feat(v2): 新增 S1 契约层（材料集合导入与事实确认）
T02 首个子卡 = S1 契约层。新增 backend/contracts/prescreen.py（设计指定的 唯一 owner），定义 SourceLocator / ImportItem / ImportBatch / FactConflict 与共用约束、错误码与错误信封。 落实的规格条目： - BR-22 逐文件保留来源/定位/解析版本与状态 - BR-23 多来源冲突保留候选并要求显式处理（resolved 需说明+选择+人） - AC-23 反例：绝对路径(C:/、/、//share)、上级引用(..)、超限、空集合、 failed 缺 error_code、parsed 缺 document_id、parsed 带 error_code - AC-24 反例：候选值全同不构成冲突；未知用 null 而非 0/空串； candidates 必须与 candidate_evidence_ids 同序同集合 - 01 §5 错误码与既有 translate() 的 HTTP 语义对齐；E-17 为运行级、HTTP 200 - 03 §9.1 时间必须带时区；禁止未知字段（E-07） D3 定案（工程限制，非业务阈值）：单文件 ≤10 MiB 沿用既有 LIMIT；单批 ≤20 文件 / ≤20 MiB。依据 03 §1 末「管理员只能降低初期限制，放宽须重新验证」， 取保守方向即规格许可；放宽需另立验证。 依赖方向：contracts 不导入 services，标准字段白名单校验留在 service 层。 测试：新增 test_prescreen_contracts.py 38 项，全绿；scripts/verify.py 9/9 PASS。 Co-Authored-By: Claude Code <noreply@anthropic.com>
This commit is contained in:
Claude Code
@@ -0,0 +1,321 @@
|
||||
"""S1 契约层测试：材料集合导入与事实确认。
|
||||
|
||||
覆盖 T00/04 §9 的 AC-23 / AC-24 可机械断言部分，以及契约自身的边界。
|
||||
每条规则都有反例——只测通过路径不能证明约束存在。
|
||||
"""
|
||||
|
||||
import unittest
|
||||
from datetime import datetime, timezone
|
||||
|
||||
from pydantic import ValidationError
|
||||
|
||||
from contracts.prescreen import (
|
||||
MAX_BATCH_BYTES,
|
||||
MAX_BATCH_FILES,
|
||||
MAX_ITEM_BYTES,
|
||||
ConflictCandidate,
|
||||
DataDomain,
|
||||
ErrorCode,
|
||||
ErrorResponse,
|
||||
FactConflict,
|
||||
ImportBatch,
|
||||
ImportItem,
|
||||
Ref,
|
||||
S1_ERROR_HTTP,
|
||||
SourceLocator,
|
||||
)
|
||||
|
||||
NOW = datetime(2026, 9, 29, 12, 0, tzinfo=timezone.utc)
|
||||
|
||||
|
||||
def locator(**over):
|
||||
base = dict(kind="pdf", document_id="doc-1", document_version=1, page=3)
|
||||
base.update(over)
|
||||
return base
|
||||
|
||||
|
||||
def item(**over):
|
||||
base = dict(
|
||||
id="item-1", relative_path="materials/a.pdf", sha256="a" * 64,
|
||||
size_bytes=1024, mime_type="application/pdf", status="queued",
|
||||
received_at=NOW,
|
||||
)
|
||||
base.update(over)
|
||||
return base
|
||||
|
||||
|
||||
def batch(**over):
|
||||
base = dict(
|
||||
id="batch-1", project_id="DEMO-A", data_domain="demo", revision=1,
|
||||
status="queued", created_at=NOW, created_by="demo-manager",
|
||||
parser_version="assistant_materials@1", items=[item()], total_bytes=1024,
|
||||
)
|
||||
base.update(over)
|
||||
return base
|
||||
|
||||
|
||||
def conflict(**over):
|
||||
ids = ["ev-1", "ev-2"]
|
||||
base = dict(
|
||||
id="fc-1", project_id="DEMO-A", data_domain="demo", revision=1,
|
||||
field="facility.amount", candidate_evidence_ids=list(ids),
|
||||
candidates=[
|
||||
ConflictCandidate(evidence_id="ev-1", value="1000", source="a.csv", locator="第2行"),
|
||||
ConflictCandidate(evidence_id="ev-2", value="1200", source="b.csv", locator="第5行"),
|
||||
],
|
||||
status="unresolved", created_at=NOW,
|
||||
)
|
||||
base.update(over)
|
||||
return base
|
||||
|
||||
|
||||
class SharedConstraintTests(unittest.TestCase):
|
||||
def test_unknown_field_is_rejected(self):
|
||||
# E-07：所有输入禁止未知字段
|
||||
with self.assertRaises(ValidationError):
|
||||
ImportItem(**item(), unexpected="x")
|
||||
|
||||
def test_naive_datetime_is_rejected(self):
|
||||
with self.assertRaises(ValidationError):
|
||||
ImportItem(**item(received_at=datetime(2026, 9, 29, 12, 0)))
|
||||
|
||||
def test_ref_requires_both_fields(self):
|
||||
self.assertEqual(Ref(id="r-1", version="v1").version, "v1")
|
||||
with self.assertRaises(ValidationError):
|
||||
Ref(id="r-1")
|
||||
