gjm
e8140f2bc5
发现的缺口：maximized_drawing_area 原先只包住取证循环，审查阶段没被包住 → LLM 在受控子循环里 zoom + 截图拿到的是未最大化的 2110x462（正是上一轮刚解决的 低分辨率问题，在最需要放大看细节的审查阶段又出现了）；且审查中的 zoom 不复位视口， 跑完把用户的 CAD 留在最后一次放大处。 改动： 1. __main__.run_review：把 maximized_drawing_area 从「只包取证」扩到「包取证 + 审查」 （两者都要截图）；report/annotate 留在 with 之外。 2. phases/review.py：新增 _VIEW_TOOLS；review_frame 拆成外壳 + _review_frame， 外壳 try/finally 里仅当确实动过视口才复位到该图框整幅视图 （zoom_to_window(win, margin=0.02)）。纯查结构化数据的路径不受影响。 3. checklist.build_system_prompt：把「可调用只读工具」改写成明确的放大取证协议—— 截图看不清时用 zoom_to_window（坐标由图框/实体 bbox 算）或 zoom_to_entity， 再 get_viewport_screenshot 复核；并约束不要为看图而反复放大。 次生问题（同轮修复）：_MAX_OBS_IMAGES 原为硬编码 3，而模型一轮内会发 8 次截图 → 拍了 8 张只回灌 3 张。改为配置项 REVIEW_OBS_IMAGES（默认 6），并把预算耗尽的占位 文案改成明确告知模型「不要再请求截图，请给出结论」。 实测（真实图纸 1 图框，新增 verify_review_maximized.py）：取证与审查两阶段窗口均为 [-4,-2,3444,1442]（最大化态），结束精确还原 [68,0,2316,960]，三项全 PASS； 子循环里 LLM 连发 8 组 zoom_to_window + get_viewport_screenshot，提示词生效。 pytest 304 passed。
342 lines
13 KiB
Python
342 lines
13 KiB
Python
"""`dwg_review.phases.review` 单元测试。"""
|
||
|
||
import json
|
||
|
||
from dwg_review.cad_tools import READ_ONLY_TOOLS
|
||
from dwg_review.mcp_client import McpToolError
|
||
from dwg_review.phases.review import (
|
||
_run_readonly_tool,
|
||
evidence_payload,
|
||
review_frame,
|
||
)
|
||
from tests.fakes import (
|
||
FakeLlmClient,
|
||
FakeMcpClient,
|
||
llm_calls,
|
||
llm_text,
|
||
make_ctx,
|
||
make_evidence,
|
||
ok_image,
|
||
ok_mixed,
|
||
ok_text,
|
||
)
|
||
|
||
|
||
# ---------------------------------------------------------------------------
|
||
# evidence_payload
|
||
# ---------------------------------------------------------------------------
|
||
def test_evidence_payload_strips_image():
|
||
ev = make_evidence(image_b64="QUJD", image_mime="image/png")
|
||
payload = evidence_payload(ev)
|
||
assert "images" not in payload
|
||
assert "image_b64" not in payload
|
||
assert payload["frame_index"] == 1
|
||
|
||
|
||
def test_evidence_payload_keeps_only_frame_metadata():
|
||
ev = make_evidence()
|
||
ev["frame"]["secret"] = "should-not-leak"
|
||
payload = evidence_payload(ev)
|
||
assert "secret" not in payload["frame"]
|
||
assert set(payload["frame"]) == {"sheet_size", "kind", "layer", "bbox", "confidence"}
|
||
|
||
|
||
def test_evidence_payload_carries_structured_data():
|
||
ev = make_evidence(entities={"total": 9, "items": []}, texts={"total": 2})
|
||
payload = evidence_payload(ev)
|
||
assert payload["entities"] == {"total": 9, "items": []}
|
||
assert payload["texts"] == {"total": 2}
|
||
|
||
|
||
# ---------------------------------------------------------------------------
|
||
# _run_readonly_tool
|
||
# ---------------------------------------------------------------------------
|
||
def test_readonly_rejects_non_whitelisted_tool():
|
||
client = FakeMcpClient()
|
||
ctx = make_ctx(client=client)
|
||
