@furongjun1999/dsh-memory 0.4.11 → 0.5.0

package/md_cg/test_p7_goals_recent.py

@@ -1,202 +1,202 @@
-# -*- coding: utf-8 -*-
-"""md_cg · 第 5 篇七件套补齐验收：目标槽 + 近期事件 + 证据计数/降级
-
-覆盖（白箱智能系列·第五篇）：
-  A 目标槽（第 3 章）：写入 / 排序 / 状态机 / 不进正排
-  B 目标定向路（第 3 章）：goal 路召回、显式启用、可审计 goal_used
-  C 近期事件滚动窗口（第 3 章）：最新在前 / 角色过滤 / 滚动截断 / recall 附带
-  D 证据计数 + 可信度降级（第 4 章）：evidence_count、正反例分离、跌破阈值降级
-  E MCP 接口面：cg(op=goal) / cg(op=recent) / mdcg_recall(goal, include_recent)
-  F 健康度覆盖：health_os 报告 goals / recent_events
-
-运行：python -m md_cg.test_p7_goals_recent
-"""
-from __future__ import annotations
-
-import os
-import shutil
-import sys
-import tempfile
-
-from .mdcos import MdCGOS
-from .mcp_server import call_tool
-
-PASS = FAIL = 0
-FAILS = []
-
-
-def check(name, cond, detail=""):
-    global PASS, FAIL
-    if cond:
-        PASS += 1
-        print(f"  [PASS] {name}" + (f"  · {detail}" if detail else ""))
-    else:
-        FAIL += 1
-        FAILS.append(name)
-        print(f"  [FAIL] {name}  · {detail}")
-
-
-def ccg(title, cond="任意情境", sub="验收", exe="直接调用", verify="test", body=""):
-    return (f"# 功能名：{title}\n"
-            f"# 生效条件：{cond}\n"
-            f"# 子功能：{sub}\n"
-            f"# 执行：{exe}\n"
-            f"# 验证方式：{verify}\n"
-            f"# 不适用条件：无\n"
-            f"{body or title}\n")
-
-
-def main():
-    root = tempfile.mkdtemp(prefix="mdcg_p7_")
-    cg = MdCGOS(root)
-    try:
-        # ---------- A. 目标槽 ----------
-        print("\n[A] 目标槽（第 5 篇第 3 章「目标」）")
-        g1 = cg.add_goal("修复登录超时", priority=0.9, deadline="2026-10-01")
-        g2 = cg.add_goal("降低内存占用", priority=0.2)
-        check("add_goal 落 goals 层",
-              (cg.get(g1) or {}).get("frontmatter", {}).get("layer") == "goals", g1)
-        check("目标文本可读回",
-              (cg.get(g1) or {}).get("frontmatter", {}).get("goal_text") == "修复登录超时")
-        check("add_goal 幂等（同文本同 id）", cg.add_goal("修复登录超时", priority=0.9) == g1)
-        act = cg.active_goals()
-        check("active_goals 按优先级排序",
-              [g["id"] for g in act] == [g1, g2], str([g["goal"] for g in act]))
-        check("goal_text 拼接活跃目标",
-              cg.goal_text() == "修复登录超时 降低内存占用", cg.goal_text())
-        cg.set_goal_status(g2, "done")
-        check("done 目标退出 active", [g["id"] for g in cg.active_goals()] == [g1])
-        check("list_goals 仍可见 done 目标",
-              {g["id"] for g in cg.list_goals()} == {g1, g2})
-        res, _m = cg.search("修复登录超时", k=20)
-        check("目标节点不参与正排（只是方向）",
-              g1 not in [n["id"] for n, _s, _q in res])
-
-        # ---------- B. 目标定向路 ----------
-        print("\n[B] 目标定向路（goal 路，显式启用）")
-        for g in cg.list_goals(status="active"):      # 先清空活跃目标
-            cg.set_goal_status(g["id"], "done")
-        a = cg.add("n_login", ccg("登录会话超时处理", cond="登录态",
-                                 body="登录 超时 session token 过期"),
-                   tags=["login"], verification_basis="test")
-        b = cg.add("n_memory", ccg("内存泄漏排查", cond="长时间运行",
-                                  body="内存 泄漏 gc 回收"),
-                   tags=["memory"], verification_basis="test")
-        # 单独启用 goal 路，隔离 lexical 等路的影响
-        _r0, m0 = cg.search_rrf("问题", k=10, paths=("goal",), record=False)
-        check("无活跃目标时 goal 路为空", not m0.get("goal_used") and not _r0,
-              f"goal_used={m0.get('goal_used')!r} n={len(_r0)}")
-        r1, m1 = cg.search_rrf("问题", k=10, paths=("goal",),
-                               record=False, goal_text="登录 超时")
-        ids = [n["id"] for n, _s, _q, _p in r1]
-        check("有目标时 goal 路召回目标相关记忆", a in ids, str(ids))
-        check("goal 路不混入与目标无关的记忆", b not in ids, str(ids))
