dd4aa6b345
T-1.21 (P1-22): Dispatch → Worker 桥接 - eng_mode: spawn_workers() 为每个 ready 任务生成标准 Skill 调用参数 - commands/eng.md: dispatch 段从意图描述改为可执行5步伪代码 Agent 不再需要猜测工具名、参数格式、task-text 来源 T-1.22 (P1-24): Merge → TaskGraph 状态同步 - merge_worker_result Phase 6.5: 写回 task-graph.json 节点 status - merge 后节点状态 DONE，ready_tasks() 过滤已完成任务 _select_ready_tasks 修复： - DAG ready 为空时不再 fallback 到 todo.md（todo.md 状态陈旧时导致重复派发） P1-23: commands/eng.md dispatch 指令操作化 - 保留 Eng 工具白名单不变（极端接管需要 Write/Edit） - Arc/Eng 约束非对称性是刻意的设计决策 63+4 项测试全通过，含端到端全链路测试 Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
272 lines
12 KiB
Python
272 lines
12 KiB
Python
#!/usr/bin/env python3
|
||
"""T-1.21 + T-1.22 端到端测试：arc → eng → do → merge → task-graph sync 全链路"""
|
||
|
||
import sys
|
||
import tempfile
|
||
import json
|
||
from pathlib import Path
|
||
|
||
sys.path.insert(0, str(Path(__file__).parent / "lib"))
|
||
|
||
|
||
def test_spawn_workers():
|
||
"""测试 spawn_workers 返回正确的 Skill 调用参数"""
|
||
from air_runtime.task_graph import TaskGraph, TaskNode
|
||
from air_runtime.modes.arc_mode import _export_task_graph_json
|
||
from air_runtime.modes.eng_mode import spawn_workers
|
||
|
||
with tempfile.TemporaryDirectory() as tmpdir:
|
||
root = Path(tmpdir)
|
||
arc_dir = root / "AirPlan" / "state" / "airarc" / "reviews"
|
||
arc_dir.mkdir(parents=True, exist_ok=True)
|
||
|
||
graph = TaskGraph()
|
||
graph.add_node(TaskNode(id="G-001", status="TODO", task="实现视频解码器",
|
||
files_dirs="src/decoder.cpp"))
|
||
graph.add_node(TaskNode(id="G-002", status="TODO", task="实现音频解码器",
|
||
files_dirs="src/audio.cpp"))
|
||
_export_task_graph_json(graph, arc_dir / "task-graph.json")
|
||
|
||
instructions = spawn_workers(root, ["G-001", "G-002"])
|
||
assert len(instructions) == 2
|
||
assert instructions[0]["skill"] == "airplan"
|
||
assert "do --sub enter --task-id G-001" in instructions[0]["args"]
|
||
assert "实现视频解码器" in instructions[0]["args"]
|
||
assert instructions[1]["taskId"] == "G-002"
|
||
|
||
print("✓ spawn_workers 测试通过")
|
||
return True
|
||
|
||
|
||
def test_merge_syncs_task_graph():
|
||
"""P1-24: 测试 merge 后 task-graph.json 节点状态同步"""
|
||
from air_runtime.task_graph import TaskGraph, TaskNode
|
||
from air_runtime.modes.arc_mode import _export_task_graph_json, ArcPhaseGate
|
||
from air_runtime.modes.eng_mode import enter_engine, merge_worker_result
|
||
from air_runtime.modes.do_mode import enter_worker, finish_worker
|
||
from air_runtime.contracts import WorkerResult
|
||
|
||
with tempfile.TemporaryDirectory() as tmpdir:
|
||
root = Path(tmpdir)
|
||
|
||
# Arc 准备
|
||
from air_runtime.modes.arc_mode import enter_mode, parallel_review_mode
|
||
enter_mode(root)
|
||
gate = ArcPhaseGate(root / "AirPlan" / "state" / "airarc" / "state.json")
|
||
gate.advance_to("confirmed")
|
||
|
||
todo = root / "AirPlan" / "todo.md"
|
||
todo.parent.mkdir(parents=True, exist_ok=True)
|
||
todo.write_text("| Task | Status | Files/Dirs | Done When |\n"
|
||
"|------|--------|------------|-----------|\n"
|
||
"| G-001 解码器 | TODO | src/decoder.cpp | 编译通过 |\n")
|
||
|
||
parallel_review_mode(root, todo)
|
||
|
||
# Eng 初始化
|
||
enter_engine(root)
|
||
|
||
# 验证 merge 前 task-graph 中 G-001 是 TODO
|
||
tg_json = root / "AirPlan" / "state" / "airarc" / "reviews" / "task-graph.json"
|
||
graph = TaskGraph.load(tg_json)
|
||
assert "G-001" in graph.nodes
|
||
assert graph.nodes["G-001"].status == "TODO"
|
||
print(f" merge 前: G-001 status = {graph.nodes['G-001'].status}")
|
||
|
||
# Do Worker 执行
|
||
w = enter_worker(root, "G-001", task_text="G-001 解码器")
|
||