content, budget = _run_readonly_tool(ctx, "get_some_unknown_tool", {}, [3])
|
||
assert "不在只读取证白名单内" in content[0]["text"]
|
||
assert client.calls == []
|
||
assert budget == [3]
|
||
|
||
|
||
def test_readonly_rejects_write_tool():
|
||
client = FakeMcpClient()
|
||
ctx = make_ctx(client=client)
|
||
content, _ = _run_readonly_tool(ctx, "add_leader_annotation", {}, [3])
|
||
assert "不在只读取证白名单内" in content[0]["text"]
|
||
assert client.calls == []
|
||
|
||
|
||
def test_readonly_whitelist_covers_expected_tools():
|
||
assert {"query_entities", "get_entity_detail", "zoom_to_window"} <= READ_ONLY_TOOLS
|
||
|
||
|
||
def test_readonly_handles_tool_error():
|
||
client = FakeMcpClient(responses={"query_entities": McpToolError("炸了")})
|
||
ctx = make_ctx(client=client)
|
||
content, _ = _run_readonly_tool(ctx, "query_entities", {"window": {}}, [3])
|
||
assert "执行失败" in content[0]["text"]
|
||
|
||
|
||
def test_readonly_text_and_image_content():
|
||
client = FakeMcpClient(
|
||
responses={"query_entities": ok_mixed(["hi"], [("QUJD", "image/png")])}
|
||
)
|
||
ctx = make_ctx(client=client)
|
||
content, budget = _run_readonly_tool(ctx, "query_entities", {}, [3])
|
||
assert content[0] == {"type": "text", "text": "hi"}
|
||
assert content[1]["type"] == "image_url"
|
||
assert budget == [2]
|
||
|
||
|
||
def test_readonly_image_budget_exhausted():
|
||
client = FakeMcpClient(
|
||
responses={"query_entities": ok_mixed(["t"], [("QUJD", "image/png")])}
|
||
)
|
||
ctx = make_ctx(client=client)
|
||
content, budget = _run_readonly_tool(ctx, "query_entities", {}, [0])
|
||
assert budget == [0]
|
||
assert any("已省略" in c.get("text", "") for c in content)
|
||
|
||
|
||
def test_readonly_empty_response_marker():
|
||
empty = {"jsonrpc": "2.0", "id": 1, "result": {"content": []}}
|
||
client = FakeMcpClient(responses={"query_entities": empty})
|
||
ctx = make_ctx(client=client)
|
||
content, _ = _run_readonly_tool(ctx, "query_entities", {}, [3])
|
||
assert content == [{"type": "text", "text": "（无返回内容）"}]
|
||
|
||
|
||
# ---------------------------------------------------------------------------
|
||
# review_frame
|
||
# ---------------------------------------------------------------------------
|
||
def test_review_frame_returns_summary_when_no_tools():
|
||
llm = FakeLlmClient(script=[llm_text("结论A")])
|
||
ctx = make_ctx(client=FakeMcpClient(), llm=llm)
|
||
out = review_frame(ctx, make_evidence(), "系统提示", log=lambda *a: None)
|
||
assert out["summary"] == "结论A" # 非 JSON → 降级为纯文本 summary
|
||
assert out["issues"] == []
|
||
assert len(llm.calls) == 1
|
||
assert llm.calls[0]["tools"] is not None # 传了只读工具 schema
|
||
|
||
|
||
def test_review_frame_includes_image_block_when_present():
|
||
llm = FakeLlmClient(script=[llm_text("ok")])
|
||
ctx = make_ctx(client=FakeMcpClient(), llm=llm)
|
||
ev = make_evidence(image_b64="QUJD", image_mime="image/png")
|
||