-        check("meta 暴露 goal_used（可审计）", m1.get("goal_used") == "登录 超时")
-        base, _mb = cg.search_rrf("问题", k=10, paths=("lexical",), record=False)
-        same, _ms = cg.search_rrf("问题", k=10, paths=("lexical",), record=False,
-                                  goal_text="登录 超时")
-        check("未启用 goal 路时目标不生效（既有基线不变）",
-              [n["id"] for n, *_ in base] == [n["id"] for n, *_ in same])
-
-        # ---------- C. 近期事件滚动窗口 ----------
-        print("\n[C] 近期事件滚动窗口（第 5 篇第 3 章「近期事件」）")
-        cg.clear_recent()
-        cg.remember_event("user", "开始重构登录模块")
-        cg.remember_event("assistant", "已定位到 token 过期逻辑")
-        cg.remember_event("command", "pytest -k login")
-        ev = cg.recent_events(limit=10)
-        check("近期事件最新在前", ev[0]["text"] == "pytest -k login", ev[0]["text"])
-        check("近期事件按角色过滤",
-              [e["text"] for e in cg.recent_events(roles=("user",))] == ["开始重构登录模块"])
-        for i in range(6):
-            cg.remember_event("user", f"事件{i}", window=3)
-        ev2 = cg.recent_events(limit=10)
-        check("滚动窗口截断到最近 N 条", len(ev2) == 3, f"n={len(ev2)}")
-        check("窗口保留最新事件", ev2[0]["text"] == "事件5", ev2[0]["text"])
-        p_no = cg.recall("登录超时", budget_tokens=400, include_recent=False)
-        p_yes = cg.recall("登录超时", budget_tokens=400,
-                          include_recent=True, recent_limit=3)
-        check("recall 默认不带近期事件", p_no.get("recent") == [])
-        check("recall(include_recent=True) 附带近期事件",
-              len(p_yes.get("recent") or []) == 3, str(len(p_yes.get("recent") or [])))
-        check("近期事件计入 token 预算（诚实口径）",
-              p_yes["tokens_used"] > p_no["tokens_used"],
-              f"{p_no['tokens_used']} → {p_yes['tokens_used']}")
-
-        # ---------- D. 证据计数 + 可信度降级 ----------
-        print("\n[D] 证据计数 + 可信度降级（第 5 篇第 4 章）")
-        n = cg.add("n_ev", ccg("临时缓存策略", cond="高并发",
-                               body="缓存 过期 时间 30s"),
-                   verification_basis="test", confidence=0.6)
-        r = cg.verify(n, "压测命中率 92%", "confirmed")
-        check("confirmed 累加 evidence_count", r["evidence_count"] == 1, str(r))
-        fm = cg.get(n)["frontmatter"]
-        check("正例计数", fm["positive_evidence"] == 1 and fm["negative_evidence"] == 0,
-              f"pos={fm['positive_evidence']} neg={fm['negative_evidence']}")
-        r = cg.verify(n, "压测命中率降至 71%", "weakened")
-        fm = cg.get(n)["frontmatter"]
-        check("weakened 累加 evidence_count", r["evidence_count"] == 2)
-        check("反例计数", fm["negative_evidence"] == 1)
-        check("反例权重 > 正例（0.6 +0.05 -0.15 = 0.5）",
-              abs(fm["confidence"] - 0.5) < 1e-9, str(fm["confidence"]))
-        demoted = None
-        for i in range(3):
-            r = cg.verify(n, f"继续走弱 {i}", "weakened")
-            demoted = r.get("demoted") or demoted
-        fm = cg.get(n)["frontmatter"]
-        check("跌破阈值后降级为 contextual", fm["layer"] == "contextual",
-              str(fm.get("layer")))
-        check("降级留 demotion 审计",
-              bool(fm.get("demotion")) and fm["demotion"]["from"] == "knowledge",
-              str(fm.get("demotion")))
-        check("降级后文件搬离 knowledge/",
-              not os.path.exists(os.path.join(root, "knowledge", f"{n}.md")))
-        check("索引层同步更新", cg.index["nodes"][n]["layer"] == "contextual")
-
-        # ---------- E. MCP 接口面 ----------
-        print("\n[E] MCP 接口面（cg op=goal / op=recent + mdcg_recall）")
-        out = call_tool(cg, "cg", {"op": "goal", "action": "add",
-                                   "goal": "优化检索延迟", "priority": 0.8})
-        check("cg(op=goal,action=add) 可用", out.get("ok") and out.get("id"), str(out))
