6130478c96
P0-8 扩大: do_mode.py finish_worker 全专家插件强制路由 - GUI→XDB, network→NDB, C/C++→SDB, done→Rvr, blocked/failed→Dbg - 证据去重: 已有 xdbSessions/ndbSessions/sdbReports/rvrReviewed 则跳过 P1-GAP17: 事件 emit 规范化 - 新增 7 个事件常量 (TASK_ENTERED, TASK_FINISHED, ENGINE_ENTERED 等) - 全部 emit 调用替换字符串字面量为常量，零残留 - 30 个事件类型常量全部定义且唯一 P1-GAP18: 事件日志原子轮转 - emit 计数器每 128 次检查轮转，避免每次 emit 读文件 - 清除未使用的 _emit_with_completion/_pending_merge_complete - 原子轮转: tempfile+os.replace 保证不损坏 eng 极端接管: 强制调用全部专家插件 (Dbg/XDB/NDB/SDB/Rvr) commands/do.md: 更新为全专家插件路由文档 全量测试: 69 通过, 0 失败 Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
248 lines
7.7 KiB
Python
Executable File
248 lines
7.7 KiB
Python
Executable File
#!/usr/bin/env python3
|
||
"""Ctx Mode 功能测试"""
|
||
|
||
import sys
|
||
import os
|
||
import tempfile
|
||
import json
|
||
from pathlib import Path
|
||
|
||
sys.path.insert(0, str(Path(__file__).parent / "lib"))
|
||
|
||
def test_ctx_enter():
|
||
"""测试上下文管理器初始化"""
|
||
from air_runtime.modes.ctx_mode import ctx_enter
|
||
|
||
with tempfile.TemporaryDirectory() as tmpdir:
|
||
project_root = Path(tmpdir)
|
||
|
||
result = ctx_enter(project_root)
|
||
|
||
assert "state_path" in result
|
||
assert Path(result["state_path"]).exists()
|
||
|
||
print("✓ 上下文管理器初始化测试通过")
|
||
return True
|
||
|
||
|
||
def test_ctx_estimate_tokens():
|
||
"""测试 Token 估算"""
|
||
from air_runtime.modes.ctx_mode import estimate_tokens
|
||
|
||
# 测试中文
|
||
chinese_text = "这是一个中文测试文本用于估算 Token 数量"
|
||
tokens = estimate_tokens(chinese_text)
|
||
assert tokens > 0
|
||
|
||
# 测试英文
|
||
english_text = "This is an English text for token estimation"
|
||
tokens_en = estimate_tokens(english_text)
|
||
assert tokens_en > 0
|
||
|
||
# 测试代码
|
||
code_text = "def main():\n print('hello world')\n return 0"
|
||
tokens_code = estimate_tokens(code_text)
|
||
assert tokens_code > 0
|
||
|
||
print(f"✓ Token 估算测试通过，中文: {tokens}, 英文: {tokens_en}, 代码: {tokens_code}")
|
||
return True
|
||
|
||
|
||
def test_ctx_validate_compression():
|
||
"""测试压缩质量校验"""
|
||
from air_runtime.modes.ctx_mode import validate_compression
|
||
|
||
original = """
|
||
# ADR-0001 决策
|
||
我们选择使用 PostgreSQL 数据库
|
||
|
||
# TODO 实现
|
||
- [ ] 创建数据库连接池
|
||
- [ ] 实现 CRUD 接口
|
||
|
||
# INV-1 约束
|
||
必须使用原子写入
|
||
"""
|
||
|
||
# 好的摘要（包含关键信息）
|
||
good_summary = "ADR-0001: PostgreSQL, TODO: 连接池/CRUD, INV-1: 原子写入"
|
||
result = validate_compression(original, good_summary)
|
||
assert result["ok"] == True
|
||
|
||
# 坏的摘要（丢失关键信息）
|
||
bad_summary = "选择了 PostgreSQL"
|
||
result_bad = validate_compression(original, bad_summary)