review_frame(ctx, ev, "sys", log=lambda *a: None)
|
||
user_content = llm.calls[0]["messages"][1]["content"]
|
||
assert any(b.get("type") == "image_url" for b in user_content)
|
||
|
||
|
||
def test_review_frame_omits_image_block_when_absent():
|
||
llm = FakeLlmClient(script=[llm_text("ok")])
|
||
ctx = make_ctx(client=FakeMcpClient(), llm=llm)
|
||
review_frame(ctx, make_evidence(), "sys", log=lambda *a: None)
|
||
user_content = llm.calls[0]["messages"][1]["content"]
|
||
assert all(b.get("type") == "text" for b in user_content)
|
||
|
||
|
||
def test_review_frame_runs_tool_then_concludes():
|
||
llm = FakeLlmClient(
|
||
script=[
|
||
llm_calls([("query_entities", {"window": {}})]),
|
||
llm_text("结论B"),
|
||
]
|
||
)
|
||
client = FakeMcpClient(responses={"query_entities": ok_text({"total": 1})})
|
||
ctx = make_ctx(client=client, llm=llm)
|
||
out = review_frame(ctx, make_evidence(), "sys", log=lambda *a: None)
|
||
assert out["summary"] == "结论B"
|
||
assert len(llm.calls) == 2
|
||
assert client.called_names() == ["query_entities"]
|
||
|
||
|
||
def test_review_frame_feeds_observation_back_as_tool_message():
|
||
llm = FakeLlmClient(script=[llm_calls([("query_entities", {})]), llm_text("done")])
|
||
ctx = make_ctx(client=FakeMcpClient(), llm=llm)
|
||
review_frame(ctx, make_evidence(), "sys", log=lambda *a: None)
|
||
tool_msgs = [m for m in llm.calls[1]["messages"] if isinstance(m, dict) and m.get("role") == "tool"]
|
||
assert tool_msgs and tool_msgs[0]["tool_call_id"] == "1"
|
||
|
||
|
||
def test_review_frame_invalid_tool_arguments_become_empty():
|
||
llm = FakeLlmClient(
|
||
script=[llm_calls([("query_entities", "not-json")]), llm_text("done")]
|
||
)
|
||
client = FakeMcpClient()
|
||
ctx = make_ctx(client=client, llm=llm)
|
||
review_frame(ctx, make_evidence(), "sys", log=lambda *a: None)
|
||
assert client.arguments_of("query_entities") == {}
|
||
|
||
|
||
def test_review_frame_force_closes_at_iteration_limit():
|
||
llm = FakeLlmClient(
|
||
script=[
|
||
llm_calls([("query_entities", {})]),
|
||
llm_text("收口结论"),
|
||
]
|
||
)
|
||
ctx = make_ctx(client=FakeMcpClient(), llm=llm, max_iters=1)
|
||
out = review_frame(ctx, make_evidence(), "sys", log=lambda *a: None)
|
||
assert out["summary"] == "收口结论"
|
||
last_msg = llm.calls[-1]["messages"][-1]
|
||
assert last_msg["role"] == "user"
|
||
assert "上限" in last_msg["content"]
|
||
|
||
|
||
def test_review_frame_with_zero_iters_forces_immediate_close():
|
||
llm = FakeLlmClient(script=[llm_text("直接结论")])
|
||
ctx = make_ctx(client=FakeMcpClient(), llm=llm, max_iters=0)
|
||
out = review_frame(ctx, make_evidence(), "sys", log=lambda *a: None)
|
||
assert out["summary"] == "直接结论"
|
||
assert len(llm.calls) == 1
|
||
|
||
|
||
def test_review_frame_logs_tool_usage():
|
||
logs = []
|
||
llm = FakeLlmClient(script=[llm_calls([("get_layers", {})]), llm_text("x")])
|
||
