#!/bin/bash
query='where is the actual train method implementation in the source code?'
filter='model.py'
echo "=== 开始 codebase 搜索测试 ==="
echo "搜索查询: $query"
echo "路径过滤器: $filter"
echo ""
# 执行搜索并处理结果
npx tsx src/cli.ts --demo --search="$query" --path-filters="$filter" --json 2>&1 | awk '/^{/ {found=1} found' | python3 -c "
import json
import sys
data = json.load(sys.stdin)
# 检查新的输出格式
if 'snippets' in data:
# 新格式: snippets 直接在根级别
all_snippets = []
for snippet in data.get('snippets', []):
snippet['source_file'] = snippet.get('filePath', 'N/A')
all_snippets.append(snippet)
print(f'检测到新格式输出')
print(f'总 snippets 数: {len(all_snippets)}')
print(f'总结果数: {data.get(\"totalResults\", \"N/A\")}')
print(f'总代码片段数: {data.get(\"totalSnippets\", \"N/A\")}')
print(f'重复移除数: {data.get(\"duplicatesRemoved\", \"N/A\")}')
elif 'files' in data:
# 旧格式: files 数组
all_snippets = []
for file_data in data.get('files', []):
file_path = file_data.get('filePath', 'N/A')
print(f'文件: {file_path}, 平均分数: {file_data.get(\"avgScore\", \"N/A\"):.3f}, 代码片段数: {file_data.get(\"snippetCount\", 0)}')
for snippet in file_data.get('snippets', []):
snippet['source_file'] = file_path
all_snippets.append(snippet)
print(f'\\n总 snippets 数: {len(all_snippets)}')
else:
print(f'错误: 无法识别的输出格式')
print(f'数据键: {list(data.keys())}')
sys.exit(1)
# 按分数降序排序
sorted_snippets = sorted(all_snippets, key=lambda x: x.get('score', 0), reverse=True)
print('\\n=== 按分数重新排序后的结果 ===')
# 1. 查找目标代码块在重新排序后的排名
target_found = False
for i, snippet in enumerate(sorted_snippets):
code = snippet.get('code', '')
if 'def train' in code and 'Model' in snippet.get('hierarchy', ''):
print(f'\\n1. 目标代码块按分数排序后的排名: 第 {i+1} 位')
print(f' 分数: {snippet.get(\"score\", \"N/A\")}')
print(f' 文件: {snippet.get(\"source_file\", \"N/A\")}')
print(f' 行范围: {snippet.get(\"lineRange\", \"N/A\")}')
print(f' 层级: {snippet.get(\"hierarchy\", \"N/A\")}')
target_found = True
break
if not target_found:
print('未找到目标代码块')
# 2. 查看前20个最高分的结果
print('\\n2. 前20个最高分的结果:')
for i, snippet in enumerate(sorted_snippets[:20]):
score = snippet.get('score', 0)
file_path = snippet.get('source_file', 'N/A')
line_range = snippet.get('lineRange', 'N/A')
hierarchy = snippet.get('hierarchy', 'N/A')
code_preview = snippet.get('code', '')[:50].replace('\n', ' ')
print(f' 第 {i+1:3d} 位: 分数={score:.3f}, 文件={file_path}')
print(f' 行范围={line_range}, 层级={hierarchy}')
print(f' 代码预览: {code_preview}...')
if i < 19:
print()
# 3. 查看分数分布
print('\\n3. 分数分布统计:')
scores = [s.get('score', 0) for s in sorted_snippets]
if scores:
print(f' 最高分: {max(scores):.3f}')
print(f' 最低分: {min(scores):.3f}')
print(f' 平均分: {sum(scores)/len(scores):.3f}')
print(f' 中位数: {sorted(scores)[len(scores)//2]:.3f}')
else:
print(' 没有分数数据')
# 4. 查看目标代码块周围的分数情况
if target_found:
target_score = None
for i, snippet in enumerate(sorted_snippets):
code = snippet.get('code', '')
if 'def train' in code and 'Model' in snippet.get('hierarchy', ''):
target_score = snippet.get('score', 0)
target_index = i
break
if target_score is not None:
print(f'\\n4. 目标代码块分数对比:')
print(f' 目标代码块分数: {target_score:.3f}')
print(f' 目标代码块文件: {sorted_snippets[target_index].get(\"source_file\", \"N/A\")}')
# 统计有多少个结果分数更高
higher_count = sum(1 for s in sorted_snippets if s.get('score', 0) > target_score)
same_count = sum(1 for s in sorted_snippets if s.get('score', 0) == target_score)
lower_count = sum(1 for s in sorted_snippets if s.get('score', 0) < target_score)
print(f' 分数更高的结果数: {higher_count}')
print(f' 分数相同的结果数: {same_count}')
print(f' 分数更低的结果数: {lower_count}')
if sorted_snippets:
print(f' 百分比位置: 前 {higher_count/len(sorted_snippets)*100:.1f}% 的结果分数更高')
# 5. 按文件统计结果
print('\\n5. 按文件统计:')
file_stats = {}
for snippet in all_snippets:
file_path = snippet.get('source_file', 'N/A')
if file_path not in file_stats:
file_stats[file_path] = {'count': 0, 'total_score': 0, 'max_score': 0}
file_stats[file_path]['count'] += 1
score = snippet.get('score', 0)
file_stats[file_path]['total_score'] += score
file_stats[file_path]['max_score'] = max(file_stats[file_path]['max_score'], score)
print(f' 文件总数: {len(file_stats)}')
for file_path, stats in sorted(file_stats.items(), key=lambda x: x[1]['count'], reverse=True)[:10]:
avg_score = stats['total_score'] / stats['count'] if stats['count'] > 0 else 0
print(f' {file_path}:')
print(f' 代码片段数: {stats[\"count\"]}, 平均分数: {avg_score:.3f}, 最高分数: {stats[\"max_score\"]:.3f}')
"
echo ""
echo "=== 测试完成 ==="