-        out = call_tool(cg, "cg", {"op": "goal", "action": "list"})
-        check("cg(op=goal,action=list) 返回目标",
-              any(g["goal"] == "优化检索延迟" for g in out.get("goals", [])),
-              str(out)[:120])
-        out = call_tool(cg, "cg", {"op": "recent", "action": "add",
-                                   "role": "user", "text": "MCP 追加的事件"})
-        check("cg(op=recent,action=add) 可用",
-              out.get("event", {}).get("text") == "MCP 追加的事件", str(out)[:120])
-        out = call_tool(cg, "cg", {"op": "recent", "action": "list", "limit": 1})
-        check("cg(op=recent,action=list) 返回事件",
-              (out.get("events") or [{}])[0].get("text") == "MCP 追加的事件",
-              str(out)[:120])
-        out = call_tool(cg, "mdcg_recall",
-                        {"query": "登录超时", "budget_tokens": 400,
-                         "goal": "登录 超时", "include_recent": True})
-        check("mdcg_recall 支持 goal + include_recent",
-              out.get("meta", {}).get("goal_used") == "登录 超时" and out.get("recent"),
-              f"goal_used={out.get('meta', {}).get('goal_used')!r} "
-              f"recent={len(out.get('recent') or [])}")
-
-        # ---------- F. 健康度覆盖 ----------
-        print("\n[F] 健康度覆盖")
-        h = cg.health_os()
-        check("health_os 报告 goals", h["os"]["goals"]["total"] >= 3,
-              str(h["os"]["goals"]))
-        check("health_os 报告 recent_events", h["os"]["recent_events"] >= 1,
-              str(h["os"]["recent_events"]))
-    finally:
-        try:
-            cg.close()
-        except Exception:  # noqa: BLE001
-            pass
-        shutil.rmtree(root, ignore_errors=True)
-
-    print(f"\n=== PASS {PASS} / FAIL {FAIL} ===")
-    if FAILS:
-        print("失败项：" + ", ".join(FAILS))
-    return 1 if FAIL else 0
-
-
-if __name__ == "__main__":
-    sys.exit(main())
+# -*- coding: utf-8 -*-
+"""md_cg · 第 5 篇七件套补齐验收：目标槽 + 近期事件 + 证据计数/降级
+
+覆盖（白箱智能系列·第五篇）：
+  A 目标槽（第 3 章）：写入 / 排序 / 状态机 / 不进正排
+  B 目标定向路（第 3 章）：goal 路召回、显式启用、可审计 goal_used
+  C 近期事件滚动窗口（第 3 章）：最新在前 / 角色过滤 / 滚动截断 / recall 附带
+  D 证据计数 + 可信度降级（第 4 章）：evidence_count、正反例分离、跌破阈值降级
+  E MCP 接口面：cg(op=goal) / cg(op=recent) / mdcg_recall(goal, include_recent)
+  F 健康度覆盖：health_os 报告 goals / recent_events
+
+运行：python -m md_cg.test_p7_goals_recent
+"""
+from __future__ import annotations
+
+import os
+import shutil
+import sys
+import tempfile
+
+from .mdcos import MdCGOS
+from .mcp_server import call_tool
+
+PASS = FAIL = 0
+FAILS = []
+
+
+def check(name, cond, detail=""):
+    global PASS, FAIL
+    if cond:
+        PASS += 1
+        print(f"  [PASS] {name}" + (f"  · {detail}" if detail else ""))
+    else:
+        FAIL += 1
+        FAILS.append(name)
+        print(f"  [FAIL] {name}  · {detail}")
+
+
+def ccg(title, cond="任意情境", sub="验收", exe="直接调用", verify="test", body=""):
+    return (f"# 功能名：{title}\n"
+            f"# 生效条件：{cond}\n"
+            f"# 子功能：{sub}\n"
+            f"# 执行：{exe}\n"
+            f"# 验证方式：{verify}\n"
+            f"# 不适用条件：无\n"
+            f"{body or title}\n")
+
+
+def main():
+    root = tempfile.mkdtemp(prefix="mdcg_p7_")
+    cg = MdCGOS(root)
+    try:
+        # ---------- A. 目标槽 ----------
+        print("\n[A] 目标槽（第 5 篇第 3 章「目标」）")
+        g1 = cg.add_goal("修复登录超时", priority=0.9, deadline="2026-10-01")
+        g2 = cg.add_goal("降低内存占用", priority=0.2)
+        check("add_goal 落 goals 层",
+              (cg.get(g1) or {}).get("frontmatter", {}).get("layer") == "goals", g1)
+        check("目标文本可读回",
+              (cg.get(g1) or {}).get("frontmatter", {}).get("goal_text") == "修复登录超时")