wr = WorkerResult(task_id="G-001", status="done", summary="完成",
|
||
files_changed=["src/decoder.cpp"], validations="编译通过")
|
||
rpath = root / "AirPlan" / "state" / "airdo" / "tasks" / "G-001" / "result.json"
|
||
rpath.parent.mkdir(parents=True, exist_ok=True)
|
||
rpath.write_text(json.dumps(wr.to_dict()))
|
||
|
||
f = finish_worker(root, "G-001", rpath)
|
||
print(f" Do finish: status={f['status']}, route={f['routingDecision']['target']}")
|
||
|
||
# Eng merge
|
||
m = merge_worker_result(root, rpath)
|
||
print(f" Eng merge: task={m['taskId']}, status={m['status']}")
|
||
|
||
# 验证 merge 后 task-graph 中 G-001 是 DONE
|
||
graph_after = TaskGraph.load(tg_json)
|
||
assert graph_after.nodes["G-001"].status == "DONE", \
|
||
f"merge 后 G-001 应为 DONE，实际: {graph_after.nodes['G-001'].status}"
|
||
print(f" merge 后: G-001 status = {graph_after.nodes['G-001'].status}")
|
||
|
||
# 验证不会重复派发（ready_tasks 不包含已完成的 G-001）
|
||
ready = graph_after.ready_tasks()
|
||
assert "G-001" not in ready, f"G-001 不应在 ready 队列中: {ready}"
|
||
print(f" ready queue: {ready}")
|
||
|
||
print("✓ merge 同步 task-graph 状态测试通过")
|
||
return True
|
||
|
||
|
||
def test_full_pipeline_e2e():
|
||
"""完整端到端流水线：arc → eng plan → dispatch → do → merge → task-graph sync"""
|
||
from air_runtime.task_graph import TaskGraph
|
||
from air_runtime.modes.arc_mode import enter_mode, parallel_review_mode, ArcPhaseGate, _export_task_graph_json
|
||
from air_runtime.modes.eng_mode import (enter_engine, build_engine_plan,
|
||
dispatch_worker_group, spawn_workers,
|
||
merge_worker_result)
|
||
from air_runtime.modes.do_mode import enter_worker, finish_worker
|
||
from air_runtime.contracts import WorkerResult
|
||
|
||
with tempfile.TemporaryDirectory() as tmpdir:
|
||
root = Path(tmpdir)
|
||
|
||
# ============ ARC ============
|
||
enter_mode(root)
|
||
gate = ArcPhaseGate(root / "AirPlan" / "state" / "airarc" / "state.json")
|
||
gate.advance_to("confirmed")
|
||
|
||
todo = root / "AirPlan" / "todo.md"
|
||
todo.parent.mkdir(parents=True, exist_ok=True)
|
||
todo.write_text("| Task | Status | Files/Dirs | Done When |\n"
|
||
"|------|--------|------------|-----------|\n"
|
||
"| H-000 前置 | DONE | | 完成 |\n"
|
||
"| G-001 解码器 | TODO | src/decoder.cpp | 编译通过 |\n"
|
||
"| G-002 编码器 | TODO | src/encoder.cpp | 编译通过 |\n")
|
||
|
||
r = parallel_review_mode(root, todo)
|
||
assert not r.get("blocked"), f"Arc blocked: {r.get('reason')}"
|
||
print("1. ARC parallel-review ✓")
|
||
|
||
# ============ ENG PLAN ============
|
||
enter_engine(root)
|
||
plan = build_engine_plan(root, todo)
|
||
assert plan["planningSource"] == "airarc-execution-plan"
|
||
assert "G-001" in plan["selectedTasks"]
|
||
assert "G-002" in plan["selectedTasks"]
|
||
print(f"2. ENG plan: tasks={plan['selectedTasks']}")
|
||
|
||
# ============ ENG DISPATCH ============
|
||
dispatch = dispatch_worker_group(root)
|
||
assert not dispatch.get("blocked"), f"dispatch blocked: {dispatch.get('reason')}"
|
||
assert len(dispatch["taskIds"]) >= 2
|
||
print(f"3. ENG dispatch: wave={dispatch['waveId']}, tasks={dispatch['taskIds']}")
|
||
|
||
# ============ SPAWN INSTRUCTIONS ============
|
||
instructions = spawn_workers(root, dispatch["taskIds"])
|
||
assert len(instructions) >= 2
|
||
for inst in instructions:
|
||
assert inst["skill"] == "airplan"
|
||
assert "do --sub enter --task-id" in inst["args"]
|
||
assert inst["taskText"] # 不能为空
|
||
print(f"4. spawn_workers: {len(instructions)} 个 Worker 指令准备完成")
|
||
for inst in instructions:
|
||
print(f" - {inst['taskId']}: {inst['taskText'][:40]}")
|
||
|
||
