patchsight 0.3.0__tar.gz → 0.4.0__tar.gz

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.
Files changed (70) hide show
  1. {patchsight-0.3.0 → patchsight-0.4.0}/PKG-INFO +12 -3
  2. {patchsight-0.3.0 → patchsight-0.4.0}/docs/pypi.md +11 -2
  3. {patchsight-0.3.0 → patchsight-0.4.0}/pyproject.toml +1 -1
  4. {patchsight-0.3.0 → patchsight-0.4.0}/pyproject.toml.orig +1 -1
  5. {patchsight-0.3.0 → patchsight-0.4.0}/src/patchsight/cli.py +13 -1
  6. {patchsight-0.3.0 → patchsight-0.4.0}/src/patchsight/db.py +4 -0
  7. {patchsight-0.3.0 → patchsight-0.4.0}/src/patchsight/pipeline/__main__.py +123 -4
  8. {patchsight-0.3.0 → patchsight-0.4.0}/src/patchsight/pipeline/benchmarks/extraction.py +1 -0
  9. {patchsight-0.3.0 → patchsight-0.4.0}/src/patchsight/pipeline/extract/extractor.py +2 -1
  10. {patchsight-0.3.0 → patchsight-0.4.0}/src/patchsight/pipeline/extract/validate.py +24 -4
  11. {patchsight-0.3.0 → patchsight-0.4.0}/src/patchsight/pipeline/gate.py +20 -0
  12. {patchsight-0.3.0 → patchsight-0.4.0}/src/patchsight/pipeline/llm/config.py +9 -0
  13. {patchsight-0.3.0 → patchsight-0.4.0}/src/patchsight/pipeline/llm_records.py +76 -0
  14. {patchsight-0.3.0 → patchsight-0.4.0}/src/patchsight/pipeline/pypi_extract.py +1 -1
  15. patchsight-0.4.0/src/patchsight/scan/sarif.py +177 -0
  16. {patchsight-0.3.0 → patchsight-0.4.0}/LICENSE +0 -0
  17. {patchsight-0.3.0 → patchsight-0.4.0}/src/patchsight/__init__.py +0 -0
  18. {patchsight-0.3.0 → patchsight-0.4.0}/src/patchsight/eval/__init__.py +0 -0
  19. {patchsight-0.3.0 → patchsight-0.4.0}/src/patchsight/eval/confidence.py +0 -0
  20. {patchsight-0.3.0 → patchsight-0.4.0}/src/patchsight/eval/failures.py +0 -0
  21. {patchsight-0.3.0 → patchsight-0.4.0}/src/patchsight/eval/metrics.py +0 -0
  22. {patchsight-0.3.0 → patchsight-0.4.0}/src/patchsight/eval/parity.py +0 -0
  23. {patchsight-0.3.0 → patchsight-0.4.0}/src/patchsight/eval/safety.py +0 -0
  24. {patchsight-0.3.0 → patchsight-0.4.0}/src/patchsight/names.py +0 -0
  25. {patchsight-0.3.0 → patchsight-0.4.0}/src/patchsight/pipeline/__init__.py +0 -0
  26. {patchsight-0.3.0 → patchsight-0.4.0}/src/patchsight/pipeline/benchmarks/__init__.py +0 -0
  27. {patchsight-0.3.0 → patchsight-0.4.0}/src/patchsight/pipeline/benchmarks/bundle.py +0 -0
  28. {patchsight-0.3.0 → patchsight-0.4.0}/src/patchsight/pipeline/benchmarks/changedfuncs.py +0 -0
  29. {patchsight-0.3.0 → patchsight-0.4.0}/src/patchsight/pipeline/benchmarks/dunlap.py +0 -0
  30. {patchsight-0.3.0 → patchsight-0.4.0}/src/patchsight/pipeline/benchmarks/replay.py +0 -0
  31. {patchsight-0.3.0 → patchsight-0.4.0}/src/patchsight/pipeline/benchmarks/splits.py +0 -0
  32. {patchsight-0.3.0 → patchsight-0.4.0}/src/patchsight/pipeline/build_db.py +0 -0
  33. {patchsight-0.3.0 → patchsight-0.4.0}/src/patchsight/pipeline/diffs/__init__.py +0 -0
  34. {patchsight-0.3.0 → patchsight-0.4.0}/src/patchsight/pipeline/diffs/commits.py +0 -0
  35. {patchsight-0.3.0 → patchsight-0.4.0}/src/patchsight/pipeline/diffs/github.py +0 -0
  36. {patchsight-0.3.0 → patchsight-0.4.0}/src/patchsight/pipeline/diffs/hunks.py +0 -0
  37. {patchsight-0.3.0 → patchsight-0.4.0}/src/patchsight/pipeline/diffs/mapping.py +0 -0
  38. {patchsight-0.3.0 → patchsight-0.4.0}/src/patchsight/pipeline/diffs/paths.py +0 -0
  39. {patchsight-0.3.0 → patchsight-0.4.0}/src/patchsight/pipeline/extract/__init__.py +0 -0
  40. {patchsight-0.3.0 → patchsight-0.4.0}/src/patchsight/pipeline/extract/prompt.py +0 -0
  41. {patchsight-0.3.0 → patchsight-0.4.0}/src/patchsight/pipeline/extract/schema.py +0 -0
  42. {patchsight-0.3.0 → patchsight-0.4.0}/src/patchsight/pipeline/fix_sample.py +0 -0
