mirror of
https://github.com/norandom/OpenUltraSAST.git
synced 2026-10-16 03:49:26 +00:00
Marius CiepluchandClaude Fable 5.1
d9d6335ab3
hunter-learning-loop groups 1-3 (first increment). The feature table is the contract: one row per touched function with diff, source, semantic and taxonomy columns, provenance per producer, labels by the benchmark's rule. Producers read the repository at head only (a planted advisory id and commit text never reach a feature). The hunt program is a DSPy ReAct loop over a signature whose inputs are the unmasked columns in a fixed order; answers outside the closed family list or citing unseen lines score as unknown. The runner writes records the benchmark scorer reads, resumable by change, program id and mask, and refuses the held-out pilot unless the run is a report run. Research extra [hunt] = dspy, pandas; the core install is unchanged. 43 tests on a scripted model, no network. Co-Authored-By: Claude Fable 5.1 <[email protected]>
159 lines
7.0 KiB
Python
159 lines
7.0 KiB
Python
"""Checks for repository-only annotations on real local git changes."""
|
|
|
|
import json
|
|
import subprocess
|
|
|
|
import pytest
|
|
|
|
from openultrasast import mapping, source
|
|
from openultrasast.hunt import producers as module
|
|
from openultrasast.hunt.producers import PRODUCERS, Context, candidate_rows, compute_rows
|
|
from openultrasast.hunt.table import FeatureTable
|
|
from openultrasast.model.taxonomy import FAMILY_IDS
|
|
|
|
|
|
@pytest.fixture
|
|
def repository(tmp_path):
|
|
def git(*args):
|
|
return subprocess.run(["git", "-C", str(tmp_path), *args], check=True, capture_output=True, text=True).stdout.strip()
|
|
|
|
git("init", "-q")
|
|
git("config", "user.name", "Fixture")
|
|
git("config", "user.email", "[email protected]")
|
|
initial = ('def first():\n value = input()\n return value\n\n'
|
|
'def second():\n value = first()\n return value\n')
|
|
(tmp_path / "app.py").write_text(initial)
|
|
(tmp_path / "notes.txt").write_text("GHSA-xxxx-xxxx-xxxx CVE-2024-0001\n")
|
|
(tmp_path / "unknown.zzz").write_text("opaque\n")
|
|
git("add", "app.py", "notes.txt", "unknown.zzz")
|
|
git("commit", "-qm", "initial")
|
|
base = git("rev-parse", "HEAD")
|
|
(tmp_path / "app.py").write_text(initial.replace("return value", "return eval(value)", 1))
|
|
(tmp_path / "unknown.zzz").write_text("opaque changed\n")
|
|
git("add", "app.py", "unknown.zzz")
|
|
git("commit", "-qm", "fix XSS CVE-2024-0001")
|
|
head = git("rev-parse", "HEAD")
|
|
data = {"id": "change", "repo": "fixture/repo", "base": base, "head": head, "kind": "vulnerability", "fold": 1,
|
|
"path": "app.py", "function": "second", "function_lines": [5, 7], "family": "injection", "case_id": "verified",
|
|
"advisory": "GHSA-xxxx-xxxx-xxxx", "commit_message": "fix XSS CVE-2024-0001",
|
|
"files": [{"path": "app.py", "head_lines": [[3, 3]]}]}
|
|
return tmp_path, data
|
|
|
|
|
|
def test_every_column_candidates_and_cached_work(repository, monkeypatch):
|
|
root, change = repository
|
|
calls = []
|
|
real = mapping.analyze_entry_points
|
|
|
|
def entries(*args):
|
|
calls.append(1)
|
|
return real(*args)
|
|
|
|
monkeypatch.setattr(mapping, "analyze_entry_points", entries)
|
|
candidates = candidate_rows(root, change)
|
|
assert [r["line"] for r in candidates] == [1, 5]
|
|
rows = compute_rows(root, change)
|
|
assert len(calls) == 1
|
|
table = FeatureTable(rows)
|
|
assert len(table.feature_columns()) == 12
|
|
assert [r["label"] for r in rows] == [None, 1]
|
|
first, second = rows
|
|
assert first["feature.diff.hunks"] == [[3, 3]]
|
|
assert first["feature.diff.distance"] == 2
|
|
assert not first["feature.diff.on_changed_line"]
|
|
assert first["feature.source.function"] == "first"
|
|
assert first["feature.source.span"] == [1, 3]
|
|
assert first["feature.source.callers"] == [["app.py", 6]]
|
|
assert first["feature.source.language"] == "python"
|
|
assert isinstance(first["feature.source.entry_points"], list)
|
|
assert first["feature.semantic.roles"]["input"] == "source"
|
|
assert first["feature.semantic.roles"]["eval"] == "sink"