# ============ DO WORKER (模拟 Agent 调 Skill) ============
|
||
for inst in instructions:
|
||
tid = inst["taskId"]
|
||
task_text = inst["taskText"]
|
||
w = enter_worker(root, tid, task_text=task_text)
|
||
wr = WorkerResult(task_id=tid, status="done", summary="完成",
|
||
files_changed=[f"src/{tid.lower()}.cpp"],
|
||
validations="编译通过")
|
||
rpath = root / "AirPlan" / "state" / "airdo" / "tasks" / tid / "result.json"
|
||
rpath.parent.mkdir(parents=True, exist_ok=True)
|
||
rpath.write_text(json.dumps(wr.to_dict()))
|
||
f = finish_worker(root, tid, rpath)
|
||
assert f["routingDecision"]["target"] == "merge"
|
||
print("5. DO Workers 全部完成 (G-001, G-002)")
|
||
|
||
# ============ ENG MERGE ============
|
||
for inst in instructions:
|
||
tid = inst["taskId"]
|
||
rpath = root / "AirPlan" / "state" / "airdo" / "tasks" / tid / "result.json"
|
||
m = merge_worker_result(root, rpath)
|
||
assert m["status"] == "done"
|
||
print("6. ENG merge 全部完成")
|
||
|
||
# ============ VERIFY TASK-GRAPH SYNC ============
|
||
tg_json = root / "AirPlan" / "state" / "airarc" / "reviews" / "task-graph.json"
|
||
graph_after = TaskGraph.load(tg_json)
|
||
assert graph_after.nodes["G-001"].status == "DONE", \
|
||
f"G-001 应为 DONE，实际: {graph_after.nodes['G-001'].status}"
|
||
assert graph_after.nodes["G-002"].status == "DONE", \
|
||
f"G-002 应为 DONE，实际: {graph_after.nodes['G-002'].status}"
|
||
print(f"7. task-graph sync: G-001={graph_after.nodes['G-001'].status}, "
|
||
f"G-002={graph_after.nodes['G-002'].status}")
|
||
|
||
# ============ VERIFY NO RE-DISPATCH ============
|
||
ready_after = graph_after.ready_tasks()
|
||
assert "G-001" not in ready_after
|
||
assert "G-002" not in ready_after
|
||
# H-000 是 DONE，也不在 ready
|
||
# 只有 T-TEST-* 还在 TODO
|
||
print(f"8. 无重复派发: ready={ready_after}")
|
||
# 再次 dispatch 不应包含已完成任务
|
||
dispatch2 = dispatch_worker_group(root)
|
||
for done_tid in ["G-001", "G-002", "H-000"]:
|
||
assert done_tid not in dispatch2.get("taskIds", []), \
|
||
f"已完成任务 {done_tid} 不应再次被派发"
|
||
|
||
print(f"9. 二次 dispatch: tasks={dispatch2.get('taskIds', [])} (无已完成任务)")
|
||
print("✓ 完整端到端流水线测试通过")
|
||
return True
|
||
|
||
|
||
def test_dispatch_no_workers_means_no_dispatch():
|
||
"""测试所有任务完成后 dispatch 返回空"""
|
||
from air_runtime.task_graph import TaskGraph, TaskNode
|
||
from air_runtime.modes.arc_mode import _export_task_graph_json
|
||
from air_runtime.modes.eng_mode import dispatch_worker_group, enter_engine
|
||
from air_runtime.modes.arc_mode import enter_mode
|
||
|
||
with tempfile.TemporaryDirectory() as tmpdir:
|
||
root = Path(tmpdir)
|
||
enter_mode(root)
|
||
enter_engine(root)
|
||
|
||
# 所有任务都是 DONE
|
||
arc_dir = root / "AirPlan" / "state" / "airarc" / "reviews"
|
||
arc_dir.mkdir(parents=True, exist_ok=True)
|
||
graph = TaskGraph()
|
||
graph.add_node(TaskNode(id="G-001", status="DONE", task="已完成"))
|
||
graph.add_node(TaskNode(id="G-002", status="DONE", task="已完成"))
|
||
_export_task_graph_json(graph, arc_dir / "task-graph.json")
|
||
|
||
result = dispatch_worker_group(root)
|
||
assert result["taskIds"] == [], f"全部完成时应为空，实际: {result['taskIds']}"
|
||
print("✓ 全部完成时 dispatch 返回空测试通过")
|
||
return True
|
||
|
||
|
||
def main():
|
||
print("=" * 50)
|
||
print("T-1.21 + T-1.22 端到端全链路测试")
|
||
print("=" * 50)
|
||
|
||
tests = [
|
||
("spawn_workers", test_spawn_workers),
|
||
("merge 同步 task-graph", test_merge_syncs_task_graph),
|
||
("完整端到端流水线", test_full_pipeline_e2e),
|
||
("全完成时 dispatch 返回空", test_dispatch_no_workers_means_no_dispatch),
|
||
]
|
||
|
||
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)
|