Spaces:
Sleeping
Sleeping
Commit Β·
12f33ae
1
Parent(s): 216a816
feat(audit): Tier 2 code-soundness checks + selftest fixtures
Browse files- audit/selftest_fixtures.py +122 -0
- audit/tier2_code.py +104 -1
audit/selftest_fixtures.py
CHANGED
|
@@ -88,10 +88,132 @@ def _f_t1_5():
|
|
| 88 |
d.rmdir()
|
| 89 |
|
| 90 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 91 |
FIXTURES.update({
|
| 92 |
"T1.1": _f_t1_1,
|
| 93 |
"T1.2": _f_t1_2,
|
| 94 |
"T1.3": _f_t1_3,
|
| 95 |
"T1.4": _f_t1_4,
|
| 96 |
"T1.5": _f_t1_5,
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 97 |
})
|
|
|
|
| 88 |
d.rmdir()
|
| 89 |
|
| 90 |
|
| 91 |
+
@contextlib.contextmanager
|
| 92 |
+
def _f_t2_1():
|
| 93 |
+
"""Track a .py with a syntax error so T2.1 FAILs."""
|
| 94 |
+
f = REPO / "_audit_selftest_syntax.py"
|
| 95 |
+
f.write_text("def broken(:\n pass\n", encoding="utf-8")
|
| 96 |
+
sh(["git", "add", "-f", "_audit_selftest_syntax.py"])
|
| 97 |
+
try:
|
| 98 |
+
yield
|
| 99 |
+
finally:
|
| 100 |
+
sh(["git", "rm", "--cached", "-q", "_audit_selftest_syntax.py"])
|
| 101 |
+
if f.exists():
|
| 102 |
+
f.unlink()
|
| 103 |
+
|
| 104 |
+
|
| 105 |
+
@contextlib.contextmanager
|
| 106 |
+
def _f_t2_2():
|
| 107 |
+
"""Track backend/_audit_selftest_mod.py that raises NameError at import."""
|
| 108 |
+
f = REPO / "backend" / "_audit_selftest_mod.py"
|
| 109 |
+
f.write_text("x = undefined_thing\n", encoding="utf-8")
|
| 110 |
+
sh(["git", "add", "-f", "backend/_audit_selftest_mod.py"])
|
| 111 |
+
try:
|
| 112 |
+
yield
|
| 113 |
+
finally:
|
| 114 |
+
sh(["git", "rm", "--cached", "-q", "backend/_audit_selftest_mod.py"])
|
| 115 |
+
if f.exists():
|
| 116 |
+
f.unlink()
|
| 117 |
+
|
| 118 |
+
|
| 119 |
+
@contextlib.contextmanager
|
| 120 |
+
def _f_t2_3():
|
| 121 |
+
"""Track backend/_audit_selftest_dead.py with a code ref to a deleted module."""
|
| 122 |
+
f = REPO / "backend" / "_audit_selftest_dead.py"
|
| 123 |
+
f.write_text("from backend.orchestrator import handle_turn\n", encoding="utf-8")
|
| 124 |
+
sh(["git", "add", "-f", "backend/_audit_selftest_dead.py"])
|
| 125 |
+
try:
|
| 126 |
+
yield
|
| 127 |
+
finally:
|
| 128 |
+
sh(["git", "rm", "--cached", "-q", "backend/_audit_selftest_dead.py"])
|
| 129 |
+
if f.exists():
|
| 130 |
+
f.unlink()
|
| 131 |
+
|
| 132 |
+
|
| 133 |
+
@contextlib.contextmanager
|
| 134 |
+
def _f_t2_4():
|
| 135 |
+
"""Track a .css file with a nested */ inside a block comment so T2.4 FAILs."""
|
| 136 |
+
f = REPO / "_audit_selftest.css"
|
| 137 |
+
f.write_text("/* a */ b */\n.x { color: red; }\n", encoding="utf-8")
|
| 138 |
+
sh(["git", "add", "-f", "_audit_selftest.css"])
|
| 139 |
+
try:
|
| 140 |
+
yield
|
| 141 |
+
finally:
|
| 142 |
+
sh(["git", "rm", "--cached", "-q", "_audit_selftest.css"])
|
| 143 |
+
if f.exists():
|
| 144 |
+
f.unlink()
|
| 145 |
+
|
| 146 |
+
|
| 147 |
+
@contextlib.contextmanager
|
| 148 |
+
def _f_t2_5():
|
| 149 |
+
"""Track backend/_audit_selftest_path.py with a hardcoded 40-data path."""