  43. {patchsight-0.3.0 → patchsight-0.4.0}/src/patchsight/pipeline/http.py +0 -0
  44. {patchsight-0.3.0 → patchsight-0.4.0}/src/patchsight/pipeline/ingest/__init__.py +0 -0
  45. {patchsight-0.3.0 → patchsight-0.4.0}/src/patchsight/pipeline/ingest/aliases.py +0 -0
  46. {patchsight-0.3.0 → patchsight-0.4.0}/src/patchsight/pipeline/ingest/ghsa_repo.py +0 -0
  47. {patchsight-0.3.0 → patchsight-0.4.0}/src/patchsight/pipeline/ingest/osv.py +0 -0
  48. {patchsight-0.3.0 → patchsight-0.4.0}/src/patchsight/pipeline/ingest/references.py +0 -0
  49. {patchsight-0.3.0 → patchsight-0.4.0}/src/patchsight/pipeline/ingest/snapshot.py +0 -0
  50. {patchsight-0.3.0 → patchsight-0.4.0}/src/patchsight/pipeline/kaggle_inputs.py +0 -0
  51. {patchsight-0.3.0 → patchsight-0.4.0}/src/patchsight/pipeline/llm/__init__.py +0 -0
  52. {patchsight-0.3.0 → patchsight-0.4.0}/src/patchsight/pipeline/llm/client.py +0 -0
  53. {patchsight-0.3.0 → patchsight-0.4.0}/src/patchsight/pipeline/llm/tokens.py +0 -0
  54. {patchsight-0.3.0 → patchsight-0.4.0}/src/patchsight/pipeline/pypi.py +0 -0
  55. {patchsight-0.3.0 → patchsight-0.4.0}/src/patchsight/pipeline/update_db.py +0 -0
  56. {patchsight-0.3.0 → patchsight-0.4.0}/src/patchsight/pipeline/week1.py +0 -0
  57. {patchsight-0.3.0 → patchsight-0.4.0}/src/patchsight/py.typed +0 -0
  58. {patchsight-0.3.0 → patchsight-0.4.0}/src/patchsight/scan/__init__.py +0 -0
  59. {patchsight-0.3.0 → patchsight-0.4.0}/src/patchsight/scan/ast_scanner.py +0 -0
  60. {patchsight-0.3.0 → patchsight-0.4.0}/src/patchsight/scan/dbcache.py +0 -0
  61. {patchsight-0.3.0 → patchsight-0.4.0}/src/patchsight/scan/explain.py +0 -0
  62. {patchsight-0.3.0 → patchsight-0.4.0}/src/patchsight/scan/lockfiles/__init__.py +0 -0
  63. {patchsight-0.3.0 → patchsight-0.4.0}/src/patchsight/scan/lockfiles/poetry.py +0 -0
  64. {patchsight-0.3.0 → patchsight-0.4.0}/src/patchsight/scan/lockfiles/requirements.py +0 -0
  65. {patchsight-0.3.0 → patchsight-0.4.0}/src/patchsight/scan/lockfiles/toml_lock.py +0 -0
  66. {patchsight-0.3.0 → patchsight-0.4.0}/src/patchsight/scan/lockfiles/uv.py +0 -0
  67. {patchsight-0.3.0 → patchsight-0.4.0}/src/patchsight/scan/osv_client.py +0 -0
  68. {patchsight-0.3.0 → patchsight-0.4.0}/src/patchsight/scan/report.py +0 -0
  69. {patchsight-0.3.0 → patchsight-0.4.0}/src/patchsight/scan/run.py +0 -0
  70. {patchsight-0.3.0 → patchsight-0.4.0}/src/patchsight/scan/verdicts.py +0 -0
@@ -1,6 +1,6 @@
1
1
  Metadata-Version: 2.4
2
2
  Name: patchsight
3
- Version: 0.3.0
3
+ Version: 0.4.0
4
4
  Summary: Know which vulnerable Python dependencies your code actually calls: symbol-level triage, no LLM at scan time.
5
5
  Keywords: security,vulnerability,dependencies,supply-chain,sca,static-analysis,reachability,osv,pypi,cli
6
6
  Author: Niroj33
@@ -62,17 +62,26 @@ patchsight scan . # scan the lockfile and the code under .
62
62
  |:---|:---|
63
63
  | `--lockfile PATH` | Scan a specific lockfile: `requirements.txt`, `uv.lock` or `poetry.lock` (default: the first found, in that order) |
64
64
  | `--format json` | Machine-readable output |
65
+ | `--format sarif` | SARIF 2.1.0 for GitHub code scanning and other SARIF readers |
65
66
  | `--fail-on called` | Exit 1 if any alert is CALLED: a CI gate (`unknown` also fails on UNKNOWN) |
66
67
  | `--db PATH` | Use a specific database file (or set `PATCHSIGHT_DB`) |
67
68
  | `patchsight explain ID` | Why the database lists an advisory's symbols: confidence, models, rationale and the changed lines |
68
69
  | `patchsight db info` | Show which database a scan uses, and its age |
69
70
 
