUDGE_PARALLEL" \
        2>&1 | tee "$RESULT_DIR/judge.log"
else
    echo "[3/4] 跳过 Judge 评分"
fi

# Step 4: 统计
if [[ "$SKIP_STAT" == false ]]; then
    echo "[4/4] 统计结果..."
    python3 "$SCRIPT_DIR/stat_judge_result.py" \
        --input "$OUTPUT_CSV" \
        --import-csv "$RESULT_DIR/import_success.csv" \
        --test-scenario "small" \
        2>&1 | tee "$RESULT_DIR/stat.log"
else
    echo "[4/4] 跳过统计"
fi

echo ""
echo "============================================"
echo "  评测完成: $DATASET_TAG"
echo "  结果目录: $RESULT_DIR"
echo "============================================"