a95e27a290
- 文字层 v2：页眉页脚带单独成句；写到页尾的句子和下一页首句接续（出处两页各一条） - 结构识别（最小版）：十进制编号、冒号引出的列表找上级条款；上级条款写明无效的，条目照样认定 - 抽取：资格、评分、日期与递交、技术参数、项目基本信息各一份提示词；实质性条款出 v2 （认上级条款、补「做无效响应处理」等写法）；各类取值核对；按首句合并； 拆开被合成一条的列举条款；描述标注方式的句子和表头碎片不当条款；* 号标注 - 项目：解析出的名称、编号、采购人、预算、截止时间、分包写回项目，用户改过的不覆盖 - 前端：清单项显示分值、参数要求、日期、证明材料 - 测试 40 个，新增两页虚构样例 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
219 lines
9.9 KiB
Python
219 lines
9.9 KiB
Python
"""不需要数据库的单元测试：证据校验、格式识别、文字层。"""
|
||
|
||
from __future__ import annotations
|
||
|
||
import pytest
|
||
|
||
from bid_doctor.checklist.extract import (
|
||
Doc,
|
||
Extracted,
|
||
amount_yuan,
|
||
check_basis,
|
||
check_evidence,
|
||
check_value,
|
||
is_marked,
|
||
merge_items,
|
||
parse_at,
|
||
)
|
||
from bid_doctor.core.errors import AppError
|
||
from bid_doctor.core.events import _sid # pyright: ignore[reportPrivateUsage]
|
||
from bid_doctor.document.formats import detect
|
||
from bid_doctor.document.pipeline.text_layer import Sentence, extract
|
||
from tests.conftest import FIXTURES
|
||
|
||
|
||
def sent(text: str, runs: list[tuple[int, int, str]] | None = None, seq: int = 1) -> Sentence:
|
||
return Sentence(key=f"d:p1:s{seq}", page_no=1, seq=seq, text=text, quads=[[0, 0, 1, 1]], fmt_runs=runs or [])
|
||
|
||
|
||
def test_evidence_checks() -> None:
|
||
s = sent("★投标人须具备有效的信息系统集成服务资质证书。")
|
||
by_id = {"p1s1": s}
|
||
assert check_evidence({"sid": "p1s1", "quote": "投标人须具备有效的信息"}, by_id) == (None, [s])
|
||
assert check_evidence({"sid": "p9s9", "quote": "投标人须具备有效的信息"}, by_id)[0] == "id"
|
||
assert check_evidence({"sid": "p1s1", "quote": "投标人须具备"}, by_id)[0] == "quote" # 少于 8 个字
|
||
assert check_evidence({"sid": "p1s1", "quote": "这句话是模型改写的"}, by_id)[0] == "quote"
|
||
|
||
|
||
def test_quote_may_run_into_next_sentence() -> None:
|
||
"""表格、跨页把一句切成两段时，引文可以从这一句接续到紧跟着的下一句，两句都算出处。"""
|
||
a = sent("支持不少于200个并发", seq=1)
|
||
b = sent("用户，连续30分钟压力测试成功率不低于99%。", seq=2)
|
||
by_id = {"p1s1": a, "p1s2": b}
|
||
assert check_evidence({"sid": "p1s1", "quote": "支持不少于200个并发用户"}, by_id, {a.key: b}) == (None, [a, b])
|
||
assert check_evidence({"sid": "p1s1", "quote": "支持不少于200个并发用户"}, by_id)[0] == "quote"
|
||
# 必须从这一句开始，不能整段都在下一句里
|
||
assert check_evidence({"sid": "p1s1", "quote": "连续30分钟压力测试"}, by_id, {a.key: b})[0] == "quote"
|
||
|
||
|
||
def test_split_listed_conditions() -> None:
|
||
from bid_doctor.checklist.extract import split_listed
|
||
|
||
heading = sent("25.3未实质性响应磋商文件的情形", seq=1)
|
||
items = [sent("25.3.1未成功解密电子响应文件的。", seq=2), sent("25.3.2未按要求上传电子响应文件的。", seq=3)]
|
||
d = doc(sentences=[heading, *items], parents={x.key: heading for x in items})
|
||
merged = Extracted(
|
||
task="substantive",
|
||
categories=["invalidity"],
|
||
summary="未实质性响应的情形",
|
||
sentences=items,
|
||
zone="formal",
|
||
basis="explicit_invalid",
|
||
importance="substantive",
|
||
)
|
||
parts = split_listed(merged, d)
|
||
assert [p.sentences[0].seq for p in parts] == [2, 3]
|
||
assert all(p.zone == "formal" for p in parts)
|
||
|
||
|
||
def doc(
|
||
declared: set[str] | None = None,
|
||