ctx = make_ctx(client=FakeMcpClient(), llm=llm)
|
||
review_frame(ctx, make_evidence(), "sys", log=logs.append)
|
||
assert any("取证" in line and "get_layers" in line for line in logs)
|
||
|
||
|
||
def test_review_frame_payload_is_json_serializable():
|
||
llm = FakeLlmClient(script=[llm_text("ok")])
|
||
ctx = make_ctx(client=FakeMcpClient(), llm=llm)
|
||
review_frame(ctx, make_evidence(), "sys", log=lambda *a: None)
|
||
text = llm.calls[0]["messages"][1]["content"][0]["text"]
|
||
json.loads(text.split("：\n", 1)[1]) # 不抛异常即通过
|
||
|
||
|
||
def test_review_frame_parses_structured_json_conclusion():
|
||
text = (
|
||
'{"summary": "整体尚可", "issues": ['
|
||
'{"category": "几何", "severity": "medium", "message": "零长度", "handle": "A1"}]}'
|
||
)
|
||
llm = FakeLlmClient(script=[llm_text(text)])
|
||
ctx = make_ctx(client=FakeMcpClient(), llm=llm)
|
||
out = review_frame(ctx, make_evidence(), "sys", log=lambda *a: None)
|
||
assert out["summary"] == "整体尚可"
|
||
assert out["issues"] == [
|
||
{"category": "几何", "severity": "medium", "message": "零长度", "handle": "A1"}
|
||
]
|
||
|
||
|
||
def test_review_frame_parses_fenced_json():
|
||
llm = FakeLlmClient(script=[llm_text('```json\n{"summary": "s", "issues": []}\n```')])
|
||
ctx = make_ctx(client=FakeMcpClient(), llm=llm)
|
||
out = review_frame(ctx, make_evidence(), "sys", log=lambda *a: None)
|
||
assert out["summary"] == "s"
|
||
assert out["issues"] == []
|
||
|
||
|
||
def test_review_frame_intermediate_tool_turn_json_not_parsed():
|
||
# 中间工具回合的 content 里即使塞了 JSON，也不能被当成最终结论
|
||
intermediate = '{"summary": "这不是最终结论", "issues": [{"message": "x"}]}'
|
||
llm = FakeLlmClient(script=[
|
||
llm_calls([("query_entities", {})], content=intermediate),
|
||
llm_text('{"summary": "最终结论", "issues": []}'),
|
||
])
|
||
ctx = make_ctx(client=FakeMcpClient(), llm=llm)
|
||
out = review_frame(ctx, make_evidence(), "sys", log=lambda *a: None)
|
||
assert out["summary"] == "最终结论"
|
||
assert out["issues"] == []
|
||
|
||
|
||
def test_review_frame_includes_multiple_image_blocks():
|
||
llm = FakeLlmClient(script=[llm_text("ok")])
|
||
ctx = make_ctx(client=FakeMcpClient(), llm=llm)
|
||
ev = make_evidence(
|
||
images=[
|
||
{"b64": "A", "mime": "image/png", "label": "overall"},
|
||
{"b64": "B", "mime": "image/png", "label": "tile-1"},
|
||
]
|
||
)
|
||
review_frame(ctx, ev, "sys", log=lambda *a: None)
|
||
user_content = llm.calls[0]["messages"][1]["content"]
|
||
image_blocks = [b for b in user_content if b.get("type") == "image_url"]
|
||
assert len(image_blocks) == 2
|
||
assert any(b.get("text") == "（截图：overall）" for b in user_content)
|
||
assert any(b.get("text") == "（截图：tile-1）" for b in user_content)
|
||
|
||
|
||
# ---------------------------------------------------------------------------
|
||
# 视口复位（放大取证不残留给用户）
|
||
# ---------------------------------------------------------------------------
|
