@furongjun1999/dsh-memory 0.5.0 → 0.6.0
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/README.md +572 -465
- package/docs/README.md +1 -0
- package/docs/eval/DSH/346/227/245/345/277/227/347/264/242/345/274/225v2_/345/217/202/350/200/203dsh-TUI_v1.0.md +248 -0
- package/docs/eval/DSH/346/227/245/345/277/227/347/264/242/345/274/225/346/225/210/346/236/234/351/252/214/350/257/201_v1.0.md +209 -0
- package/docs/eval/DSH/347/253/257/347/274/272/351/231/267/344/270/223/351/241/271_v1.0.md +254 -0
- package/docs/eval/P1b2_/350/257/273/351/235/242/344/273/243/351/231/205/344/277/256/345/244/215_v1.0.md +124 -0
- package/docs/eval/bench_lingshu_self/bench_self.py +140 -0
- package/docs/eval/bench_lingshu_self/self_bench_result.json +404 -0
- package/docs/eval/bench_lingshu_self//347/201/265/346/236/242/350/207/252/345/272/223/347/253/257/345/210/260/347/253/257/346/243/200/347/264/242/345/256/236/346/265/213_v1.0.md +38 -0
- package/docs/eval//344/270/215/345/217/257/351/235/240/346/200/247/350/220/275/345/234/260_P0_v1.0.md +185 -0
- package/docs/eval//345/217/221/345/270/20306_/344/270/200/351/224/256/351/205/215/347/275/256/344/270/216DSH0172_v1.0.md +171 -0
- package/docs/eval//346/225/205/351/232/234/346/263/250/345/205/245/345/256/236/346/265/213_v1.0.md +422 -0
- package/docs/eval//347/253/257/345/210/260/347/253/257LoCoMoQA/345/220/214/345/217/243/345/276/204/345/257/271/347/205/247_v1.0.md +100 -0
- package/docs/eval//347/253/257/345/210/260/347/253/257/345/271/262/346/211/260/346/261/240/350/257/204/346/265/213_/347/241/256/345/256/232/346/200/247/350/243/201/345/206/263vsLLM_judge_v1.1.md +197 -0
- package/docs/eval//347/274/272/351/231/267/346/214/226/346/216/230_/350/207/252/344/270/273/350/277/255/344/273/243_v1.md +156 -0
- package/docs/eval//347/274/272/351/231/267/346/214/226/346/216/230_/350/207/252/344/270/273/350/277/255/344/273/243_v10.md +210 -0
- package/docs/eval//347/274/272/351/231/267/346/214/226/346/216/230_/350/207/252/344/270/273/350/277/255/344/273/243_v11.md +227 -0
- package/docs/eval//347/274/272/351/231/267/346/214/226/346/216/230_/350/207/252/344/270/273/350/277/255/344/273/243_v12.md +203 -0
- package/docs/eval//347/274/272/351/231/267/346/214/226/346/216/230_/350/207/252/344/270/273/350/277/255/344/273/243_v13.md +233 -0
- package/docs/eval//347/274/272/351/231/267/346/214/226/346/216/230_/350/207/252/344/270/273/350/277/255/344/273/243_v14.md +191 -0
- package/docs/eval//347/274/272/351/231/267/346/214/226/346/216/230_/350/207/252/344/270/273/350/277/255/344/273/243_v15.md +213 -0
- package/docs/eval//347/274/272/351/231/267/346/214/226/346/216/230_/350/207/252/344/270/273/350/277/255/344/273/243_v16.md +214 -0
- package/docs/eval//347/274/272/351/231/267/346/214/226/346/216/230_/350/207/252/344/270/273/350/277/255/344/273/243_v17.md +199 -0
- package/docs/eval//347/274/272/351/231/267/346/214/226/346/216/230_/350/207/252/344/270/273/350/277/255/344/273/243_v18.md +183 -0
- package/docs/eval//347/274/272/351/231/267/346/214/226/346/216/230_/350/207/252/344/270/273/350/277/255/344/273/243_v19.md +207 -0
- package/docs/eval//347/274/272/351/231/267/346/214/226/346/216/230_/350/207/252/344/270/273/350/277/255/344/273/243_v2.md +156 -0
- package/docs/eval//347/274/272/351/231/267/346/214/226/346/216/230_/350/207/252/344/270/273/350/277/255/344/273/243_v3.md +152 -0
- package/docs/eval//347/274/272/351/231/267/346/214/226/346/216/230_/350/207/252/344/270/273/350/277/255/344/273/243_v4.md +128 -0
- package/docs/eval//347/274/272/351/231/267/346/214/226/346/216/230_/350/207/252/344/270/273/350/277/255/344/273/243_v5.md +114 -0
- package/docs/eval//347/274/272/351/231/267/346/214/226/346/216/230_/350/207/252/344/270/273/350/277/255/344/273/243_v6.md +192 -0
