105 lines
3.8 KiB
Python
105 lines
3.8 KiB
Python
"""解析器契约测试。
|
|
|
|
除 brief 给定的 7 条契约外，另有一条**对真实源文件**的计数测试：`RAG-` 前缀
|
|
只覆盖 62/105 条，漏掉全部 43 条 `NF-*`（真实文件里共 12 种前缀）。真实源文件是
|
|
工作区外的交付物，缺失时 skip（不算失败），存在时按期望值强校验。
|
|
"""
|
|
|
|
from __future__ import annotations
|
|
|
|
import hashlib
|
|
from dataclasses import FrozenInstanceError
|
|
from pathlib import Path
|
|
|
|
import pytest
|
|
|
|
from tools.qa_source_parser import (
|
|
EXPECTED_RECORD_COUNT,
|
|
QaRecord,
|
|
normalize_phrase,
|
|
parse_qa_source,
|
|
phrase_hash,
|
|
)
|
|
|
|
SAMPLE = """
|
|
[RAG-PER-001]
|
|
问题：你是谁？
|
|
相似问法：你是人工吗？|你是真人客服吗？
|
|
回答：我是奶龙基金智能助手。
|
|
|
|
[NF-SVC-001]
|
|
问题：怎么联系你们？
|
|
相似问法：客服入口在哪？
|
|
回答：您可通过官网获取服务。
|
|
"""
|
|
|
|
#: 已审核源文件的实际位置（工作区外）。用于验证"11 种前缀"这一前提是否成立。
|
|
REAL_SOURCE = Path(
|
|
r"C:\Users\Windows\Desktop\DSH开发智能助手\胜宇前期资料"
|
|
r"\客服Agent知识库_QA问答对_v5_RAG发布候选版.txt"
|
|
)
|
|
|
|
|
|
def test_expected_count_is_105() -> None:
|
|
assert EXPECTED_RECORD_COUNT == 105
|
|
|
|
|
|
def test_parser_accepts_both_id_prefixes() -> None:
|
|
records = parse_qa_source(SAMPLE, expected_count=None)
|
|
assert [r.qa_id for r in records] == ["RAG-PER-001", "NF-SVC-001"]
|
|
|
|
|
|
def test_parser_splits_synonyms_on_pipe() -> None:
|
|
records = parse_qa_source(SAMPLE, expected_count=None)
|
|
assert records[0].synonyms == ("你是人工吗？", "你是真人客服吗？")
|
|
|
|
|
|
def test_parser_keeps_answer_verbatim() -> None:
|
|
records = parse_qa_source(SAMPLE, expected_count=None)
|
|
assert records[0].answer == "我是奶龙基金智能助手。"
|
|
|
|
|
|
def test_parser_rejects_record_without_synonyms() -> None:
|
|
with pytest.raises(ValueError, match="相似问法"):
|
|
parse_qa_source("[RAG-PER-001]\n问题：问？\n回答：答。\n", expected_count=None)
|
|
|
|
|
|
def test_parser_rejects_wrong_record_count_by_default() -> None:
|
|
with pytest.raises(ValueError, match="记录数不符"):
|
|
parse_qa_source(SAMPLE)
|
|
|
|
|
|
def test_phrase_hash_is_sha256_of_normalized_phrase() -> None:
|
|
value = "你是人工吗？"
|
|
assert phrase_hash(value) == hashlib.sha256(
|
|
normalize_phrase(value).encode("utf-8")).hexdigest()
|
|
assert len(phrase_hash(value)) == 64
|
|
|
|
|
|
def test_record_is_frozen_dataclass() -> None:
|
|
record = parse_qa_source(SAMPLE, expected_count=None)[0]
|
|
assert isinstance(record, QaRecord)
|
|
assert isinstance(record.synonyms, tuple)
|
|
with pytest.raises(FrozenInstanceError):
|
|
record.question = "改不了" # type: ignore[misc]
|
|
|
|
|
|
def test_parser_rejects_duplicate_synonym_within_record_after_normalization() -> None:
|
|
"""`uk_faq_synonym (knowledge_id, phrase_hash)` 只认归一化后的哈希。"""
|
|
text = "[RAG-PER-001]\n问题：你是谁？\n相似问法：你是谁？| 你是谁？ |你 是 谁？\n回答：答。\n"
|
|
synonyms = parse_qa_source(text, expected_count=None)[0].synonyms
|
|
assert len({phrase_hash(item) for item in synonyms}) == len(synonyms)
|
|
|
|
|
|
@pytest.mark.skipif(not REAL_SOURCE.exists(), reason="已审核 QA 源文件不在本机")
|
|
def test_parser_counts_real_source_file() -> None:
|
|
records = parse_qa_source(REAL_SOURCE.read_text(encoding="utf-8"))
|
|
assert len(records) == EXPECTED_RECORD_COUNT
|
|
rag = [r for r in records if r.qa_id.startswith("RAG-")]
|
|
nf = [r for r in records if r.qa_id.startswith("NF-")]
|
|
assert (len(rag), len(nf)) == (62, 43)
|
|
assert all(r.question and r.answer and r.synonyms for r in records)
|
|
assert sum(len(r.synonyms) for r in records) == 421
|
|
# 前缀多于"RAG-"/"NF-"两类：只按 RAG- 解析会静默丢掉 NF-* 全部记录。
|
|
assert len([r for r in records if not r.qa_id.startswith(("RAG-", "NF-"))]) == 0
|