71
+ In GitHub Actions, the [PatchSight action](https://github.com/professor3333/PatchSight#in-ci-the-github-action) scans and uploads the SARIF report to code scanning:
72
+
73
+ ```yaml
74
+ - uses: professor3333/PatchSight@v0.4.0
75
+ with:
76
+ fail-on: called
77
+ ```
78
+
70
79
  ## Verdicts
71
80
 
72
81
  | | Verdict | Meaning |
73
82
  |:---:|:---|:---|
74
83
  | 🔴 | **CALLED** | A reference in your code resolves to a vulnerable function, its class, or an entry point that reaches it. **Fix these first.** |
75
- | 🟢 | **IMPORTED, NOT CALLED** | The package is imported, but no vulnerable symbol is referenced. Given only for high-confidence data. |
84
+ | 🟢 | **IMPORTED, NOT CALLED** | The package is imported, but no vulnerable symbol is referenced. Given only for high-confidence data, which is [withheld for now](https://github.com/professor3333/PatchSight/blob/main/docs/adr/0008-withhold-high-confidence.md). |
76
85
  | ⚪ | **NOT IMPORTED** | Your code never imports the package directly. It may still be used through other dependencies. |
77
86
  | 🟡 | **UNKNOWN** | No symbol data, only low-confidence data, or dynamic code the scanner can't follow. |
78
87
 
@@ -93,7 +102,7 @@ The F1 95% interval is 0.62 to 0.74. The reports and a one-command offline repro
93
102
  ## Status: early release
94
103
 