|
||
# 可能不 ok，因为丢失了 TODO 和 INV
|
||
|
||
print("✓ 压缩质量校验测试通过")
|
||
return True
|
||
|
||
|
||
def test_ctx_compress_fallback():
|
||
"""测试三级降级压缩"""
|
||
from air_runtime.modes.ctx_mode import compress_with_fallback, CompressionLevel
|
||
|
||
# 创建大文本
|
||
long_text = "这是测试内容\n" * 1000
|
||
|
||
result = compress_with_fallback(long_text, max_tokens=100)
|
||
|
||
# 应该返回结果
|
||
assert "level" in result
|
||
assert "result" in result
|
||
assert result["level"] in [CompressionLevel.RETRY, CompressionLevel.FALLBACK_MODEL, CompressionLevel.TRUNCATE]
|
||
|
||
print(f"✓ 三级降级压缩测试通过，压缩级别: {result['level']}")
|
||
return True
|
||
|
||
|
||
def test_ctx_lock():
|
||
"""测试 compactor 锁"""
|
||
from air_runtime.modes.ctx_mode import acquire_compactor_lock
|
||
|
||
with tempfile.TemporaryDirectory() as tmpdir:
|
||
lock_path = Path(tmpdir) / "test.lock"
|
||
|
||
# 首次获取锁应该成功
|
||
acquired = acquire_compactor_lock(lock_path)
|
||
assert acquired == True
|
||
assert lock_path.exists()
|
||
|
||
# 再次获取应该失败（锁已被持有）
|
||
acquired2 = acquire_compactor_lock(lock_path)
|
||
assert acquired2 == False
|
||
|
||
print("✓ Compactor 锁测试通过")
|
||
return True
|
||
|
||
|
||
def test_event_log_emit():
|
||
"""P1-GAP17: 测试事件日志 emit 和读取"""
|
||
from air_runtime.events import EventLog, TASK_COMPLETED, TASK_ENTERED, TASK_FINISHED
|
||
|
||
with tempfile.TemporaryDirectory() as tmpdir:
|
||
log_path = Path(tmpdir) / "events.jsonl"
|
||
log = EventLog(log_path)
|
||
|
||
log.emit(TASK_ENTERED, {"taskId": "T-001"})
|
||
log.emit(TASK_FINISHED, {"taskId": "T-001"})
|
||
log.emit(TASK_COMPLETED, {"taskId": "T-001", "status": "done"})
|
||
|
||
assert log_path.exists()
|
||
lines = log_path.read_text().strip().split("\n")
|
||
assert len(lines) == 3
|
||
|
||
entries = [json.loads(line) for line in lines]
|
||
assert entries[0]["type"] == "task.entered"
|
||
assert entries[1]["type"] == "task.finished"
|
||
assert entries[2]["type"] == "task.completed"
|
||
|
||
# 检查 ts 字段均为 ISO 格式
|
||
for e in entries:
|
||
assert "ts" in e
|
||
assert "T" in e["ts"]
|
||
|
||
print("✓ 事件日志 emit 测试通过")
|
||
return True
|
||
|
||
|
||
def test_event_log_rotate():
|
||
"""P1-GAP18: 测试事件日志原子轮转"""
|
||
from air_runtime.events import EventLog, TASK_COMPLETED
|
||
|
||
with tempfile.TemporaryDirectory() as tmpdir:
|
||
log_path = Path(tmpdir) / "events.jsonl"
|
||
# 小上限 + 密集写入触发轮转
|
||
log = EventLog(log_path, max_lines=50)
|
||
# 改写 ROTATE_CHECK_EVERY 为小值确保多次检查
|
||
log.ROTATE_CHECK_EVERY = 10
|
||
|
||
# 写入 200 条，因 ROTATE_CHECK_EVERY=10 会多次触发轮转
|
||
for i in range(200):