- package/docs/eval//347/274/272/351/231/267/346/214/226/346/216/230_/350/207/252/344/270/273/350/277/255/344/273/243_v7.md +187 -0
- package/docs/eval//347/274/272/351/231/267/346/214/226/346/216/230_/350/207/252/344/270/273/350/277/255/344/273/243_v8.md +207 -0
- package/docs/eval//347/274/272/351/231/267/346/214/226/346/216/230_/350/207/252/344/270/273/350/277/255/344/273/243_v9.md +203 -0
- package/docs/hive//345/256/211/345/205/250/345/256/241/350/256/241/345/256/236/351/224/232_v0.1.md +202 -0
- package/docs/hive//350/234/202/345/267/242/345/217/214/345/256/236/344/276/213/344/272/222/351/252/214_/350/256/276/350/256/241/345/256/232/347/250/277.md +19 -3
- package/docs/images/lingshu-moonlight-covenant-poster-preview.jpg +0 -0
- package/docs/images/lingshu-moonlight-covenant-poster.png +0 -0
- package/docs/mdcg//345/212/237/350/203/275/350/260/203/347/224/250/346/230/240/345/260/204/350/241/250_v0.1.md +40 -40
- package/docs/theory//344/270/215/345/217/257/351/235/240/345/256/232/347/220/206/344/270/216/345/244/261/346/225/210/344/274/230/345/205/210/346/241/206/346/236/266_v0.3.md +365 -0
- package/docs//344/270/215/345/217/257/351/235/240/346/200/247/347/220/206/350/256/272_v0.1.md +343 -0
- package/dsh/cordis-patch-profile-web.example.yml +35 -0
- package/lib/bridge.d.ts +9 -0
- package/lib/bridge.js +35 -0
- package/lib/cli.d.ts +3 -0
- package/lib/cli.js +66 -0
- package/lib/index.js +7 -1
- package/lib/init.d.ts +88 -0
- package/lib/init.js +287 -0
- package/lib/lib/roleplay_web.js +530 -443
- package/lib/lib/token_store.d.ts +7 -1
- package/lib/lib/token_store.js +12 -3
- package/md_cg/audit.py +12 -1
- package/md_cg/backfill.py +16 -15
- package/md_cg/backfill_bucket_zh.py +35 -0
- package/md_cg/bench_e2e_judge.py +532 -0
- package/md_cg/bench_e2e_locomo_qa.py +368 -0
- package/md_cg/bench_e2e_qa.py +256 -0
- package/md_cg/branches.py +18 -2
- package/md_cg/ccgc.py +3 -2
- package/md_cg/chain.py +19 -4
- package/md_cg/consolidate.py +7 -6
- package/md_cg/crosscheck.py +4 -3
- package/md_cg/crypto.py +439 -437
- package/md_cg/datapath.py +395 -335
- package/md_cg/evidence.py +8 -3
- package/md_cg/export.py +3 -1
- package/md_cg/forgetting.py +2 -2
- package/md_cg/fsutil.py +48 -0
- package/md_cg/hotcache.py +255 -238
- package/md_cg/interop.py +161 -22
- package/md_cg/judgment_manifest.py +177 -0
- package/md_cg/links.py +655 -622
- package/md_cg/mcp_server.py +282 -31
- package/md_cg/mdcg.py +702 -126
- package/md_cg/mdcos.py +454 -68
- package/md_cg/mreview/govern.py +5 -4
- package/md_cg/postings.py +4 -2
- package/md_cg/readcache.py +76 -18
- package/md_cg/reconcile.py +228 -0
- package/md_cg/review_cli.py +215 -0
- package/md_cg/routing.py +28 -0
- package/md_cg/run_tests.py +211 -0
- package/md_cg/scrub.py +862 -852
- package/md_cg/security.py +385 -275
- package/md_cg/selfreport.py +3 -2
- package/md_cg/signer.py +565 -562
- package/md_cg/sources.py +3 -2
- package/md_cg/stg.py +6 -0
- package/md_cg/sustain.py +1168 -1138
- package/md_cg/test_access_hints.py +147 -0
- package/md_cg/test_branch_discard_tombstone.py +136 -0
- package/md_cg/test_branches.py +259 -249
- package/md_cg/test_chain_read_isolate.py +168 -0
- package/md_cg/test_datapath_device_name.py +203 -0
- package/md_cg/test_emit_negtail_cache.py +156 -0
- package/md_cg/test_en_pipeline.py +186 -166
- package/md_cg/test_govern_directread.py +421 -0
- package/md_cg/test_i32_hotcache_env_key.py +218 -0
- package/md_cg/test_identity_attribution.py +228 -147
- package/md_cg/test_index_crossprocess_reload.py +301 -0
- package/md_cg/test_index_durability.py +238 -224
- package/md_cg/test_interop.py +4 -2
- package/md_cg/test_interop_judgment.py +228 -0