|
| 150 |
+
f = REPO / "backend" / "_audit_selftest_path.py"
|
| 151 |
+
f.write_text('x = something / "40-data" / "y"\n', encoding="utf-8")
|
| 152 |
+
sh(["git", "add", "-f", "backend/_audit_selftest_path.py"])
|
| 153 |
+
try:
|
| 154 |
+
yield
|
| 155 |
+
finally:
|
| 156 |
+
sh(["git", "rm", "--cached", "-q", "backend/_audit_selftest_path.py"])
|
| 157 |
+
if f.exists():
|
| 158 |
+
f.unlink()
|
| 159 |
+
|
| 160 |
+
|
| 161 |
+
def _ruff_available() -> bool:
|
| 162 |
+
return (REPO / ".venv" / "bin" / "ruff").exists()
|
| 163 |
+
|
| 164 |
+
|
| 165 |
+
def _tsc_available() -> bool:
|
| 166 |
+
return (REPO / "frontend" / "node_modules" / ".bin" / "tsc").exists()
|
| 167 |
+
|
| 168 |
+
|
| 169 |
+
@contextlib.contextmanager
|
| 170 |
+
def _f_t2_6():
|
| 171 |
+
"""Force a lint/type failure so T2.6 FAILs.
|
| 172 |
+
|
| 173 |
+
Prefer ruff (a flagrant unused-import + bare-except .py under audit/). If
|
| 174 |
+
ruff is unavailable in this env T2.6 would SKIP (not FAIL) on a ruff-only
|
| 175 |
+
fixture, so fall back to forcing a tsc error via a broken .ts under
|
| 176 |
+
frontend/src/. If NEITHER ruff nor tsc is available this single fixture
|
| 177 |
+
cannot legitimately force a FAIL β raise a clear RuntimeError so selftest
|
| 178 |
+
surfaces it rather than reporting a false pass.
|
| 179 |
+
"""
|
| 180 |
+
have_ruff = _ruff_available()
|
| 181 |
+
have_tsc = _tsc_available()
|
| 182 |
+
if not have_ruff and not have_tsc:
|
| 183 |
+
raise RuntimeError("T2.6 selftest needs ruff or tsc")
|
| 184 |
+
|
| 185 |
+
created = []
|
| 186 |
+
try:
|
| 187 |
+
if have_ruff:
|
| 188 |
+
f = REPO / "audit" / "_audit_selftest_lint.py"
|
| 189 |
+
f.write_text("import os\ntry:\n pass\nexcept:\n pass\n",
|
| 190 |
+
encoding="utf-8")
|
| 191 |
+
sh(["git", "add", "-f", "audit/_audit_selftest_lint.py"])
|
| 192 |
+
created.append("audit/_audit_selftest_lint.py")
|
| 193 |
+
else:
|
| 194 |
+
f = REPO / "frontend" / "src" / "_audit_selftest_bad.ts"
|
| 195 |
+
f.write_text("export const x: number = ;\n", encoding="utf-8")
|
| 196 |
+
sh(["git", "add", "-f", "frontend/src/_audit_selftest_bad.ts"])
|
| 197 |
+
created.append("frontend/src/_audit_selftest_bad.ts")
|
| 198 |
+
yield
|
| 199 |
+
finally:
|
| 200 |
+
for rel in created:
|
| 201 |
+
sh(["git", "rm", "--cached", "-q", rel])
|
| 202 |
+
fp = REPO / rel
|
| 203 |
+
if fp.exists():
|
| 204 |
+
fp.unlink()
|
| 205 |
+
|
| 206 |
+
|
| 207 |
FIXTURES.update({
|
| 208 |
"T1.1": _f_t1_1,
|
| 209 |
"T1.2": _f_t1_2,
|
| 210 |
"T1.3": _f_t1_3,
|
| 211 |
"T1.4": _f_t1_4,
|
| 212 |
"T1.5": _f_t1_5,
|
| 213 |
+
"T2.1": _f_t2_1,
|
| 214 |
+
"T2.2": _f_t2_2,
|
| 215 |
+
"T2.3": _f_t2_3,
|
| 216 |
+
"T2.4": _f_t2_4,
|
| 217 |
+
"T2.5": _f_t2_5,
|
| 218 |
+
"T2.6": _f_t2_6,
|
| 219 |
})
|
audit/tier2_code.py
CHANGED
|
@@ -1 +1,104 @@
|
|
| 1 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
"""Tier 2 β code soundness (pre-commit)."""