+        check("add_goal 幂等（同文本同 id）", cg.add_goal("修复登录超时", priority=0.9) == g1)
+        act = cg.active_goals()
+        check("active_goals 按优先级排序",
+              [g["id"] for g in act] == [g1, g2], str([g["goal"] for g in act]))
+        check("goal_text 拼接活跃目标",
+              cg.goal_text() == "修复登录超时 降低内存占用", cg.goal_text())
+        cg.set_goal_status(g2, "done")
+        check("done 目标退出 active", [g["id"] for g in cg.active_goals()] == [g1])
+        check("list_goals 仍可见 done 目标",
+              {g["id"] for g in cg.list_goals()} == {g1, g2})
+        res, _m = cg.search("修复登录超时", k=20)
+        check("目标节点不参与正排（只是方向）",
+              g1 not in [n["id"] for n, _s, _q in res])
+
+        # ---------- B. 目标定向路 ----------
+        print("\n[B] 目标定向路（goal 路，显式启用）")
+        for g in cg.list_goals(status="active"):      # 先清空活跃目标
+            cg.set_goal_status(g["id"], "done")
+        a = cg.add("n_login", ccg("登录会话超时处理", cond="登录态",
+                                 body="登录 超时 session token 过期"),
+                   tags=["login"], verification_basis="test")
+        b = cg.add("n_memory", ccg("内存泄漏排查", cond="长时间运行",
+                                  body="内存 泄漏 gc 回收"),
+                   tags=["memory"], verification_basis="test")
+        # 单独启用 goal 路，隔离 lexical 等路的影响
+        _r0, m0 = cg.search_rrf("问题", k=10, paths=("goal",), record=False)
+        check("无活跃目标时 goal 路为空", not m0.get("goal_used") and not _r0,
+              f"goal_used={m0.get('goal_used')!r} n={len(_r0)}")
+        r1, m1 = cg.search_rrf("问题", k=10, paths=("goal",),
+                               record=False, goal_text="登录 超时")
+        ids = [n["id"] for n, _s, _q, _p in r1]
+        check("有目标时 goal 路召回目标相关记忆", a in ids, str(ids))
+        check("goal 路不混入与目标无关的记忆", b not in ids, str(ids))
+        check("meta 暴露 goal_used（可审计）", m1.get("goal_used") == "登录 超时")
+        base, _mb = cg.search_rrf("问题", k=10, paths=("lexical",), record=False)
+        same, _ms = cg.search_rrf("问题", k=10, paths=("lexical",), record=False,
+                                  goal_text="登录 超时")
+        check("未启用 goal 路时目标不生效（既有基线不变）",
+              [n["id"] for n, *_ in base] == [n["id"] for n, *_ in same])
+
+        # ---------- C. 近期事件滚动窗口 ----------
+        print("\n[C] 近期事件滚动窗口（第 5 篇第 3 章「近期事件」）")
+        cg.clear_recent()
+        cg.remember_event("user", "开始重构登录模块")
+        cg.remember_event("assistant", "已定位到 token 过期逻辑")
+        cg.remember_event("command", "pytest -k login")
+        ev = cg.recent_events(limit=10)
+        check("近期事件最新在前", ev[0]["text"] == "pytest -k login", ev[0]["text"])
+        check("近期事件按角色过滤",
+              [e["text"] for e in cg.recent_events(roles=("user",))] == ["开始重构登录模块"])
+        for i in range(6):
+            cg.remember_event("user", f"事件{i}", window=3)
+        ev2 = cg.recent_events(limit=10)
+        check("滚动窗口截断到最近 N 条", len(ev2) == 3, f"n={len(ev2)}")
+        check("窗口保留最新事件", ev2[0]["text"] == "事件5", ev2[0]["text"])
+        p_no = cg.recall("登录超时", budget_tokens=400, include_recent=False)
+        p_yes = cg.recall("登录超时", budget_tokens=400,
+                          include_recent=True, recent_limit=3)
+        check("recall 默认不带近期事件", p_no.get("recent") == [])
+        check("recall(include_recent=True) 附带近期事件",
+              len(p_yes.get("recent") or []) == 3, str(len(p_yes.get("recent") or [])))
+        check("近期事件计入 token 预算（诚实口径）",
+              p_yes["tokens_used"] > p_no["tokens_used"],
+              f"{p_no['tokens_used']} → {p_yes['tokens_used']}")
+
+        # ---------- D. 证据计数 + 可信度降级 ----------
+        print("\n[D] 证据计数 + 可信度降级（第 5 篇第 4 章）")
+        n = cg.add("n_ev", ccg("临时缓存策略", cond="高并发",