- package/md_cg/test_issue39_utf8_stdio.py +273 -0
- package/md_cg/test_links_concurrent_write.py +188 -0
- package/md_cg/test_merge_upsert.py +168 -0
- package/md_cg/test_n123_derive_expiry_chain.py +205 -0
- package/md_cg/test_n130_verify_falsified_protect.py +185 -0
- package/md_cg/test_n131_merge_gate.py +205 -0
- package/md_cg/test_none_id_write_guard.py +165 -0
- package/md_cg/test_p1x_ref_root.py +160 -0
- package/md_cg/test_p27_docindex.py +774 -765
- package/md_cg/test_p2_mcp.py +3 -0
- package/md_cg/test_p32_backfill.py +304 -298
- package/md_cg/test_p39_verify_flow.py +90 -50
- package/md_cg/test_p47_session_view.py +60 -25
- package/md_cg/test_propose_tail_index.py +157 -0
- package/md_cg/test_read_scope_b27.py +277 -0
- package/md_cg/test_readcache_default_on.py +168 -0
- package/md_cg/test_readcache_precise_inval.py +270 -0
- package/md_cg/test_readcache_prodpath.py +55 -7
- package/md_cg/test_reconcile_v0.py +294 -0
- package/md_cg/test_retr_s1.py +6 -2
- package/md_cg/test_retr_s1b.py +67 -0
- package/md_cg/test_retr_s7.py +8 -0
- package/md_cg/test_retr_s9_entity_ctx.py +181 -175
- package/md_cg/test_retr_score_once.py +208 -0
- package/md_cg/test_review_cli_attribution.py +177 -0
- package/md_cg/test_review_cli_visibility.py +235 -0
- package/md_cg/test_review_onepass.py +170 -0
- package/md_cg/test_rrf_graph_seed_cache.py +154 -0
- package/md_cg/test_security_audit.py +155 -0
- package/md_cg/test_security_audit_b26.py +161 -0
- package/md_cg/test_security_audit_v21.py +250 -0
- package/md_cg/test_semantic_canonical.py +255 -241
- package/md_cg/test_session_isolation.py +168 -0
- package/md_cg/test_snapshot_autoclose.py +187 -0
- package/md_cg/test_tail_watermark_race.py +208 -0
- package/md_cg/test_tenant_env_override_warn.py +139 -0
- package/md_cg/test_tenant_registry_corrupt_warn.py +151 -0
- package/md_cg/test_v14_fixes.py +415 -397
- package/md_cg/test_verify_dirty_reconcile.py +157 -0
- package/md_cg/theory.py +276 -273
- package/md_cg/tokens.py +734 -677
- package/md_cg/units.py +3 -2
- package/md_cg/vision_evidence.py +4 -3
- package/md_cg/whitebox_kb/wisdom/code_solidified.json +6195 -6195
- package/md_cg/whitebox_kb/wisdom/multilang_ir.py +128 -124
- package/md_cg/writepipe.py +554 -550
- package/package.json +11 -2
- package/src/bridge.ts +434 -401
- package/src/cli.ts +65 -0
- package/src/index.ts +526 -518
- package/src/init.ts +347 -0
- package/src/lib/roleplay_web.ts +1019 -932
- package/src/lib/token_store.ts +202 -192
package/md_cg/interop.py
CHANGED
|
@@ -20,13 +20,30 @@ import sys
|
|
|
20
20
|
import time
|
|
21
21
|
|
|
22
22
|
HERE = os.path.dirname(os.path.dirname(os.path.abspath(__file__)))
|
|
23
|
-
|
|
23
|
+
# 判据面实现(2026-09-24 修复):包内模块;出货包不含 scripts/,且 DSH 沙箱
|
|
24
|
+
# 禁子进程管道,故**进程内**调用。MDCG_JUDGMENT_MANIFEST 可指向外部实现文件
|
|
25
|
+
# (自定义判据面时用;按文件路径加载,需提供 collect()/digest())。
|
|
26
|
+
MANIFEST_ENV = "MDCG_JUDGMENT_MANIFEST"
|
|
24
27
|
|
|
25
28
|
_FORBIDDEN_RES = [
|
|
26
|
-
(re.compile(r"[A-Za-z]:[\\\\/]"
|
|
29
|
+
(re.compile(r"[A-Za-z]:[\\\\/]"), "本机绝对路径(盘符)"),
|
|
27
30
|
(re.compile(r"/Users/|/home/"), "本机绝对路径(unix 家目录)"),
|
|
28
|
-
(re.compile(r"sk[-_][A-Za-z0-9]{8,}"), "疑似 API key"),
|
|
29
31
|
(re.compile(r"https?://[^\s\"']*(api|key|token)", re.I), "API 端点/凭证 URL"),
|
|
32
|
+
(re.compile(r"sk[-_][A-Za-z0-9]{8,}"), "疑似 API key"),
|
|
33
|
+
# issue #37 J5 扩面(纵深防御的真实宽度对齐声明):
|
|
34
|
+
(re.compile(r"\\\\[A-Za-z0-9_$.-]+\\"), "本机绝对路径(UNC)"),
|
|
35
|
+
(re.compile(r"/Volumes/"), "本机绝对路径(macOS 挂载)"),
|
|
36
|
+
(re.compile(r"(?:^|[\s\"'(=,])/(?:etc|usr|var|opt|root|tmp|private)/"),
|
|
37
|
+
"本机绝对路径(unix)"),
|
|
38
|
+
(re.compile(r"(?:^|[\s\"'(=,])~/"), "家目录缩写 ~"),
|
|
39
|
+
(re.compile(r"\$HOME"), "家目录变量"),