|
|
assert first["feature.semantic.def_use"]["binds"] == [{"line": 2, "name": "value", "names": ["input"], "call_name": "input"}]
|
|
assert {c["name"] for c in first["feature.semantic.def_use"]["calls"]} == {"input", "eval"}
|
|
assert isinstance(first["feature.semantic.obligations"], list)
|
|
assert [f["id"] for f in first["feature.taxonomy.families"]] == list(FAMILY_IDS)
|
|
assert second["feature.diff.distance"] == 2
|
|
for row in rows:
|
|
for producer in PRODUCERS:
|
|
assert set(producer.columns) <= set(row)
|
|
assert row[f"provenance.{producer.name}.ok"] is True
|
|
assert row[f"provenance.{producer.name}.reason"] is None
|
|
assert row[f"provenance.{producer.name}.version"]
|
|
assert row[f"provenance.{producer.name}.bytes_read"] > 0
|
|
ctx = Context(root, change)
|
|
diff = PRODUCERS[0]
|
|
assert diff.compute(root, change, "app.py", 3, ctx)["feature.diff.on_changed_line"]
|
|
assert diff.compute(root, change, "notes.txt", 1, ctx)["feature.diff.distance"] is None
|
|
|
|
|
|
def test_unsupported_language_is_absent_with_reason(repository):
|
|
root, change = repository
|
|
change = {**change, "kind": "ordinary", "files": [{"path": "unknown.zzz", "head_lines": [[1, 1]]}]}
|
|
rows = compute_rows(root, change)
|
|
assert len(rows) == 1
|
|
assert rows[0]["provenance.semantic.ok"] is False
|
|
assert rows[0]["provenance.semantic.reason"] == "language_unsupported"
|
|
assert all(rows[0][column] is None for column in PRODUCERS[2].columns if column.startswith("feature."))
|
|
assert rows[0]["label"] == 0
|
|
FeatureTable(rows)
|
|
|
|
|
|
def test_planted_advisory_and_commit_text_never_enter_features(repository):
|
|
root, change = repository
|
|
rows = compute_rows(root, change)
|
|
values = json.dumps([{k: v for k, v in row.items() if k.startswith("feature.")} for row in rows])
|
|
for planted in ("GHSA-xxxx-xxxx-xxxx", "CVE-2024-0001", "fix XSS"):
|
|
assert planted not in values
|
|
|
|
|
|
def test_producer_exceptions_are_type_only(repository, monkeypatch):
|
|
root, change = repository
|
|
|
|
def failed(*args, **kwargs):
|
|
raise RuntimeError("secret credential")
|
|
|
|
monkeypatch.setattr(source, "facts_for", failed)
|
|
row = compute_rows(root, change)[0]
|
|
assert row["provenance.source.reason"] == "RuntimeError"
|
|
assert row["provenance.source.ok"] is False
|
|
assert all(row[c] is None for c in PRODUCERS[1].columns if c.startswith("feature."))
|
|
assert "secret credential" not in json.dumps(row)
|
|
|
|
|
|
def test_range_rejection_retains_row_and_reason(repository, monkeypatch):
|
|
root, change = repository
|
|
|
|
def failed(*args):
|
|
raise ValueError("private text")
|
|
|
|
monkeypatch.setattr(source, "ranges", failed)
|
|
rows = compute_rows(root, {**change, "kind": "ordinary"})
|
|
assert len(rows) == 1
|
|
assert rows[0]["provenance.source.reason"] == "ValueError"
|
|
assert rows[0]["provenance.source.ok"] is False
|
|
|
|
|
|
def test_private_or_outside_paths_are_not_read(repository):
|
|
root, change = repository
|
|
for path in (".env", "../outside.py"):
|
|
with pytest.raises(ValueError):
|
|
candidate_rows(root, {**change, "files": [{"path": path, "head_lines": [[1, 1]]}]})
|
|
|
|
|
|
def test_head_mismatch_is_not_a_clean_measurement(repository):
|
|
root, change = repository
|
|
rows = compute_rows(root, {**change, "head": change["base"], "kind": "ordinary"})
|
|
assert rows[0]["provenance.diff.ok"] is False
|
|
assert rows[0]["provenance.diff.reason"] == "head_mismatch"
|
|
|
|
|
|
def test_obligation_result_uses_dominance_not_absence_of_findings(repository, monkeypatch):
|
|
root, change = repository
|
|
text = ('from flask import request\n@app.route("/item")\ndef first():\n'
|
|
' return User.query.get(request.args["id"])\n')
|
|
(root / "app.py").write_text(text)
|
|
monkeypatch.setattr(module, "_git", lambda *args: change["head"])
|
|
row = compute_rows(root, {**change, "kind": "ordinary", "files": [{"path": "app.py", "head_lines": [[4, 4]]}]})[0]
|
|
assert row["feature.semantic.obligations"]
|
|
assert row["feature.semantic.obligations"][0]["discharged"] is False
|