sentences: list[Sentence] | None = None,
|
||
parents: dict[str, Sentence] | None = None,
|
||
) -> Doc:
|
||
return Doc(sentences=sentences or [], parents=parents or {}, continuations={}, notes=[], declared=declared or set())
|
||
|
||
|
||
def test_basis_check() -> None:
|
||
star = sent("★服务器内存不低于32GB。")
|
||
plain = sent("投标人应按要求提供售后服务方案。")
|
||
iu = sent("投标人须提供原厂授权书。", [(0, 12, "iu")])
|
||
assert check_basis("star", [star], [star], doc())
|
||
assert not check_basis("star", [plain], [plain], doc())
|
||
# 斜体下划线只在文件声明过时才算（07 F1）
|
||
assert not check_basis("italic_underline", [iu], [iu], doc())
|
||
assert check_basis("italic_underline", [iu], [iu], doc({"italic_underline"}))
|
||
invalid = sent("未提供的，其投标无效。", seq=2)
|
||
assert check_basis("explicit_invalid", [plain], [plain, invalid], doc())
|
||
assert not check_basis("unknown", [star], [star], doc())
|
||
|
||
|
||
def test_invalid_wordings_and_parent_heading() -> None:
|
||
"""「做无效响应处理」「视为非实质性响应」也算写明无效；列在无效情形标题下的条目，上级标题就是依据。"""
|
||
for text in ("否则做无效响应处理。", "将被视为非实质性响应。", "其投标将予以拒绝。"):
|
||
s = sent(text)
|
||
assert check_basis("explicit_invalid", [s], [s], doc()), text
|
||
heading = sent("25.3未实质性响应磋商文件的情形", seq=1)
|
||
item = sent("25.3.5供应商的最后报价超过了采购预算。", seq=9)
|
||
assert not check_basis("explicit_invalid", [item], [item], doc())
|
||
assert check_basis("explicit_invalid", [item], [item], doc(parents={item.key: heading}))
|
||
|
||
|
||
def test_marker_talk_is_not_a_clause() -> None:
|
||
"""国动样本 p67：偏离表表头「序号｜实质性要求（斜体且有下划线部分或加*号条款）」被切成碎片，不能当条款。"""
|
||
frag = sent("号（斜体且有下划线部分或加", [(0, 13, "biu")])
|
||
assert is_marked(frag, {"italic_underline", "asterisk"}) is None
|
||
assert not check_basis("italic_underline", [frag], [frag], doc({"italic_underline"}))
|
||
|
||
|
||
def test_asterisk_needs_declaration() -> None:
|
||
s = sent("*供应商须提供原厂授权书。")
|
||
assert is_marked(s, set()) is None
|
||
assert is_marked(s, {"asterisk"}) == "asterisk"
|
||
assert is_marked(sent("3.*支持国产化CPU。"), {"asterisk"}) == "asterisk"
|
||
|
||
|
||
def test_value_checks() -> None:
|
||
s = [sent("响应文件递交截止时间：2026年12月1日9时30分（北京时间）。")]
|
||
assert parse_at("2026-12-01T09:30", s) == ("2026-12-01T09:30:00+08:00", True)
|
||
assert parse_at("2026-12-02T09:30", s)[1] is False # 日期对不上原文：进待确认
|
||
assert check_value("200", [sent("系统支持不少于200个并发连接。")])
|
||
assert not check_value("300", [sent("系统支持不少于200个并发连接。")])
|
||
assert check_value("0～5", [sent("类似业绩每个得1分，最高5分，满分0～5分。")])
|
||
assert amount_yuan("45", "万元") == "450000.00"
|
||
assert amount_yuan("8,000", "元") == "8000.00"
|
||
|
||
|
||
def test_parents_and_continuations() -> None:
|
||
from bid_doctor.document.pipeline.structure import parents
|
||
from bid_doctor.document.pipeline.text_layer import continuations
|
||
|
||
ss = [
|
||
sent("25.3未实质性响应磋商文件的情形", seq=1),
|
||
sent("25.3.1未加盖电子公章的。", seq=2),
|
||
sent("有下列情形之一的，投标无效：", seq=3),
|
||
sent("（1）未提交保证金的；", seq=4),
|
||
sent("26.评审", seq=5),
|
||
]
|
||
par = parents(ss)
|
||
assert par["d:p1:s2"].seq == 1 and par["d:p1:s4"].seq == 3 and "d:p1:s5" not in par
|
||
|
||