|
|
40
|
+
(re.compile(r"\b(?:ghp_[A-Za-z0-9]{20,}|gho_[A-Za-z0-9]{20,}"
|
|
41
|
+
r"|github_pat_[A-Za-z0-9_]{20,}|AIza[A-Za-z0-9_-]{20,}"
|
|
42
|
+
r"|AKIA[0-9A-Z]{16}|xox[baprs]-[A-Za-z0-9-]{10,})"),
|
|
43
|
+
"公开密钥形态"),
|
|
44
|
+
(re.compile(r"Bearer\s+[A-Za-z0-9._~+/=-]{20,}"), "Bearer 凭证"),
|
|
45
|
+
(re.compile(r"\b(?:user_prompt|result\.content|spec_body|system_prompt)"
|
|
46
|
+
r"\b\s*[::=]", re.I), "prompt/content 文本形态"),
|
|
30
47
|
]
|
|
31
48
|
_FORBIDDEN_KEYS = {"prompt", "content", "spec", "api_key", "base", "model",
|
|
32
49
|
"system_prompt", "user_prompt", "content_head"}
|
|
@@ -36,27 +53,78 @@ class InteropSanityError(ValueError):
|
|
|
36
53
|
"""verdict 脱敏门禁违规(入库即公开,违规内容不得落盘)。"""
|
|
37
54
|
|
|
38
55
|
|
|
56
|
+
def _manifest_module():
|
|
57
|
+
"""取判据面实现模块,优先级:
|
|
58
|
+
|
|
59
|
+
① `MDCG_JUDGMENT_MANIFEST` 外部实现(自定义判据面)
|
|
60
|
+
② **源码树的 `scripts/judgment_manifest.py`**——验证器侧(hive runner /
|
|
61
|
+
serve 心跳)用它算 digest,A3 要求两侧**同一份实现**;源码树在时
|
|
62
|
+
必须用它,否则两侧判据面构成一旦分叉,digest 必不相等(A3 全红)。
|
|
63
|
+
③ 包内 `md_cg/judgment_manifest.py`——出货包不含 `scripts/`,
|
|
64
|
+
安装态回落此处(构成与 ② 同步维护)。
|
|
65
|
+
"""
|
|
66
|
+
override = (os.environ.get(MANIFEST_ENV) or "").strip()
|
|
67
|
+
if override:
|
|
68
|
+
if not os.path.isfile(override):
|
|
69
|
+
raise RuntimeError(
|
|
70
|
+
f"{MANIFEST_ENV} 指向的文件不存在:{override}")
|
|
71
|
+
return _load_manifest_file(override)
|
|
72
|
+
src = os.path.join(HERE, "scripts", "judgment_manifest.py")
|
|
73
|
+
if os.path.isfile(src):
|
|
74
|
+
return _load_manifest_file(src)
|
|
75
|
+
from . import judgment_manifest as mod
|
|
76
|
+
return mod
|
|
77
|
+
|
|
78
|
+
|
|
79
|
+
def _load_manifest_file(path):
|
|
80
|
+
"""按路径加载判据面实现(需提供 collect()/digest())。"""
|
|
81
|
+
import importlib.util
|
|
82
|
+
spec = importlib.util.spec_from_file_location(
|
|
83
|
+
"_mdcg_judgment_manifest_ext", path)
|
|
84
|
+
mod = importlib.util.module_from_spec(spec)
|
|
85
|
+
spec.loader.exec_module(mod)
|
|
86
|
+
for fn in ("collect", "digest"):
|
|
87
|
+
if not callable(getattr(mod, fn, None)):
|
|
88
|
+
raise RuntimeError(f"{path} 不是合法判据面实现(缺 {fn}())")
|
|
89
|
+
return mod
|
|
90
|
+
|
|
91
|
+
|
|
39
92
|
def _manifest(action: str):
|
|
40
|
-
|
|
41
|
-
|
|
42
|
-
|
|
43
|
-
|
|
44
|
-
|
|
45
|
-
|
|
46
|
-
|
|
93
|
+
"""判据面清单/指纹(进程内,无子进程)。
|
|
94
|
+
|
|
95
|
+
历史(2026-09-24 修复):0.4.x 以 `subprocess.run([python, scripts/
|
|
96
|
+
judgment_manifest.py, ...])` 取清单——出货包不含 scripts/,安装态必抛
|
|
97
|
+
FileNotFoundError(判据面是运行时依赖,不是可选工具);且 DSH 文件沙箱
|
|
98
|
+
禁止子进程管道(CreatePipe → WinError 5),该写法在受限宿主里连空库都跑不动。
|
|
99
|
+
现改为直接调用包内模块,语义(PATTERNS/排序/digest 算法)与原脚本逐字一致。
|
|
100
|
+
"""
|
|
101
|
+
mod = _manifest_module()
|
|
102
|
+
manifest = mod.collect()
|
|
103
|
+
if action == "--digest":
|
|
104
|
+
return mod.digest(manifest)
|
|
105
|
+
return json.dumps(manifest, ensure_ascii=False)
|
|
47
106
|
|
|
48
107
|
|
|
49
108
|
def freeze(iter_id: str, out_dir: str = None) -> dict:
|
|
50
109
|
"""第 0 步:记录本轮判据面冻结凭证(A3 的比对基准)。
|
|
51
110
|
|
|
52
111
|
out_dir 缺省 = hive/interop/<iter_id>/;显式传入时视为最终目录(不拼 iter_id)。
|
|
112
|
+
注意(issue #37 J6):verify_runner 固定读 repo 标准位置——显式 out_dir
|
|
113
|
+
冻结的凭证**不被 runner 识别**(仅适用于单测/离线留痕),此处显式警告。
|
|
53
114
|
"""
|
|
115
|
+
if out_dir is not None:
|
|
116
|
+
sys.stderr.write("[interop] 警告:显式 out_dir 的冻结凭证不被 "
|
|
117
|
+
"verify_runner 识别(runner 固定读 hive/interop/"
|
|
118
|
+
"<iter_id>/frozen.json)——仅适用于单测/离线留痕\n")
|
|
54
119
|
manifest = json.loads(_manifest(""))
|
|
55
120
|
credential = {
|
|
56
121
|
"iter_id": iter_id,
|
|
57
122
|
"frozen_at": time.strftime("%Y-%m-%dT%H:%M:%S%z"),
|
|
58
123
|
"digest": _manifest("--digest"),
|
|
59
124
|
"files": manifest["files"],
