d909d1bd1b
实现侧： - 上下文：任务范围拆分为 范围校验/范围授权；索引按可发现口径重建、索引新鲜度改对称差；依赖校验统一快照漂移说明。 - 知识方法：方法与材料读取口径统一；超限方法材料按可选省略，核对路径不再二次计费；删除无合同的读时重算。 - 任务运行：新增 context.usage/tool.denied 事件类型；连接池常驻并在装配生命周期内开关；调用结算与核对分列。 - 效果评测/审校修订/交付连载/作者经验/作品规划：凭据冻结、标定消费、导出补证、事实引文核对等收尾修复。 - 资源加载：能力正文不再夹带索引用的导航注记（该注记此前进入角色与技能的模型提示）。 - 元数据：受保护骨架与代码保护属性对齐；字段校验与内置结构口径同步。 - 基础设施：环境预检进入装配生命周期；数据库连接运行期字段不参与相等比较；索引指纹归一化 jsonb 浮点。 - 删除被替代实现：7 份旧提示词模板与空壳 资料来源 读取器。 用例侧： - 用例身份与导航元信息迁移；夹具补生命周期、同库暴露与模板封存； - 本轮定向修复：方法材料省略、事实引文、迁移回执、额度与暂停用例、慢用例超时预算等。
264 lines
10 KiB
Python
264 lines
10 KiB
Python
"""空规则覆盖、触发依据和同事务读取的独立边界；只用合成输入。"""
|
|
|
|
from copy import deepcopy
|
|
from types import SimpleNamespace
|
|
from uuid import UUID
|
|
|
|
import pytest
|
|
from pydantic import ValidationError
|
|
|
|
from muse.审校修订 import 持久返修
|
|
from muse.审校修订.持久返修 import 返修比较定位
|
|
from muse.审校修订.接口 import 读取必需检查结论于
|
|
from muse.审校修订.机器味规则 import 核对规则, 核对规则库, 读取内置规则
|
|
from muse.审校修订.机器味诊断 import 诊断文本
|
|
from muse.审校修订.模型 import 审校错误
|
|
from muse.审校修订.返修策略 import 选择返修
|
|
from muse.正式变更.接口 import 固定哈希
|
|
|
|
|
|
def _规则库(触发: dict, *, active=True):
|
|
库 = 读取内置规则()
|
|
规则 = deepcopy(库["rules"]["l002"])
|
|
规则["trigger"] = 触发
|
|
规则["status"] = "active" if active else "candidate"
|
|
return 核对规则库([规则], list(库["samples"].values()))
|
|
|
|
|
|
@pytest.mark.case_id("NC-o03-b06-0001")
|
|
def test_空库不得声称已执行全文规则覆盖__b06001():
|
|
文本 = "这是合成诊断输入。"
|
|
报告 = 诊断文本(文本, work_id="synthetic", 规则库=核对规则库([], []))
|
|
assert 报告["mode"] == "active"
|
|
assert 报告["rule_count"] == 0 and 报告["empty_library"] is True
|
|
assert 报告["findings"] == 报告["executed_rules"] == 报告["unreviewed_rules"] == []
|
|
assert 报告["coverage"] == {
|
|
"deterministic": [],
|
|
"text_length": len(文本),
|
|
"rules_executed": 0,
|
|
}
|
|
|
|
|
|
@pytest.mark.case_id("NC-o03-b06-0002")
|
|
def test_有效规则零命中仍区别于空库__b06002():
|
|
文本 = "这是合成诊断输入。"
|
|
报告 = 诊断文本(
|
|
文本, work_id="synthetic", 规则库=_规则库({"type": "regex", "pattern": "不可能命中"})
|
|
)
|
|
assert 报告["findings"] == []
|
|
assert 报告["rule_count"] == 1 and 报告["empty_library"] is False
|
|
assert 报告["executed_rules"] == ["l002"]
|
|
assert 报告["coverage"]["deterministic"] == [[0, len(文本)]]
|
|
assert 报告["coverage"]["rules_executed"] == 1
|
|
|
|
|
|
@pytest.mark.case_id("NC-o03-b06-0003")
|
|
def test_只有语义规则未审不计确定性覆盖__b06003():
|
|
报告 = 诊断文本(
|
|
"这是合成诊断输入。",
|
|
work_id="synthetic",
|
|
规则库=_规则库({"type": "model_judgment", "criteria": "需要受控语义证据"}),
|
|
)
|
|
assert 报告["rule_count"] == 1 and 报告["empty_library"] is False
|
|
assert 报告["coverage"]["deterministic"] == []
|
|
assert 报告["coverage"]["rules_executed"] == 0
|
|
assert 报告["unreviewed_rules"] == [{"rule_id": "l002", "reason": "缺少受控语义交付"}]
|
|
|
|
|
|
@pytest.mark.case_id("NC-o03-b06-0004")
|
|
def test_候选规则不充当已激活覆盖__b06004():
|
|
库 = _规则库({"type": "regex", "pattern": "合成"}, active=False)
|
|
正式 = 诊断文本("这是合成诊断输入。", work_id="synthetic", 规则库=库)
|
|
预览 = 诊断文本("这是合成诊断输入。", work_id="synthetic", 规则库=库, 预览候选=True)