a = Sentence(key="d:p1:s9", page_no=1, seq=9, text="采购人有权", quads=[], open_end=True)
|
||
b = Sentence(key="d:p2:s1", page_no=2, seq=1, text="永久使用、修改、维护。", quads=[])
|
||
h = Sentence(key="d:p2:s1", page_no=2, seq=1, text="11.1供应商应按规定格式填报。", quads=[])
|
||
assert continuations([a, b]) == {"d:p1:s9": b}
|
||
assert continuations([a, h]) == {} # 下一页是新的编号条目：不是同一句
|
||
|
||
|
||
def test_merge_items_across_tasks() -> None:
|
||
s = sent("★服务器内存不低于32GB。")
|
||
sub = Extracted(
|
||
task="substantive",
|
||
categories=["invalidity"],
|
||
summary="内存要求",
|
||
sentences=[s],
|
||
zone="formal",
|
||
importance="substantive",
|
||
spec={"invalidity": {"source": "substantive_marker"}},
|
||
)
|
||
tech = Extracted(
|
||
task="tech_param",
|
||
categories=["tech_param", "invalidity"],
|
||
summary="内存不低于 32GB",
|
||
sentences=[s],
|
||
zone="formal",
|
||
importance="substantive",
|
||
spec={"tech_param": {"param": "内存", "op": ">=", "value": "32", "unit": "GB"}},
|
||
)
|
||
(m,) = merge_items([sub, tech])
|
||
assert m.summary == "内存不低于 32GB"
|
||
assert set(m.categories) == {"tech_param", "invalidity"}
|
||
assert set(m.spec) == {"tech_param", "invalidity"}
|
||
|
||
|
||
def test_marker_fallback() -> None:
|
||
assert is_marked(sent("★采用国产化CPU芯片。"), set()) == "star"
|
||
assert is_marked(sent("带★号的条款为实质性要求。"), set()) is None # 规则说明本身
|
||
iu = sent("投标人须提供原厂授权书。", [(0, 12, "iu")])
|
||
assert is_marked(iu, set()) is None
|
||
assert is_marked(iu, {"italic_underline"}) == "italic_underline"
|
||
|
||
|
||
def test_marked_text_for_model() -> None:
|
||
s = sent("须提供授权书，否则无效", [(3, 6, "iu"), (9, 11, "u")])
|
||
assert s.marked == "须提供<iu>授权书</iu>，否则<u>无效</u>"
|
||
|
||
|
||
def test_detect_formats() -> None:
|
||
pdf = (FIXTURES / "tender-mini.pdf").read_bytes()
|
||
d = detect(pdf, "采购文件.pdf")
|
||
assert (d.format, d.page_count, d.extension_mismatch) == ("pdf", 1, False)
|
||
assert detect(pdf, "采购文件.doc").extension_mismatch
|
||
with pytest.raises(AppError) as e:
|
||
detect(b"hello", "a.doc")
|
||
assert e.value.code == "file_format_unsupported"
|
||
|
||
|
||
def test_text_layer_normalizes_radicals_and_keys() -> None:
|
||
pages = extract("doc1", (FIXTURES / "tender-mini.pdf").read_bytes())
|
||
texts = [s.text for s in pages[0].sentences]
|
||
assert any("项目" in t for t in texts) and not any("⽬" in t for t in texts) # 「目」不是康熙部首
|
||
assert any("彩页" in t for t in texts)
|
||
keys = [s.key for s in pages[0].sentences]
|
||
assert keys[0] == "doc1:p1:s1" and len(set(keys)) == len(keys)
|
||
|
||
|
||
def test_stream_id_compare() -> None:
|
||
assert _sid("1700000000000-1") > _sid("999-5")
|
||
assert _sid("bad") == (0, 0)
|
||
|
||
|
||
def test_heading_is_context_not_the_requirement() -> None:
|
||
from bid_doctor.checklist.extract import without_headings
|
||
|
||
heading = sent("25.3未实质性响应磋商文件的情形", seq=1)
|
||
item = sent("25.3.1未成功解密电子响应文件的。", seq=2)
|
||
d = doc(sentences=[heading, item], parents={item.key: heading})
|
||
assert without_headings([heading, item], d) == [item]
|
||
assert without_headings([heading], d) == [heading] # 只引了标题：保留
|