|
||
log.emit(TASK_COMPLETED, {"taskId": f"T-{i:03d}"})
|
||
|
||
lines = log_path.read_text().strip().split("\n")
|
||
# 经过多次轮转后，行数不应超过 max_lines * 2（两倍内都合理）
|
||
assert len(lines) < 200, f"轮转后应少于写入量 200，实际: {len(lines)}"
|
||
assert len(lines) >= 1, f"轮转后应有数据, 实际: {len(lines)}"
|
||
|
||
# 验证写入操作后文件仍然有效 JSONL
|
||
for line in lines:
|
||
entry = json.loads(line)
|
||
assert entry["type"] == "task.completed"
|
||
assert "ts" in entry
|
||
|
||
print(f"✓ 事件日志原子轮转测试通过，轮转后保留 {len(lines)} 条")
|
||
return True
|
||
|
||
|
||
def test_event_constants_defined():
|
||
"""P1-GAP17: 验证所有事件类型常量均已定义且唯一"""
|
||
import air_runtime.events as evt
|
||
|
||
expected_names = [
|
||
"TASK_DISPATCHED", "TASK_ENTERED", "TASK_FINISHED",
|
||
"TASK_COMPLETED", "TASK_BLOCKED",
|
||
"MERGE_STARTED", "MERGE_COMPLETED",
|
||
"REPAIR_CREATED", "REPAIR_RESOLVED",
|
||
"INTERVENTION_STALL", "XDB_CAPTURED", "DEBUG_SESSION",
|
||
"CONTEXT_COMPACTED", "DEPLOY_COMPLETED", "TEST_RUN", "SEC_SCAN", "REVIEW_SESSION",
|
||
"ENGINE_ENTERED", "ENGINE_CYCLE", "ENG_REPLAN_TRIGGERED", "ENG_BLOCKED",
|
||
"WORKER_TIMEOUT", "WORKTREE_MERGE_CONFLICT",
|
||
"ARC_REPLANNED",
|
||
"ADR_CHANGE_DETECTED", "ADR_INVALIDATION", "ADR_UNFREEZED",
|
||
"LOCK_ACQUIRED", "LOCK_RELEASED", "STALE_LOCK_CLEANED",
|
||
]
|
||
|
||
values = []
|
||
for name in expected_names:
|
||
val = getattr(evt, name, None)
|
||
assert val is not None, f"缺少事件常量: {name}"
|
||
values.append(val)
|
||
|
||
# 无重复值
|
||
assert len(set(values)) == len(values), f"事件常量存在重复值: {len(set(values))} vs {len(values)}"
|
||
|
||
print(f"✓ 事件常量定义测试通过，共 {len(values)} 个事件类型")
|
||
return True
|
||
|
||
|
||
def main():
|
||
print("=" * 50)
|
||
print("Ctx Mode 功能测试")
|
||
print("=" * 50)
|
||
|
||
tests = [
|
||
("上下文管理器初始化", test_ctx_enter),
|
||
("Token 估算", test_ctx_estimate_tokens),
|
||
("压缩质量校验", test_ctx_validate_compression),
|
||
("三级降级压缩", test_ctx_compress_fallback),
|
||
("Compactor 锁", test_ctx_lock),
|
||
("事件日志 emit", test_event_log_emit),
|
||
("事件日志原子轮转", test_event_log_rotate),
|
||
("事件常量定义", test_event_constants_defined),
|
||
]
|
||
|
||
passed = 0
|
||
failed = 0
|
||
|
||
for name, test_fn in tests:
|
||
try:
|
||
test_fn()
|
||
passed += 1
|
||
except Exception as e:
|
||
print(f"✗ {name} 失败: {e}")
|
||
import traceback
|
||
traceback.print_exc()
|
||
failed += 1
|
||
|
||
print("=" * 50)
|
||
print(f"测试结果: {passed} 通过, {failed} 失败")
|
||
print("=" * 50)
|
||
|
||
return failed == 0
|
||
|
||
|
||
if __name__ == "__main__":
|
||
success = main()
|
||
sys.exit(0 if success else 1) |