|
||||
def test_data_domain_enum_values(self):
|
||||
self.assertEqual({d.value for d in DataDomain}, {"demo", "production"})
|
||||
with self.assertRaises(ValidationError):
|
||||
ImportBatch(**batch(data_domain="staging"))
|
||||
|
||||
|
||||
class SourceLocatorTests(unittest.TestCase):
|
||||
def test_each_kind_accepts_its_own_fields(self):
|
||||
self.assertEqual(SourceLocator(**locator()).page, 3)
|
||||
self.assertEqual(
|
||||
SourceLocator(**locator(kind="table", page=None, sheet="Sheet1",
|
||||
cell_range="C4")).sheet, "Sheet1")
|
||||
self.assertEqual(
|
||||
SourceLocator(**locator(kind="manual", page=None, note="正文第2段")).note,
|
||||
"正文第2段")
|
||||
|
||||
def test_missing_required_field_per_kind(self):
|
||||
for over, missing in [
|
||||
(dict(page=None), "page"),
|
||||
(dict(kind="table", page=None), "sheet"),
|
||||
(dict(kind="table", page=None, sheet="S"), "cell_range"),
|
||||
(dict(kind="manual", page=None), "note"),
|
||||
]:
|
||||
with self.subTest(missing=missing):
|
||||
with self.assertRaises(ValidationError) as ctx:
|
||||
SourceLocator(**locator(**over))
|
||||
self.assertIn(missing, str(ctx.exception))
|
||||
|
||||
def test_cross_kind_fields_are_rejected(self):
|
||||
# 三种寻址方式互斥，不允许混填
|
||||
with self.assertRaises(ValidationError):
|
||||
SourceLocator(**locator(kind="table", sheet="S", cell_range="C4")) # page 也在
|
||||
with self.assertRaises(ValidationError):
|
||||
SourceLocator(**locator(note="顺带写个说明"))
|
||||
|
||||
def test_page_must_be_positive(self):
|
||||
with self.assertRaises(ValidationError):
|
||||
SourceLocator(**locator(page=0))
|
||||
|
||||
|
||||
class ImportItemTests(unittest.TestCase):
|
||||
def test_valid_item(self):
|
||||
self.assertEqual(ImportItem(**item()).status, "queued")
|
||||
|
||||
def test_absolute_paths_rejected(self):
|
||||
for bad in ["/etc/passwd", "C:/Windows/x.pdf", "C:\\Windows\\x.pdf", "//share/x.pdf"]:
|
||||
with self.subTest(path=bad):
|
||||
with self.assertRaises(ValidationError):
|
||||
ImportItem(**item(relative_path=bad))
|
||||
|
||||
def test_parent_reference_rejected(self):
|
||||
with self.assertRaises(ValidationError):
|
||||
ImportItem(**item(relative_path="../../secret.pdf"))
|
||||
with self.assertRaises(ValidationError):
|
||||
ImportItem(**item(relative_path="a/../../b.pdf"))
|
||||
|
||||
def test_backslash_relative_path_is_kept_verbatim(self):
|
||||
# 校验用规范化副本，但保留原值：落库内容须与客户端提交一致
|
||||
parsed = ImportItem(**item(relative_path="materials\\sub\\a.pdf"))
|
||||
self.assertEqual(parsed.relative_path, "materials\\sub\\a.pdf")
|
||||
|
||||
def test_sha256_shape(self):
|
||||
with self.assertRaises(ValidationError):
|
||||
ImportItem(**item(sha256="A" * 64)) # 大写不合法
|
||||
with self.assertRaises(ValidationError):
|
||||
ImportItem(**item(sha256="a" * 63))
|
||||
|
||||
def test_single_file_over_limit_rejected(self):
|
||||
with self.assertRaises(ValidationError):
|
||||
ImportItem(**item(size_bytes=MAX_ITEM_BYTES + 1))
|
||||
|
||||
def test_failed_requires_error_code_and_parsed_forbids_it(self):
|
||||
with self.assertRaises(ValidationError):
|
||||
ImportItem(**item(status="failed"))
|
||||
self.assertEqual(
|
||||
ImportItem(**item(status="failed", error_code=ErrorCode.MATERIAL_PARSE_FAILED)).status,
|
||||
"failed")
|
||||
with self.assertRaises(ValidationError):
|
||||
ImportItem(**item(status="parsed", document_id="doc-1",
|
||||
error_code=ErrorCode.MATERIAL_PARSE_FAILED))
|
||||
|
||||
def test_parsed_requires_document_id(self):