|
|
125
|
+
# 判据面是源码树概念:安装态(出货包)只有 md_cg/test_*.py 一组可得。
|
|
126
|
+
# 缺失组随凭证落盘 → A3 两侧布局不一致时可判因,而不是只见「指纹不符」。
|
|
127
|
+
"missing_patterns": manifest.get("missing_patterns") or [],
|
|
60
128
|
}
|
|
61
129
|
if out_dir is None:
|
|
62
130
|
out_dir = os.path.join(HERE, "hive", "interop", iter_id)
|
|
@@ -83,10 +151,15 @@ def assert_a1(verifier_instance: str, subject_instance: str) -> dict:
|
|
|
83
151
|
|
|
84
152
|
|
|
85
153
|
def assert_a2(verifier_fingerprint: str, subject_fingerprint: str) -> dict:
|
|
86
|
-
|
|
87
|
-
|
|
88
|
-
|
|
89
|
-
|
|
154
|
+
"""A2(辅助断言):两实例指纹不同。**subject_fp 为派发方自报值,本断言不核验
|
|
155
|
+
其真伪**(issue #37 J2:复用 #27 的 self-reported 分层标注)——设计定稿
|
|
156
|
+
(§7.5)明示 A1/A2 起两进程即自动成立、不构成保证,防线在 A3(冻结值)。
|
|
157
|
+
"""
|
|
158
|
+
ok = bool(verifier_fingerprint and subject_fingerprint
|
|
159
|
+
and verifier_fingerprint != subject_fingerprint)
|
|
160
|
+
return {"assertion": "A2", "ok": ok,
|
|
161
|
+
"verifier": verifier_fingerprint, "subject": subject_fingerprint,
|
|
162
|
+
"subject_fp_source": "self-reported"}
|
|
90
163
|
|
|
91
164
|
|
|
92
165
|
def sanity_check_verdict(verdict: dict) -> None:
|
|
@@ -110,11 +183,66 @@ def sanity_check_verdict(verdict: dict) -> None:
|
|
|
110
183
|
walk(verdict)
|
|
111
184
|
|
|
112
185
|
|
|
186
|
+
# issue #37 J5:黑名单天然漏——补「字符集 + 形状」白名单校验(与本仓
|
|
187
|
+
# 「能机械判的绝不猜」纪律同源)。verdict 的结构是固定契约,任何越权字段
|
|
188
|
+
# 或坏形状(指纹非十六进制、计数非整数、枚举外值)在落盘前即拒。
|
|
189
|
+
_ALLOWED_TOP = {"iter_id", "verifier_instance", "verifier_fingerprint",
|
|
190
|
+
"subject_instance", "subject_fingerprint", "suite_origin",
|
|
191
|
+
"frozen_at", "verdict", "passed", "failed", "details",
|
|
192
|
+
"valid", "assertions_ok", "failure_reason"}
|
|
193
|
+
_SHAPE_FP = re.compile(r"[0-9a-f]{12,64}")
|
|
194
|
+
|
|
195
|
+
|
|
196
|
+
def shape_check_verdict(verdict: dict) -> None:
|
|
197
|
+
"""形状白名单:顶层字段集 ⊆ 契约集;指纹=12-64 位十六进制;枚举与计数
|
|
198
|
+
类型受检。与 sanity_check_verdict(脱敏黑名单)互补,构成双门禁。"""
|
|
199
|
+
extra = set(verdict) - _ALLOWED_TOP
|
|
200
|
+
if extra:
|
|
201
|
+
raise InteropSanityError(f"未知顶层字段(越权): {sorted(extra)}")
|
|
202
|
+
for k in ("verifier_fingerprint", "subject_fingerprint"):
|
|
203
|
+
v = verdict.get(k)
|
|
204
|
+
if v and not _SHAPE_FP.fullmatch(str(v)):
|
|
205
|
+
raise InteropSanityError(
|
|
206
|
+
f"{k}: 非指纹形状(期望 12-64 位十六进制): {str(v)[:24]!r}")
|
|
207
|
+
if verdict.get("verdict") not in ("pass", "fail"):
|
|
208
|
+
raise InteropSanityError(
|
|
209
|
+
f"verdict 非法枚举: {verdict.get('verdict')!r}")
|
|
210
|
+
if verdict.get("failure_reason") not in (
|
|
211
|
+
None, "suite_failed", "assertions_failed", "both"):
|
|
212
|
+
raise InteropSanityError(
|
|
213
|
+
f"failure_reason 非法枚举: {verdict.get('failure_reason')!r}")
|
|
214
|
+
for k in ("passed", "failed"):
|
|
215
|
+
v = verdict.get(k)
|
|
216
|
+
if not isinstance(v, int) or isinstance(v, bool) or v < 0:
|
|
217
|
+
raise InteropSanityError(f"{k}: 非非负整数: {v!r}")
|
|
218
|
+
|
|
219
|
+
|
|
113
220
|
def make_verdict(iter_id: str, verifier_instance: str, verifier_fingerprint: str,
|
|
114
221
|
subject_instance: str, subject_fingerprint: str,
|
|
115
222
|
suite_origin: str, frozen_at: str, verdict: str,
|
|
116
|
-
passed: int, failed: int, details=None
|
|
117
|
-
|
|
223
|
+
passed: int, failed: int, details=None,
|
|
224
|
+
suite_ok: bool = None) -> dict:
|
|
225
|
+
"""§7.3 契约构造 + 脱敏门禁(违规即抛,绝不落盘)。
|
|
226
|
+
|
|
227
|
+
批次 23(v20 API 一致性):产出**直接含门禁 `_REQUIRED` 全部字段**
|
|
228
|
+
(valid/assertions_ok/failure_reason,J1/J3 语义与
|
|
229
|
+
`verify_runner._compose_semantics` 同口径)——此前 make_verdict 不产
|
|
230
|
+
这三字段、只有 verify_runner 写盘前补齐,直连 make_verdict →
|
|
231
|
+
write_verdict_to_repo 会被门禁判 exit 3(形状非法)。suite_ok 可选:
|
|