+                               body="缓存 过期 时间 30s"),
+                   verification_basis="test", confidence=0.6)
+        r = cg.verify(n, "压测命中率 92%", "confirmed")
+        check("confirmed 累加 evidence_count", r["evidence_count"] == 1, str(r))
+        fm = cg.get(n)["frontmatter"]
+        check("正例计数", fm["positive_evidence"] == 1 and fm["negative_evidence"] == 0,
+              f"pos={fm['positive_evidence']} neg={fm['negative_evidence']}")
+        r = cg.verify(n, "压测命中率降至 71%", "weakened")
+        fm = cg.get(n)["frontmatter"]
+        check("weakened 累加 evidence_count", r["evidence_count"] == 2)
+        check("反例计数", fm["negative_evidence"] == 1)
+        check("反例权重 > 正例（0.6 +0.05 -0.15 = 0.5）",
+              abs(fm["confidence"] - 0.5) < 1e-9, str(fm["confidence"]))
+        demoted = None
+        for i in range(3):
+            r = cg.verify(n, f"继续走弱 {i}", "weakened")
+            demoted = r.get("demoted") or demoted
+        fm = cg.get(n)["frontmatter"]
+        check("跌破阈值后降级为 contextual", fm["layer"] == "contextual",
+              str(fm.get("layer")))
+        check("降级留 demotion 审计",
+              bool(fm.get("demotion")) and fm["demotion"]["from"] == "knowledge",
+              str(fm.get("demotion")))
+        check("降级后文件搬离 knowledge/",
+              not os.path.exists(os.path.join(root, "knowledge", f"{n}.md")))
+        check("索引层同步更新", cg.index["nodes"][n]["layer"] == "contextual")
+
+        # ---------- E. MCP 接口面 ----------
+        print("\n[E] MCP 接口面（cg op=goal / op=recent + mdcg_recall）")
+        out = call_tool(cg, "cg", {"op": "goal", "action": "add",
+                                   "goal": "优化检索延迟", "priority": 0.8})
+        check("cg(op=goal,action=add) 可用", out.get("ok") and out.get("id"), str(out))
+        out = call_tool(cg, "cg", {"op": "goal", "action": "list"})
+        check("cg(op=goal,action=list) 返回目标",
+              any(g["goal"] == "优化检索延迟" for g in out.get("goals", [])),
+              str(out)[:120])
+        out = call_tool(cg, "cg", {"op": "recent", "action": "add",
+                                   "role": "user", "text": "MCP 追加的事件"})
+        check("cg(op=recent,action=add) 可用",
+              out.get("event", {}).get("text") == "MCP 追加的事件", str(out)[:120])
+        out = call_tool(cg, "cg", {"op": "recent", "action": "list", "limit": 1})
+        check("cg(op=recent,action=list) 返回事件",
+              (out.get("events") or [{}])[0].get("text") == "MCP 追加的事件",
+              str(out)[:120])
+        out = call_tool(cg, "mdcg_recall",
+                        {"query": "登录超时", "budget_tokens": 400,
+                         "goal": "登录 超时", "include_recent": True})
+        check("mdcg_recall 支持 goal + include_recent",
+              out.get("meta", {}).get("goal_used") == "登录 超时" and out.get("recent"),
+              f"goal_used={out.get('meta', {}).get('goal_used')!r} "
+              f"recent={len(out.get('recent') or [])}")
+
+        # ---------- F. 健康度覆盖 ----------
+        print("\n[F] 健康度覆盖")
+        h = cg.health_os()
+        check("health_os 报告 goals", h["os"]["goals"]["total"] >= 3,
+              str(h["os"]["goals"]))
+        check("health_os 报告 recent_events", h["os"]["recent_events"] >= 1,
+              str(h["os"]["recent_events"]))
+    finally:
+        try:
+            cg.close()
+        except Exception:  # noqa: BLE001
+            pass
+        shutil.rmtree(root, ignore_errors=True)
+
+    print(f"\n=== PASS {PASS} / FAIL {FAIL} ===")
+    if FAILS:
+        print("失败项：" + ", ".join(FAILS))
+    return 1 if FAIL else 0
+
+
+if __name__ == "__main__":
+    sys.exit(main())