|
|
assert 正式["empty_library"] is True and 正式["coverage"]["deterministic"] == []
|
|
assert 预览["rule_count"] == 1 and 预览["empty_library"] is False
|
|
assert 预览["mode"] == "candidate_preview" and 预览["findings"]
|
|
assert 库["rules"]["l002"]["status"] == "candidate"
|
|
|
|
|
|
@pytest.mark.case_id("NC-o03-b06-0005")
|
|
@pytest.mark.parametrize(
|
|
"触发",
|
|
[
|
|
{"type": "regex", "pattern": "合成"},
|
|
{"type": "density", "pattern": "合成", "window_chars": 50, "min_hits": 2},
|
|
{"type": "handler", "handler": "short_sentence_run"},
|
|
{"type": "handler", "handler": "repeated_sentence_start"},
|
|
{"type": "handler", "handler": "camera_action_list", "pattern": "走|停"},
|
|
{"type": "handler", "handler": "uniform_paragraph_length"},
|
|
{"type": "model_judgment", "criteria": "需要受控语义证据"},
|
|
],
|
|
ids=["regex", "density", "short", "repeat", "camera", "uniform", "semantic"],
|
|
)
|
|
def test_四类有效触发依据保留规范载荷__b06005(触发):
|
|
库 = _规则库(触发)
|
|
规范 = 库["rules"]["l002"]["trigger"]
|
|
期望 = {
|
|
"type": 触发["type"],
|
|
"pattern": None,
|
|
"criteria": None,
|
|
"handler": None,
|
|
"max_chars": 8,
|
|
"min_run": 4,
|
|
"window_chars": 500,
|
|
"min_hits": 3,
|
|
"tolerance": 15,
|
|
**触发,
|
|
}
|
|
assert 规范 == 期望
|
|
assert 核对规则(库["rules"]["l002"], 库["samples"]) == 库["rules"]["l002"]
|
|
|
|
|
|
@pytest.mark.case_id("NC-o03-b06-0006")
|
|
@pytest.mark.parametrize(
|
|
"触发",
|
|
[
|
|
{"type": "regex"},
|
|
{"type": "regex", "pattern": "["},
|
|
{"type": "regex", "pattern": "合成", "criteria": "不应混用"},
|
|
{"type": "density", "pattern": "合成", "handler": "short_sentence_run"},
|
|
{"type": "handler"},
|
|
{"type": "handler", "handler": "camera_action_list"},
|
|
{"type": "handler", "handler": "short_sentence_run", "pattern": "不被消费"},
|
|
{"type": "handler", "handler": "uniform_paragraph_length", "criteria": "不应混用"},
|
|
{"type": "model_judgment", "criteria": " "},
|
|
{"type": "model_judgment", "criteria": "语义判据", "pattern": "不应混用"},
|
|
],
|
|
ids=[
|
|
"missing-pattern",
|
|
"bad-regex",
|
|
"regex-criteria",
|
|
"density-handler",
|
|
"missing-handler",
|
|
"camera-pattern",
|
|
"unused-pattern",
|
|
"handler-criteria",
|
|
"blank-criteria",
|
|
"semantic-pattern",
|
|
],
|
|
)
|
|
def test_非法或互斥触发依据由owner拒绝__b06006(触发):
|
|
with pytest.raises(审校错误, match="规则或四类例证不完整"):
|
|
_规则库(触发)
|
|
|
|
|
|
@pytest.mark.case_id("NC-o03-b06-0007")
|
|
@pytest.mark.parametrize("判定", [None, "passed", "failed"], ids=["missing", "passed", "failed"])
|
|
def test_必需检查按确切候选版本沿用调用连接__b06007(判定):
|
|
class 连接:
|
|
def __init__(self):
|
|
self.调用 = []
|
|
|
|
def execute(self, sql, params):
|
|
self.调用.append((sql, params))
|
|
return SimpleNamespace(fetchone=lambda: None if 判定 is None else (判定,))
|
|
|
|
连 = 连接()
|
|
assert 读取必需检查结论于(连, "synthetic-candidate", 7) == 判定
|
|
assert 连.调用 == [
|
|
(
|
|
"SELECT verdict FROM muse_review_report "
|
|
"WHERE candidate_id=%s AND candidate_revision=%s",
|
|