232
|
+
调用方知道套件面成败时传入以精确 failure_reason;不传则按可推导部分
|
|
233
|
+
诚实标注(failed>0 → assertions_failed),不编造 suite 面。
|
|
234
|
+
"""
|
|
235
|
+
valid = (verdict == "pass")
|
|
236
|
+
assertions_ok = (int(failed) == 0)
|
|
237
|
+
if valid and assertions_ok:
|
|
238
|
+
failure_reason = None
|
|
239
|
+
elif assertions_ok:
|
|
240
|
+
failure_reason = "suite_failed"
|
|
241
|
+
elif suite_ok is False:
|
|
242
|
+
failure_reason = "both"
|
|
243
|
+
else:
|
|
244
|
+
# failed>0:断言面确定败;suite 面未知(None)或已知真 → 枚举取断言面
|
|
245
|
+
failure_reason = "assertions_failed"
|
|
118
246
|
v = {
|
|
119
247
|
"iter_id": iter_id,
|
|
120
248
|
"verifier_instance": verifier_instance,
|
|
@@ -124,6 +252,9 @@ def make_verdict(iter_id: str, verifier_instance: str, verifier_fingerprint: str
|
|
|
124
252
|
"suite_origin": suite_origin,
|
|
125
253
|
"frozen_at": frozen_at,
|
|
126
254
|
"verdict": verdict,
|
|
255
|
+
"valid": valid,
|
|
256
|
+
"assertions_ok": assertions_ok,
|
|
257
|
+
"failure_reason": failure_reason,
|
|
127
258
|
"passed": passed,
|
|
128
259
|
"failed": failed,
|
|
129
260
|
"details": details or [],
|
|
@@ -144,11 +275,11 @@ def dispatch_verify_job(verifier_jobs_dir: str, iter_id: str,
|
|
|
144
275
|
A2 的比对输入)。
|
|
145
276
|
"""
|
|
146
277
|
jobs = os.path.abspath(verifier_jobs_dir)
|
|
278
|
+
runner = os.path.join(HERE, "hive", "verify_runner.py")
|
|
147
279
|
spec = {
|
|
148
280
|
"model": "cmd",
|
|
149
281
|
"user_prompt": f"互验执行 iter={iter_id}",
|
|
150
|
-
"command": [sys.executable,
|
|
151
|
-
os.path.join(HERE, "hive", "verify_runner.py"), iter_id],
|
|
282
|
+
"command": [sys.executable, runner, iter_id],
|
|
152
283
|
"timeout_s": max(5, min(3600, timeout_s)),
|
|
153
284
|
"env": {"SUBJECT_FP": subject_fingerprint, "ITER_ID": iter_id},
|
|
154
285
|
}
|
|
@@ -159,8 +290,14 @@ def dispatch_verify_job(verifier_jobs_dir: str, iter_id: str,
|
|
|
159
290
|
json.dump(spec, f, ensure_ascii=False)
|
|
160
291
|
with open(os.path.join(jd, "status.json"), "w", encoding="utf-8") as f:
|
|
161
292
|
json.dump({"state": "pending"}, f)
|
|
162
|
-
|
|
163
|
-
|
|
293
|
+
out = {"ok": True, "dispatched": jd, "iter_id": iter_id,
|
|
294
|
+
"note": "非阻塞投递完成——进度靠验证实例心跳拉取,不轮询不等待"}
|
|
295
|
+
# hive/ 是源码树子系统(出货包不含):投递本身成功,但没有可执行 runner 时
|
|
296
|
+
# 消费端会静默失败 → 显式报警,不假装这一票能跑(2026-09-24 补齐)。
|
|
297
|
+
if not os.path.isfile(runner):
|
|
298
|
+
out["warning"] = (f"验证 runner 不存在:{runner}——投递已落盘,但本次"
|
|
299
|
+
"验证不会被消费(hive/ 为源码树子系统,安装态需另行提供)")
|
|
300
|
+
return out
|
|
164
301
|
|
|
165
302
|
|
|
166
303
|
def write_verdict_to_repo(verdict: dict, repo: str = HERE,
|
|
@@ -168,11 +305,13 @@ def write_verdict_to_repo(verdict: dict, repo: str = HERE,
|
|
|
168
305
|
"""§7.3 留痕:verdict 写 hive/interop/<iter_id>/(受版本控制目录,
|
|
169
306
|
**不用** hive/jobs/——该目录 .gitignore 且含运行态)。
|
|
170
307
|
|
|
171
|
-
脱敏门禁在 make_verdict
|
|
172
|
-
|
|
308
|
+
脱敏门禁在 make_verdict 已过;此处落盘前再过一次(纵深)——issue #37 J5
|
|
309
|
+
后为 sanity(脱敏黑名单)+ shape(形状白名单)双门禁。do_commit=True 时
|
|
310
|
+
git add+commit(**不自动 push**——推送由使用者/编排触发,§7.3 双清单
|
|
173
311
|
人工确认环节保留)。
|
|
174
312
|
"""
|
|
175
313
|
sanity_check_verdict(verdict)
|
|
314
|
+
shape_check_verdict(verdict)
|
|
176
315
|
it = verdict.get("iter_id") or ""
|
|
177
316
|
if not re.fullmatch(r"[A-Za-z0-9_.-]{1,80}", it):
|
|
178
317
|
raise ValueError(f"iter_id 非法(入库路径组成部分): {it!r}")
|
|
@@ -0,0 +1,177 @@
|
|
|
1
|
+
# -*- coding: utf-8 -*-
|
|
2
|
+
"""判据面清单生成器(批次8b,互验定稿 §7.6「机器可读清单未定义」缺口收口)。
|
|
3
|
+
|
|
4
|
+
**本模块是 `scripts/judgment_manifest.py` 的包内等价实现**(2026-09-24 修复):
|
|
5
|
+
出货包 `files` 不含 `scripts/`,而 `md_cg/interop.py` 原先以子进程调用该脚本
|
|
6
|
+
——安装态必抛 FileNotFoundError,且受限宿主(DSH 文件沙箱)禁子进程管道
|
|
7
|
+
(CreatePipe → WinError 5)。判据面是**运行时依赖**,故搬进包内。
