217 lines
8.4 KiB
Python
217 lines
8.4 KiB
Python
# -*- coding: utf-8 -*-
|
||
"""
|
||
Intention (L7) 召回 + 惰性过期转换 端到端测试（chroma 后端）。
|
||
|
||
覆盖：
|
||
- 未过期 L7 意图被 recall_intentions 召回，layer 标记 l7_intention；
|
||
- 已过期 L7 意图不被召回，且其存储 layer 被持久化改写为 l2_fact（再查确认）；
|
||
- valid_until=None 的意图永不过期，始终召回；
|
||
- intention_limit=0 时 reader 不返回意图（gate）；
|
||
- extractor 新 schema：facts/intentions 解析；writer 不再产 L4_IDENTITY。
|
||
"""
|
||
|
||
import asyncio
|
||
import random
|
||
import uuid
|
||
from datetime import datetime, timedelta
|
||
|
||
import pytest
|
||
|
||
from hy_memory.config import MemoryConfig
|
||
from hy_memory.models.memory import MemoryNode, MemoryLayer, MemoryStatus
|
||
from hy_memory.pipelines._retrieval.intention import recall_intentions
|
||
|
||
|
||
def run(coro):
|
||
loop = asyncio.new_event_loop()
|
||
try:
|
||
return loop.run_until_complete(coro)
|
||
finally:
|
||
loop.close()
|
||
|
||
|
||
def _emb(dims: int = 32):
|
||
vec = [random.gauss(0, 1) for _ in range(dims)]
|
||
norm = sum(v * v for v in vec) ** 0.5 or 1.0
|
||
return [v / norm for v in vec]
|
||
|
||
|
||
def _make_intention(content, *, user_id="u1", valid_until=None, dims=32):
|
||
node = MemoryNode(
|
||
node_id=str(uuid.uuid4()),
|
||
user_id=user_id,
|
||
agent_id="a1",
|
||
session_id="s1",
|
||
content=content,
|
||
layer=MemoryLayer.L7_INTENTION,
|
||
status=MemoryStatus.ACTIVE,
|
||
is_latest=True,
|
||
valid_until=valid_until,
|
||
)
|
||
node.embedding = _emb(dims)
|
||
return node
|
||
|
||
|
||
@pytest.fixture
|
||
def store(tmp_path):
|
||
config = MemoryConfig.from_dict({
|
||
"vector_store": {
|
||
"provider": "chroma",
|
||
"collection_name": f"test_intent_{uuid.uuid4().hex[:8]}",
|
||
"persist_directory": str(tmp_path / "chroma_data"),
|
||
"embedding_dims": 32,
|
||
},
|
||
})
|
||
from hy_memory.data.vector_store_chroma import ChromaVectorStore
|
||
vs = ChromaVectorStore(config)
|
||
run(vs.initialize())
|
||
yield vs
|
||
run(vs.close())
|
||
|
||
|
||
class TestRecallIntentions:
|
||
|
||
def test_unexpired_recalled(self, store):
|
||
future = datetime.now() + timedelta(days=7)
|
||
n = _make_intention("用户下周二要面试字节", valid_until=future)
|
||
run(store.upsert(n))
|
||
q = list(n.embedding) # 同向量保证 cosine=1 命中
|
||
hits = run(recall_intentions(store, q, user_ids=["u1"], limit=5))
|
||
ids = {h["node_id"] for h in hits}
|
||
assert n.node_id in ids
|
||
hit = next(h for h in hits if h["node_id"] == n.node_id)
|
||
assert hit["layer"] == "l7_intention"
|
||
assert hit["retrieval_source"] == "vdb_intention"
|
||
|
||
def test_no_expiry_always_recalled(self, store):
|
||
n = _make_intention("用户打算学习吉他", valid_until=None)
|
||
run(store.upsert(n))
|
||
q = list(n.embedding)
|
||
hits = run(recall_intentions(store, q, user_ids=["u1"], limit=5))
|
||
assert n.node_id in {h["node_id"] for h in hits}
|
||
|
||
def test_expired_excluded_and_converted_to_fact(self, store):
|
||
past = datetime.now() - timedelta(days=3)
|
||
n = _make_intention("用户上周要交报告", valid_until=past)
|
||
run(store.upsert(n))
|
||
q = list(n.embedding)
|
||
# 过期 → 不在意图结果里
|
||
hits = run(recall_intentions(store, q, user_ids=["u1"], limit=5))
|
||
assert n.node_id not in {h["node_id"] for h in hits}
|
||
# 且存储 layer 被持久化改写为 l2_fact
|
||
reloaded = run(store.get_by_id(n.node_id))
|
||
assert reloaded is not None
|
||
assert reloaded.layer == MemoryLayer.L2_FACT
|
||
|
||
def test_limit_zero_returns_empty(self, store):
|
||
n = _make_intention("用户计划去旅行", valid_until=None)
|
||
run(store.upsert(n))
|
||
q = _emb()
|
||
assert run(recall_intentions(store, q, user_ids=["u1"], limit=0)) == []
|
||
|
||
|
||
class TestExtractorSchema:
|
||
"""新 schema：memory→L2_FACT、intentions 单独解析；不再产 L4_IDENTITY。"""
|
||
|
||
def test_collect_new_memories_memory_bucket_to_l2(self):
|
||
from hy_memory.pipelines.writer import MemoryWriter
|
||
info = {
|
||