|
| 2 |
+
from __future__ import annotations
|
| 3 |
+
import ast, re
|
| 4 |
+
from audit.core import register, Result, Status, REPO, git, sh
|
| 5 |
+
|
| 6 |
+
PY = [p for p in git("ls-files").splitlines() if p.endswith(".py")]
|
| 7 |
+
DEAD = ("backend.orchestrator", "import sales_brain", "qa_brain", "faithfulness",
|
| 8 |
+
"backend.translator", "profile_extractor", "get_judge_llm", "get_fast_brain_llm")
|
| 9 |
+
|
| 10 |
+
|
| 11 |
+
@register("T2.1", "static", "all .py parse (AST)")
|
| 12 |
+
def t2_1() -> Result:
|
| 13 |
+
bad = []
|
| 14 |
+
for p in PY:
|
| 15 |
+
try:
|
| 16 |
+
ast.parse((REPO / p).read_text(encoding="utf-8", errors="replace"), p)
|
| 17 |
+
except SyntaxError as e:
|
| 18 |
+
bad.append(f"{p}: {e}")
|
| 19 |
+
return (Result("T2.1", Status.FAIL, "; ".join(bad[:5]), "fix the syntax error")
|
| 20 |
+
if bad else Result("T2.1", Status.PASS, f"{len(PY)} files parse"))
|
| 21 |
+
|
| 22 |
+
|
| 23 |
+
@register("T2.2", "static", "runtime-import every backend/rag module")
|
| 24 |
+
def t2_2() -> Result:
|
| 25 |
+
mods = []
|
| 26 |
+
for p in PY:
|
| 27 |
+
if (p.startswith("backend/") or p.startswith("rag/")) and not p.endswith("__init__.py") \
|
| 28 |
+
and "_smoke_test" not in p and "/tests/" not in p:
|
| 29 |
+
mods.append(p[:-3].replace("/", "."))
|
| 30 |
+
code = "import importlib,sys\nbad=[]\n" + \
|
| 31 |
+
"".join(f"try:\n importlib.import_module({m!r})\nexcept Exception as e:\n bad.append(({m!r},repr(e)))\n"
|
| 32 |
+
for m in mods) + "print(bad)\nsys.exit(1 if bad else 0)"
|
| 33 |
+
r = sh([".venv/bin/python", "-c", code], timeout=300)
|
| 34 |
+
if r.returncode != 0:
|
| 35 |
+
return Result("T2.2", Status.FAIL, r.stdout.strip()[:600],
|
| 36 |
+
"fix the import (often: import wrongly placed inside a docstring)")
|
| 37 |
+
return Result("T2.2", Status.PASS, f"{len(mods)} modules import clean")
|
| 38 |
+
|
| 39 |
+
|
| 40 |
+
@register("T2.3", "static", "no refs to deleted modules/symbols")
|
| 41 |
+
def t2_3() -> Result:
|
| 42 |
+
code_hits, doc_hits = [], []
|
| 43 |
+
for p in PY:
|
| 44 |
+
for i, ln in enumerate(open(REPO / p, encoding="utf-8", errors="replace"), 1):
|
| 45 |
+
for d in DEAD:
|
| 46 |
+
if d in ln:
|
| 47 |
+
s = ln.strip()
|
| 48 |
+
(doc_hits if s.startswith("#") or s.startswith(('"', "'", "*")) else code_hits
|
| 49 |
+
).append(f"{p}:{i} {d}")
|
| 50 |
+
if code_hits:
|
| 51 |
+
return Result("T2.3", Status.FAIL, "; ".join(code_hits[:6]),
|
| 52 |
+
"remove/replace the dead reference (e.g. get_judge_llm -> get_brain_llm)")
|
| 53 |
+
if doc_hits:
|
| 54 |
+
return Result("T2.3", Status.WARN, f"{len(doc_hits)} stale comment refs e.g. {doc_hits[:3]}",