|
|
8
|
+
|
|
9
|
+
⚠ **判据面构成必须与源码树逐条同源**(A3 的硬约束):验证器侧(hive runner /
|
|
10
|
+
hive serve 心跳)用的是 `scripts/judgment_manifest.py`,两侧 PATTERNS 一旦
|
|
11
|
+
不同,digest 必不相等 → A3 红。因此:
|
|
12
|
+
① 本文件的 PATTERNS / 排序 / digest 算法与 `scripts/` 版**逐字同步**
|
|
13
|
+
(含 issue #36 的七组构成);
|
|
14
|
+
② `md_cg/interop.py` **优先加载源码树的 `scripts/judgment_manifest.py`**,
|
|
15
|
+
只有它不在(安装态)才回落到本模块。
|
|
16
|
+
|
|
17
|
+
判据面 = 验证实例 A3 断言(判据面文件集合 hash==冻结值)的覆盖对象。
|
|
18
|
+
候选面 = hive/src/**(被验证的源码)。
|
|
19
|
+
物理分离成立后:候选弱化判据面任一文件 → 判据面 hash 不变、候选面变化被
|
|
20
|
+
判据面覆盖 → 弱化必红(zcode 外评 break#3 的结构性收口)。
|
|
21
|
+
|
|
22
|
+
用法(包内入口):
|
|
23
|
+
python -m md_cg.judgment_manifest # 输出 JSON(清单+sha256+分组计数)
|
|
24
|
+
python -m md_cg.judgment_manifest --digest # 仅输出组合指纹(纯值,A3 输入)
|
|
25
|
+
python -m md_cg.judgment_manifest --check-coverage # 覆盖完备性守卫(issue #36)
|
|
26
|
+
python -m md_cg.judgment_manifest --verify <冻结的manifest.json>
|
|
27
|
+
|
|
28
|
+
清单组成(§7.6 定稿;issue #36 补齐「跑什么=冻结什么」同源):
|
|
29
|
+
hive/tests/*.rs Rust 侧承重断言(judgment_surface.rs)
|
|
30
|
+
hive/test_*.py hive Python 测试套件
|
|
31
|
+
scripts/run_tests.py 全量测试入口
|
|
32
|
+
scripts/test_*.py scripts 侧测试
|
|
33
|
+
md_cg/test_*.py Python 侧测试套件
|
|
34
|
+
compiler/tests/*.py compiler 测试
|
|
35
|
+
swarm/tests/*.py swarm 测试
|
|
36
|
+
|
|
37
|
+
安装态(出货包)只有 `md_cg/test_*.py` 一组可得——其余组命中数为 0,
|
|
38
|
+
`missing_patterns` 显式随清单与冻结凭证落盘(不静默少算);A3 两侧必须在
|
|
39
|
+
同一种布局下取值,否则 digest 天然不等(见 `md_cg/interop.py` 的说明)。
|
|
40
|
+
"""
|
|
41
|
+
import hashlib
|
|
42
|
+
import json
|
|
43
|
+
import os
|
|
44
|
+
import sys
|
|
45
|
+
|
|
46
|
+
# 包内位置 = <插件根>/md_cg/judgment_manifest.py → 上溯两级回到插件根
|
|
47
|
+
# (与 scripts/ 版的 os.path.dirname(dirname(__file__)) 同值)。
|
|
48
|
+
HERE = os.path.dirname(os.path.dirname(os.path.abspath(__file__)))
|
|
49
|
+
|
|
50
|
+
PATTERNS = [
|
|
51
|
+
("hive/tests", "*.rs"),
|
|
52
|
+
("hive", "test_*.py"),
|
|
53
|
+
("scripts", "run_tests.py"),
|
|
54
|
+
("scripts", "test_*.py"),
|
|
55
|
+
("md_cg", "test_*.py"),
|
|
56
|
+
("compiler/tests", "*.py"),
|
|
57
|
+
("swarm/tests", "*.py"),
|
|
58
|
+
]
|
|
59
|
+
|
|
60
|
+
|
|
61
|
+
# 生效条件:无必需形参;遍历 PATTERNS 逐组取 HERE 下子目录中匹配的文件(目录不存在或零命中则 groups 计数为 0 且该组进 missing_patterns),按相对路径排序后返回 {"algorithm","groups","files","missing_patterns"}。
|
|
62
|
+
def collect() -> dict:
|
|
63
|
+
"""收集判据面清单(分组计数 + 缺失组显式上报)。"""
|
|
64
|
+
files = []
|
|
65
|
+
groups = {}
|
|
66
|
+
for sub, pat in PATTERNS:
|
|
67
|
+
base = os.path.join(HERE, sub)
|
|
68
|
+
n = 0
|
|
69
|
+
if os.path.isdir(base):
|
|
70
|
+
import fnmatch
|
|
71
|
+
for name in sorted(os.listdir(base)):
|
|
72
|
+
full = os.path.join(base, name)
|
|
73
|
+
if not os.path.isfile(full):
|
|
74
|
+
continue
|
|
75
|
+
if fnmatch.fnmatch(name, pat):
|
|
76
|
+
rel = os.path.relpath(full, HERE).replace("\\", "/")
|
|
77
|
+
files.append(rel)
|
|
78
|
+
n += 1
|
|
79
|
+
groups[f"{sub}/{pat}"] = n
|
|
80
|
+
files.sort()
|
|
81
|
+
out = {"algorithm": "sha256", "groups": groups, "files": [],
|
|
82
|
+
# 零命中组(目录不存在、或目录在但空):安装态/裸 clone 的判据面不完整,
|
|
83
|
+
# 必须随清单可见——否则「少算」会被当成「判据面完好」。
|
|
84
|
+
"missing_patterns": sorted(k for k, v in groups.items() if not v)}
|
|
85
|
+
for rel in files:
|
|
86
|
+
with open(os.path.join(HERE, rel), "rb") as fh:
|
|
87
|
+
h = hashlib.sha256(fh.read()).hexdigest()
|
|
88
|
+
out["files"].append({"path": rel, "sha256": h})
|
|
89
|
+
return out
|
|
90
|
+
|
|
91
|
+
|
|
92
|
+
# 生效条件:scripts/run_tests.py 在(源码树)时以 _discovered_files() 为唯一执行清单真源,逐文件判定是否落在冻结 PATTERNS 覆盖域内(域外=collect 永不收=漏冻)返回缺失列表;rt 不在(安装态)时返回 None——调用方须如实说明「无法核对」,不得当作通过。批次 23 域覆盖语义(与 scripts 版同口径,v20 D-36-1:集合差形式在 collect 超集下恒空=假牙)。
|
|
93
|
+
def coverage_gap(manifest: dict):