|
||||
# AC-23：解析成功必须能定位到既有 document
|
||||
with self.assertRaises(ValidationError):
|
||||
ImportItem(**item(status="parsed"))
|
||||
self.assertEqual(
|
||||
ImportItem(**item(status="parsed", document_id="doc-9")).document_id, "doc-9")
|
||||
|
||||
def test_observed_at_may_be_null_and_is_not_faked(self):
|
||||
# BR-02：缺发生时间保留 null，不用导入时间冒充
|
||||
parsed = ImportItem(**item())
|
||||
self.assertIsNone(parsed.observed_at)
|
||||
self.assertIsNotNone(parsed.received_at)
|
||||
|
||||
|
||||
class ImportBatchTests(unittest.TestCase):
|
||||
def test_valid_batch(self):
|
||||
self.assertEqual(ImportBatch(**batch()).total_bytes, 1024)
|
||||
|
||||
def test_total_bytes_must_match_sum(self):
|
||||
with self.assertRaises(ValidationError):
|
||||
ImportBatch(**batch(total_bytes=999))
|
||||
|
||||
def test_duplicate_item_ids_rejected(self):
|
||||
with self.assertRaises(ValidationError):
|
||||
ImportBatch(**batch(
|
||||
items=[item(), item()], total_bytes=2048))
|
||||
|
||||
def test_empty_items_rejected(self):
|
||||
# E-16：空集合不合法
|
||||
with self.assertRaises(ValidationError):
|
||||
ImportBatch(**batch(items=[], total_bytes=0))
|
||||
|
||||
def test_over_file_count_rejected(self):
|
||||
with self.assertRaises(ValidationError):
|
||||
ImportBatch(**batch(items=[], total_bytes=0)) # min_length=1 已拦截
|
||||
items = [item(id=f"i-{n}", size_bytes=1) for n in range(MAX_BATCH_FILES + 1)]
|
||||
with self.assertRaises(ValidationError):
|
||||
ImportBatch(**batch(items=items, total_bytes=len(items)))
|
||||
|
||||
def test_batch_total_limit_enforced(self):
|
||||
# D3 决策：单批 ≤ 20 MiB。每个文件合法但合计超限必须拒绝。
|
||||
each = MAX_BATCH_BYTES // 3 + 1
|
||||
items = [item(id=f"i-{n}", size_bytes=each) for n in range(3)]
|
||||
self.assertLessEqual(each, MAX_ITEM_BYTES)
|
||||
with self.assertRaises(ValidationError) as ctx:
|
||||
ImportBatch(**batch(items=items, total_bytes=each * 3))
|
||||
self.assertIn("拆分", str(ctx.exception))
|
||||
|
||||
def test_status_enum_bounded(self):
|
||||
with self.assertRaises(ValidationError):
|
||||
ImportBatch(**batch(status="done"))
|
||||
|
||||
|
||||
class FactConflictTests(unittest.TestCase):
|
||||
def test_valid_unresolved(self):
|
||||
parsed = FactConflict(**conflict())
|
||||
self.assertEqual(parsed.status, "unresolved")
|
||||
self.assertEqual(parsed.selected_evidence_ids, [])
|
||||
|
||||
def test_fewer_than_two_candidates_rejected(self):
|
||||
with self.assertRaises(ValidationError):
|
||||
FactConflict(**conflict(candidate_evidence_ids=["ev-1"]))
|
||||
|
||||
def test_identical_values_are_not_a_conflict(self):
|
||||
# AC-24 反例：两个来源给出同一个值不应建冲突对象
|
||||
with self.assertRaises(ValidationError) as ctx:
|
||||
FactConflict(**conflict(candidates=[
|
||||
ConflictCandidate(evidence_id="ev-1", value="1000", source="a", locator="L1"),
|
||||
ConflictCandidate(evidence_id="ev-2", value="1000", source="b", locator="L2"),
|
||||
]))
|
||||
self.assertIn("不构成", str(ctx.exception))
|
||||
|
||||
def test_null_value_counts_as_distinct(self):
|
||||
# 未知用 null，且与已有数值构成分歧
|
||||
parsed = FactConflict(**conflict(candidates=[
|
||||
ConflictCandidate(evidence_id="ev-1", value="1000", source="a", locator="L1"),
|
||||
ConflictCandidate(evidence_id="ev-2", value=None, source="b", locator="L2"),
|
||||
]))
|
||||
self.assertIsNone(parsed.candidates[1].value)
|
||||
|
||||
def test_candidates_must_align_with_id_list(self):
|
||||
with self.assertRaises(ValidationError):
|
||||