"memory": [{"content": "用户喜欢篮球", "tags": ["sport"]}],
|
||
"intentions": [{"content": "用户下周面试", "valid_until": "2099-01-01"}],
|
||
}
|
||
texts, meta = MemoryWriter._collect_new_memories(info)
|
||
assert "用户喜欢篮球" in texts
|
||
assert all(m["layer"] == "L2_FACT" for m in meta)
|
||
# intentions 不进 normal memories
|
||
assert "用户下周面试" not in texts
|
||
|
||
def test_collect_new_memories_legacy_facts_to_l2(self):
|
||
"""向后兼容：旧 extractor 的 facts 字段仍可解析。"""
|
||
from hy_memory.pipelines.writer import MemoryWriter
|
||
info = {"facts": [{"content": "用户喜欢篮球", "tags": ["sport"]}]}
|
||
texts, meta = MemoryWriter._collect_new_memories(info)
|
||
assert "用户喜欢篮球" in texts
|
||
assert all(m["layer"] == "L2_FACT" for m in meta)
|
||
|
||
def test_collect_new_memories_legacy_identity_to_l2(self):
|
||
from hy_memory.pipelines.writer import MemoryWriter
|
||
info = {"identity": [{"content": "用户性格内向", "tags": []}]}
|
||
texts, meta = MemoryWriter._collect_new_memories(info)
|
||
assert "用户性格内向" in texts
|
||
assert all(m["layer"] == "L2_FACT" for m in meta)
|
||
|
||
def test_collect_intentions_parses_valid_until(self):
|
||
from hy_memory.pipelines.writer import MemoryWriter
|
||
info = {
|
||
"intentions": [
|
||
{"content": "用户下周面试", "valid_until": "2099-01-01"},
|
||
{"content": "用户想学吉他", "valid_until": None},
|
||
{"content": "坏日期", "valid_until": "not-a-date"},
|
||
]
|
||
}
|
||
out = MemoryWriter._collect_intentions(info)
|
||
assert len(out) == 3
|
||
assert out[0]["valid_until"] is not None
|
||
assert out[0]["valid_until"].year == 2099
|
||
assert out[1]["valid_until"] is None
|
||
assert out[2]["valid_until"] is None # 坏日期降级 None
|
||
|
||
|
||
class TestOwner:
|
||
"""owner 字段：extractor→meta、intention、归一化、持久化 round-trip。"""
|
||
|
||
def test_collect_new_memories_owner(self):
|
||
from hy_memory.pipelines.writer import MemoryWriter
|
||
info = {"memory": [
|
||
{"content": "用户喜欢篮球", "owner": "user"},
|
||
{"content": "助手推荐了川菜馆", "owner": "agent"},
|
||
{"content": "无 owner", "tags": []},
|
||
{"content": "脏 owner", "owner": "robot"},
|
||
]}
|
||
_, meta = MemoryWriter._collect_new_memories(info)
|
||
owners = {m["content"]: m["owner"] for m in meta}
|
||
assert owners["用户喜欢篮球"] == "user"
|
||
assert owners["助手推荐了川菜馆"] == "agent"
|
||
assert owners["无 owner"] is None # 缺失 → None
|
||
assert owners["脏 owner"] is None # 非法 → None
|
||
|
||
def test_collect_new_memories_owner_assistant_alias(self):
|
||
from hy_memory.pipelines.writer import MemoryWriter
|
||
info = {"memory": [{"content": "x", "owner": "assistant"}]}
|
||
_, meta = MemoryWriter._collect_new_memories(info)
|
||
assert meta[0]["owner"] == "agent" # mem0 风格 assistant → agent
|
||
|
||
def test_collect_intentions_owner(self):
|
||
from hy_memory.pipelines.writer import MemoryWriter
|
||
info = {"intentions": [
|
||
{"content": "用户下周面试", "owner": "user"},
|
||
{"content": "8小时后提醒用户", "owner": "agent"},
|
||
]}
|
||
out = MemoryWriter._collect_intentions(info)
|
||
assert out[0]["owner"] == "user"
|
||
assert out[1]["owner"] == "agent"
|
||
|
||
def test_node_owner_roundtrip(self, store):
|
||
n = MemoryNode(
|
||
node_id=str(uuid.uuid4()), user_id="u1", agent_id="a1", session_id="s1",
|
||
content="助手推荐了科比纪录片", layer=MemoryLayer.L2_FACT,
|
||
status=MemoryStatus.ACTIVE, owner="agent",
|
||
)
|
||
n.embedding = _emb()
|
||
run(store.upsert(n))
|
||
got = run(store.get_by_id(n.node_id))
|
||
assert got is not None
|
||
assert got.owner == "agent"
|
||
|
||
def test_node_owner_empty_roundtrip(self, store):
|
||
n = MemoryNode(
|
||
node_id=str(uuid.uuid4()), user_id="u1", agent_id="a1", session_id="s1",
|
||
content="一条 summary", layer=MemoryLayer.L3_SUMMARY,
|
||
status=MemoryStatus.ACTIVE, # owner 留空
|
||
)
|
||
n.embedding = _emb()
|
||
run(store.upsert(n))
|
||
got = run(store.get_by_id(n.node_id))
|
||
assert got is not None
|
||
assert got.owner is None
|