|
|
94
|
+
"""覆盖完备性守卫(issue #36):「跑什么」⊆「冻结什么」。
|
|
95
|
+
|
|
96
|
+
批次 23 起与 `scripts/` 版同口径——**域覆盖判定**:每个 run_tests
|
|
97
|
+
会跑的文件必须落在冻结 PATTERNS 的覆盖域内(域外 = collect 永不收
|
|
98
|
+
→ 不进判据指纹 → 漏冻)。v20 D-36-1:集合差形式在 collect 超集下
|
|
99
|
+
恒空(假牙)。安装态(run_tests 不在)返回 None——调用方须如实
|
|
100
|
+
说明「无法核对」,不得当作通过。
|
|
101
|
+
"""
|
|
102
|
+
import fnmatch
|
|
103
|
+
import importlib.util
|
|
104
|
+
rt_path = os.path.join(HERE, "scripts", "run_tests.py")
|
|
105
|
+
if not os.path.isfile(rt_path):
|
|
106
|
+
return None
|
|
107
|
+
spec = importlib.util.spec_from_file_location("run_tests", rt_path)
|
|
108
|
+
rt = importlib.util.module_from_spec(spec)
|
|
109
|
+
spec.loader.exec_module(rt)
|
|
110
|
+
gap = []
|
|
111
|
+
for rel in rt._discovered_files():
|
|
112
|
+
rel = str(rel).replace("\\", "/")
|
|
113
|
+
if not any(fnmatch.fnmatch(rel, f"{sub}/{pat}")
|
|
114
|
+
for sub, pat in PATTERNS):
|
|
115
|
+
gap.append(rel)
|
|
116
|
+
return sorted(gap)
|
|
117
|
+
|
|
118
|
+
|
|
119
|
+
# 生效条件:manifest 含 "files"(每项 path/sha256)时返回 sha256("path:hash\n" 按清单序拼接);files 为空时返回空串拼接结果(sha256 of "")。
|
|
120
|
+
def digest(manifest: dict) -> str:
|
|
121
|
+
"""判据面组合指纹(批次10,A3/A2 的输入):sha256("path:hash\\n" 按清单序拼接)。
|
|
122
|
+
|
|
123
|
+
服务端(hive serve 心跳 fingerprint)与验证器共用同一算法——清单变更
|
|
124
|
+
(新增/删除/任一文件改动)必改 digest,缺一不可。
|
|
125
|
+
"""
|
|
126
|
+
lines = "".join(f"{f['path']}:{f['sha256']}\n" for f in manifest["files"])
|
|
127
|
+
return hashlib.sha256(lines.encode("utf-8")).hexdigest()
|
|
128
|
+
|
|
129
|
+
|
|
130
|
+
# 生效条件:argv 为空时打印全量清单 JSON 并返回 0;--digest 打印指纹返回 0;--check-coverage 在源码树缺 run_tests.py 时打印 UNVERIFIABLE 返回 2、否则按缺口 PASS/FAIL 返回 0/1;--verify <path> 比对冻结清单 PASS/FAIL 返回 0/1。
|
|
131
|
+
def main(argv=None) -> int:
|
|
132
|
+
argv = list(sys.argv[1:] if argv is None else argv)
|
|
133
|
+
manifest = collect()
|
|
134
|
+
if argv[:1] == ["--digest"]:
|
|
135
|
+
print(digest(manifest))
|
|
136
|
+
return 0
|
|
137
|
+
if argv[:1] == ["--check-coverage"]:
|
|
138
|
+
gap = coverage_gap(manifest)
|
|
139
|
+
if gap is None:
|
|
140
|
+
print(json.dumps({
|
|
141
|
+
"verdict": "UNVERIFIABLE",
|
|
142
|
+
"reason": "scripts/run_tests.py 不在(安装态)——覆盖完备性守卫"
|
|
143
|
+
"属源码树门禁,本环境无法核对(不冒充通过)",
|
|
144
|
+
"groups": manifest["groups"],
|
|
145
|
+
"file_count": len(manifest["files"]),
|
|
146
|
+
}, ensure_ascii=False, indent=2))
|
|
147
|
+
return 2
|
|
148
|
+
print(json.dumps({
|
|
149
|
+
"verdict": "PASS" if not gap else "FAIL",
|
|
150
|
+
"missing": gap,
|
|
151
|
+
"groups": manifest["groups"],
|
|
152
|
+
"file_count": len(manifest["files"]),
|
|
153
|
+
}, ensure_ascii=False, indent=2))
|
|
154
|
+
return 0 if not gap else 1
|
|
155
|
+
if argv[:1] == ["--verify"]:
|
|
156
|
+
if len(argv) < 2:
|
|
157
|
+
print("用法:--verify <冻结的manifest.json>", file=sys.stderr)
|
|
158
|
+
return 2
|
|
159
|
+
with open(argv[1], encoding="utf-8") as fh:
|
|
160
|
+
frozen = json.load(fh)
|
|
161
|
+
cur = {f["path"]: f["sha256"] for f in manifest["files"]}
|
|
162
|
+
froz = {f["path"]: f["sha256"] for f in frozen["files"]}
|
|
163
|
+
added = sorted(set(cur) - set(froz))
|
|
164
|
+
removed = sorted(set(froz) - set(cur))
|
|
165
|
+
changed = sorted(p for p in set(cur) & set(froz) if cur[p] != froz[p])
|
|
166
|
+
ok = not (added or removed or changed)
|
|
167
|
+
print(json.dumps({
|
|
168
|
+
"verdict": "PASS" if ok else "FAIL",
|
|
169
|
+
"added": added, "removed": removed, "changed": changed,
|
|
170
|
+
}, ensure_ascii=False, indent=2))
|
|
171
|
+
return 0 if ok else 1
|
|
172
|
+
print(json.dumps(manifest, ensure_ascii=False, indent=2))
|
|
173
|
+
return 0
|
|
174
|
+
|
|
175
|
+
|
|
176
|
+
if __name__ == "__main__":
|
|
177
|
+
sys.exit(main())
|