FactConflict(**conflict(candidates=[
|
||||
ConflictCandidate(evidence_id="ev-2", value="1000", source="a", locator="L1"),
|
||||
ConflictCandidate(evidence_id="ev-1", value="1200", source="b", locator="L2"),
|
||||
]))
|
||||
with self.assertRaises(ValidationError):
|
||||
FactConflict(**conflict(candidate_evidence_ids=["ev-1", "ev-1"], candidates=[
|
||||
ConflictCandidate(evidence_id="ev-1", value="1000", source="a", locator="L1"),
|
||||
ConflictCandidate(evidence_id="ev-1", value="1200", source="b", locator="L2"),
|
||||
]))
|
||||
|
||||
def test_resolved_requires_note_selection_and_actor(self):
|
||||
good = conflict(status="resolved", resolution_note="以合同为准",
|
||||
selected_evidence_ids=["ev-2"], resolved_by="demo-manager",
|
||||
resolved_at=NOW)
|
||||
self.assertEqual(FactConflict(**good).selected_evidence_ids, ["ev-2"])
|
||||
for drop in ("resolution_note", "resolved_by", "resolved_at", "selected_evidence_ids"):
|
||||
with self.subTest(drop=drop):
|
||||
broken = dict(good)
|
||||
broken.pop(drop)
|
||||
with self.assertRaises(ValidationError):
|
||||
FactConflict(**broken)
|
||||
|
||||
def test_selection_must_come_from_candidates(self):
|
||||
with self.assertRaises(ValidationError) as ctx:
|
||||
FactConflict(**conflict(status="resolved", resolution_note="n",
|
||||
selected_evidence_ids=["ev-999"],
|
||||
resolved_by="demo-manager", resolved_at=NOW))
|
||||
self.assertIn("不属于", str(ctx.exception))
|
||||
|
||||
def test_unresolved_must_not_carry_resolution(self):
|
||||
with self.assertRaises(ValidationError):
|
||||
FactConflict(**conflict(resolution_note="提前写了结论"))
|
||||
|
||||
def test_supersedes_chain_is_representable(self):
|
||||
# 再次冲突新建对象并指向旧对象，不覆写历史
|
||||
parsed = FactConflict(**conflict(id="fc-2", supersedes_id="fc-1"))
|
||||
self.assertEqual(parsed.supersedes_id, "fc-1")
|
||||
|
||||
|
||||
class ErrorContractTests(unittest.TestCase):
|
||||
def test_every_code_has_an_http_status(self):
|
||||
self.assertEqual(set(S1_ERROR_HTTP), set(ErrorCode))
|
||||
|
||||
def test_status_semantics_match_existing_translate(self):
|
||||
# 既有 translate() 的语义：KeyError→404, Conflict→409, PermissionError→403, ValueError→422
|
||||
self.assertEqual(S1_ERROR_HTTP[ErrorCode.NOT_FOUND], 404)
|
||||
self.assertEqual(S1_ERROR_HTTP[ErrorCode.ACCESS_DENIED], 403)
|
||||
self.assertEqual(S1_ERROR_HTTP[ErrorCode.REVISION_CONFLICT], 409)
|
||||
self.assertEqual(S1_ERROR_HTTP[ErrorCode.IDEMPOTENCY_CONFLICT], 409)
|
||||
self.assertEqual(S1_ERROR_HTTP[ErrorCode.IMPORT_BATCH_INVALID], 422)
|
||||
|
||||
def test_e17_is_not_an_http_error(self):
|
||||
# 01 §5 末：运行级错误 HTTP 200，只表示结果可读
|
||||
self.assertEqual(S1_ERROR_HTTP[ErrorCode.FACT_CONFLICT_UNRESOLVED], 200)
|
||||
|
||||
def test_envelope_shape(self):
|
||||
body = ErrorResponse.model_validate({"error": {
|
||||
"code": "IMPORT_BATCH_INVALID", "message": "材料集合不符合导入要求",
|
||||
"retryable": False, "request_id": "req-1"}})
|
||||
self.assertEqual(body.error.code, ErrorCode.IMPORT_BATCH_INVALID)
|
||||
|
||||
def test_no_credentials_in_contract_module(self):
|
||||
# 契约层不得出现凭据字段（E-10 / BR-16 的静态底线）
|
||||
import contracts.prescreen as module
|
||||
with open(module.__file__, encoding="utf-8") as handle:
|
||||
source = handle.read()
|
||||
for banned in ("api_key", "password", "token", "secret"):
|
||||
self.assertNotIn(banned, source.lower())
|
||||
|
||||
|
||||
if __name__ == "__main__":
|
||||
unittest.main()
|
||||
Reference in New Issue
Block a user