("synthetic-candidate", 7),
|
|
)
|
|
]
|
|
|
|
|
|
def _返修轮次():
|
|
候选ID = "00000000-0000-4000-8000-000000000002"
|
|
定位 = 返修比较定位(
|
|
session_id=UUID("00000000-0000-4000-8000-000000000001"),
|
|
round=1,
|
|
candidate_id=UUID(候选ID),
|
|
)
|
|
结果 = {
|
|
"candidate_id": 候选ID,
|
|
"outcome": "candidate",
|
|
"source_kind": "selection",
|
|
"diagnosis_id": None,
|
|
}
|
|
轮 = {
|
|
"round_number": 1,
|
|
"task_id": "synthetic-task",
|
|
"outcome": "candidate",
|
|
"candidate_id": 候选ID,
|
|
"result": 结果,
|
|
"result_hash": 固定哈希(结果),
|
|
}
|
|
return {"source_kind": "selection", "rounds": [轮]}, 定位
|
|
|
|
|
|
@pytest.mark.case_id("NC-o03-b06-0008")
|
|
@pytest.mark.parametrize(
|
|
"坏处,说明",
|
|
[
|
|
("none", None),
|
|
("wrong-candidate", "同一模型候选"),
|
|
("missing-task", "同一模型候选"),
|
|
("stale-hash", "结果哈希"),
|
|
("wrong-source", "来源类型"),
|
|
],
|
|
)
|
|
def test_比较轮次独立校验保留拒绝原因__b06008(坏处, 说明):
|
|
会话, 定位 = _返修轮次()
|
|
轮 = 会话["rounds"][0]
|
|
if 坏处 == "wrong-candidate":
|
|
轮["candidate_id"] = "other"
|
|
elif 坏处 == "missing-task":
|
|
轮["task_id"] = None
|
|
elif 坏处 == "stale-hash":
|
|
轮["result_hash"] = "0" * 64
|
|
elif 坏处 == "wrong-source":
|
|
轮["result"]["source_kind"] = "diagnosis"
|
|
轮["result_hash"] = 固定哈希(轮["result"])
|
|
if 说明 is None:
|
|
assert 持久返修._核对比较轮次(会话, 定位) is 轮
|
|
else:
|
|
with pytest.raises(审校错误, match=说明):
|
|
持久返修._核对比较轮次(会话, 定位)
|
|
|
|
|
|
@pytest.mark.case_id("NC-o03-b06-0009")
|
|
@pytest.mark.parametrize("坏处", ["none", "missing", "wrong-candidate", "failed", "false-gate"])
|
|
def test_比较报告不以引用代替保真证据__b06009(坏处):
|
|
报告 = {"candidate_id": "candidate", "verdict": "passed", "snapshot_id": "snapshot"}
|
|
结果 = {"snapshot_id": "snapshot", "checks": {"mechanical_pass": True}}
|
|
if 坏处 == "missing":
|
|
报告 = None
|
|
elif 坏处 == "wrong-candidate":
|
|
报告["candidate_id"] = "other"
|
|
elif 坏处 == "failed":
|
|
报告["verdict"] = "failed"
|
|
elif 坏处 == "false-gate":
|
|
结果["checks"]["mechanical_pass"] = False
|
|
if 坏处 == "none":
|
|
assert 持久返修._核对比较报告(报告, {"verdict": "passed"}, "candidate", 结果) is 报告
|
|
else:
|
|
with pytest.raises(审校错误, match="保真结果"):
|
|
持久返修._核对比较报告(报告, {"verdict": "passed"}, "candidate", 结果)
|
|
|
|
|
|
@pytest.mark.case_id("NC-o03-b06-0010")
|
|
def test_同事务比较读取不另建连接__b0600a(monkeypatch):
|
|
连, 运行, 结果 = object(), object(), object()
|
|
_, 定位 = _返修轮次()
|
|
调用 = []
|
|
|
|
def 核验(连接, 服务, 作者, 位置):
|
|
调用.append((连接, 服务, 作者, 位置))
|
|
return 结果
|
|
|
|
monkeypatch.setattr(持久返修, "_读取返修比较来源于", 核验)
|
|
assert 持久返修.读取返修比较来源于(连, 运行, "author", 定位) is 结果
|
|
assert 调用 == [(连, 运行, "author", 定位)]
|
|
|
|
|
|
@pytest.mark.case_id("NC-o03-b06-0011")
|
|
def test_有限返修公开合同仍以五轮为界__b0600b():
|
|
assert 选择返修(已用轮数=5, 最大轮数=5, 硬门通过=True, 有修改=True)["code"] == "round_limit"
|
|
with pytest.raises(审校错误, match="最多5轮"):
|
|
选择返修(已用轮数=5, 最大轮数=6, 硬门通过=True, 有修改=True)
|
|
_, 定位 = _返修轮次()
|
|
with pytest.raises(ValidationError):
|
|
返修比较定位(**{**定位.model_dump(), "round": 6})
|