|
| 55 |
+
"tidy the stale comment")
|
| 56 |
+
return Result("T2.3", Status.PASS, "no dead-symbol references")
|
| 57 |
+
|
| 58 |
+
|
| 59 |
+
@register("T2.4", "static", "no */ inside CSS/JS block-comment body")
|
| 60 |
+
def t2_4() -> Result:
|
| 61 |
+
bad = []
|
| 62 |
+
for p in git("ls-files").splitlines():
|
| 63 |
+
if not p.endswith((".css", ".scss")):
|
| 64 |
+
continue
|
| 65 |
+
txt = (REPO / p).read_text(encoding="utf-8", errors="replace")
|
| 66 |
+
for m in re.finditer(r"/\*.*?\*/", txt, re.S):
|
| 67 |
+
body = m.group(0)[2:-2]
|
| 68 |
+
if "*/" in body:
|
| 69 |
+
bad.append(f"{p}: nested */ in comment")
|
| 70 |
+
return (Result("T2.4", Status.FAIL, "; ".join(bad), "space the token: '* /' or reword")
|
| 71 |
+
if bad else Result("T2.4", Status.PASS, "no comment-terminator footgun"))
|
| 72 |
+
|
| 73 |
+
|
| 74 |
+
@register("T2.5", "static", "no hardcoded 40-data path construction")
|
| 75 |
+
def t2_5() -> Result:
|
| 76 |
+
pat = re.compile(r'/\s*["\']40-data["\']')
|
| 77 |
+
bad = []
|
| 78 |
+
for p in PY:
|
| 79 |
+
if not (p.startswith("backend/") or p.startswith("rag/")):
|
| 80 |
+
continue
|
| 81 |
+
if p.endswith("config.py"):
|
| 82 |
+
continue
|
| 83 |
+
for i, ln in enumerate(open(REPO / p, encoding="utf-8", errors="replace"), 1):
|
| 84 |
+
if pat.search(ln) and not ln.strip().startswith("#"):
|
| 85 |
+
bad.append(f"{p}:{i}")
|
| 86 |
+
return (Result("T2.5", Status.FAIL, "; ".join(bad[:8]), "use settings.DATA_DIR")
|
| 87 |
+
if bad else Result("T2.5", Status.PASS, "DATA_DIR centralized"))
|
| 88 |
+
|
| 89 |
+
|
| 90 |
+
@register("T2.6", "static", "ruff + tsc clean")
|
| 91 |
+
def t2_6() -> Result:
|
| 92 |
+
ruff = sh([".venv/bin/ruff", "check", "backend", "rag", "audit"], timeout=120)
|
| 93 |
+
tsc = sh(["npx", "--prefix", "frontend", "--no-install", "tsc", "-p", "frontend", "--noEmit"], timeout=240)
|
| 94 |
+
probs = []
|
| 95 |
+
if ruff.returncode not in (0, 127):
|
| 96 |
+
tail = (ruff.stdout or ruff.stderr).strip().splitlines()
|
| 97 |
+
probs.append("ruff: " + (tail[-1][:200] if tail else "error"))
|
| 98 |
+
if tsc.returncode not in (0, 127):
|
| 99 |
+
tail = (tsc.stdout or tsc.stderr).strip().splitlines()
|
| 100 |
+
probs.append("tsc: " + (tail[-1][:200] if tail else "error"))
|
| 101 |
+
if 127 in (ruff.returncode, tsc.returncode) and not probs:
|
| 102 |
+
return Result("T2.6", Status.SKIP, "ruff/tsc not installed", "pip install ruff / npm i")
|
| 103 |
+
return (Result("T2.6", Status.FAIL, " | ".join(probs), "fix lint/type errors")
|
| 104 |
+
if probs else Result("T2.6", Status.PASS, "ruff + tsc clean"))
|