||
def test_review_frame_resets_viewport_after_zoom():
|
||
llm = FakeLlmClient(
|
||
script=[
|
||
llm_calls([("zoom_to_window", {"min_x": 1, "min_y": 2, "max_x": 3, "max_y": 4})]),
|
||
llm_text("done"),
|
||
]
|
||
)
|
||
client = FakeMcpClient()
|
||
ctx = make_ctx(client=client, llm=llm)
|
||
review_frame(ctx, make_evidence(), "sys", log=lambda *a: None)
|
||
zooms = client.called("zoom_to_window")
|
||
assert len(zooms) == 2 # LLM 放大一次 + 结束时复位一次
|
||
reset = zooms[-1]["arguments"]
|
||
assert (reset["min_x"], reset["min_y"]) == (0.0, 0.0) # 复位到图框整幅视图
|
||
assert (reset["max_x"], reset["max_y"]) == (100.0, 50.0)
|
||
assert reset["margin"] == 0.02
|
||
|
||
|
||
def test_review_frame_resets_viewport_after_zoom_to_entity():
|
||
llm = FakeLlmClient(
|
||
script=[llm_calls([("zoom_to_entity", {"handle": "A1"})]), llm_text("done")]
|
||
)
|
||
client = FakeMcpClient()
|
||
ctx = make_ctx(client=client, llm=llm)
|
||
review_frame(ctx, make_evidence(), "sys", log=lambda *a: None)
|
||
assert client.called_names().count("zoom_to_window") == 1
|
||
|
||
|
||
def test_review_frame_resets_viewport_at_iteration_limit():
|
||
llm = FakeLlmClient(
|
||
script=[llm_calls([("zoom_to_window", {})]), llm_text("收口结论")]
|
||
)
|
||
client = FakeMcpClient()
|
||
ctx = make_ctx(client=client, llm=llm, max_iters=1)
|
||
out = review_frame(ctx, make_evidence(), "sys", log=lambda *a: None)
|
||
assert out["summary"] == "收口结论"
|
||
assert client.called_names().count("zoom_to_window") == 2 # 强制收口路径也复位
|
||
|
||
|
||
def test_review_frame_does_not_touch_viewport_without_zoom():
|
||
llm = FakeLlmClient(script=[llm_calls([("get_layers", {})]), llm_text("x")])
|
||
client = FakeMcpClient()
|
||
ctx = make_ctx(client=client, llm=llm)
|
||
review_frame(ctx, make_evidence(), "sys", log=lambda *a: None)
|
||
assert "zoom_to_window" not in client.called_names()
|
||
|
||
|
||
def test_review_frame_viewport_reset_failure_is_swallowed():
|
||
seen = {"n": 0}
|
||
|
||
def zoom(_args):
|
||
seen["n"] += 1
|
||
if seen["n"] > 1:
|
||
raise McpToolError("复位失败")
|
||
return ok_text({})
|
||
|
||
llm = FakeLlmClient(
|
||
script=[llm_calls([("zoom_to_window", {})]), llm_text("结论")] # noqa: E501
|
||
)
|
||
client = FakeMcpClient(responses={"zoom_to_window": zoom})
|
||
ctx = make_ctx(client=client, llm=llm)
|
||
out = review_frame(ctx, make_evidence(), "sys", log=lambda *a: None)
|
||
assert out["summary"] == "结论" # 复位失败不影响结论
|
||
assert seen["n"] == 2
|
||
|
||
|
||
def test_review_frame_no_viewport_reset_without_window():
|
||
llm = FakeLlmClient(script=[llm_calls([("zoom_to_window", {})]), llm_text("x")])
|
||
client = FakeMcpClient()
|
||
ctx = make_ctx(client=client, llm=llm)
|
||
ev = make_evidence()
|
||
ev["window"] = None # 非 CAD 源（无窗口）→ 无从复位
|
||
review_frame(ctx, ev, "sys", log=lambda *a: None)
|
||
assert client.called_names().count("zoom_to_window") == 1
|