95
104
  - Reads `requirements.txt` (exact `==` pins only), `uv.lock` and `poetry.lock`. Packages without a PyPI release version (unpinned requirements, git, URL or local sources) are listed, not guessed.
96
- - The [symbol database](https://github.com/professor3333/PatchSight/releases) is updated every night and covers about 3,500 PyPI vulnerabilities, all at **low confidence** so far: every function the fix changed, plus the model's picks where the backfill has reached. Expect many UNKNOWN verdicts until high-confidence records are published.
105
+ - The [symbol database](https://github.com/professor3333/PatchSight/releases) is updated every night and covers about 3,500 PyPI vulnerabilities, all at **low confidence** so far: every function the fix changed, plus the model's picks where the backfill has reached. High-confidence records are withheld until the confidence rule meets its safety target, so expect UNKNOWN where a later release may say IMPORTED, NOT CALLED.
97
106
  - Transitive use and dynamic code (`getattr` with a computed name, `importlib`, star imports) aren't analysed. They give UNKNOWN or NOT IMPORTED, never a false "not called".
98
107
 
99
108
  ## Links
@@ -24,17 +24,26 @@ patchsight scan . # scan the lockfile and the code under .
24
24
  |:---|:---|
25
25
  | `--lockfile PATH` | Scan a specific lockfile: `requirements.txt`, `uv.lock` or `poetry.lock` (default: the first found, in that order) |
26
26
  | `--format json` | Machine-readable output |
27
+ | `--format sarif` | SARIF 2.1.0 for GitHub code scanning and other SARIF readers |
27
28
  | `--fail-on called` | Exit 1 if any alert is CALLED: a CI gate (`unknown` also fails on UNKNOWN) |
28
29
  | `--db PATH` | Use a specific database file (or set `PATCHSIGHT_DB`) |
29
30
  | `patchsight explain ID` | Why the database lists an advisory's symbols: confidence, models, rationale and the changed lines |
30
31
  | `patchsight db info` | Show which database a scan uses, and its age |
31
32
 
33
+ In GitHub Actions, the [PatchSight action](https://github.com/professor3333/PatchSight#in-ci-the-github-action) scans and uploads the SARIF report to code scanning:
34
+
35
+ ```yaml
36
+ - uses: professor3333/PatchSight@v0.4.0
37
+ with:
38
+ fail-on: called
39
+ ```
40
+
32
41
  ## Verdicts
33
42
 
34
43
  | | Verdict | Meaning |
35
44
  |:---:|:---|:---|
36
45
  | 🔴 | **CALLED** | A reference in your code resolves to a vulnerable function, its class, or an entry point that reaches it. **Fix these first.** |
37
- | 🟢 | **IMPORTED, NOT CALLED** | The package is imported, but no vulnerable symbol is referenced. Given only for high-confidence data. |
46
+ | 🟢 | **IMPORTED, NOT CALLED** | The package is imported, but no vulnerable symbol is referenced. Given only for high-confidence data, which is [withheld for now](https://github.com/professor3333/PatchSight/blob/main/docs/adr/0008-withhold-high-confidence.md). |
38
47
  | ⚪ | **NOT IMPORTED** | Your code never imports the package directly. It may still be used through other dependencies. |
39
48
  | 🟡 | **UNKNOWN** | No symbol data, only low-confidence data, or dynamic code the scanner can't follow. |
40
49
 
@@ -55,7 +64,7 @@ The F1 95% interval is 0.62 to 0.74. The reports and a one-command offline repro
55
64
  ## Status: early release
56
65
 
57
66
  - Reads `requirements.txt` (exact `==` pins only), `uv.lock` and `poetry.lock`. Packages without a PyPI release version (unpinned requirements, git, URL or local sources) are listed, not guessed.
58
- - The [symbol database](https://github.com/professor3333/PatchSight/releases) is updated every night and covers about 3,500 PyPI vulnerabilities, all at **low confidence** so far: every function the fix changed, plus the model's picks where the backfill has reached. Expect many UNKNOWN verdicts until high-confidence records are published.
67
+ - The [symbol database](https://github.com/professor3333/PatchSight/releases) is updated every night and covers about 3,500 PyPI vulnerabilities, all at **low confidence** so far: every function the fix changed, plus the model's picks where the backfill has reached. High-confidence records are withheld until the confidence rule meets its safety target, so expect UNKNOWN where a later release may say IMPORTED, NOT CALLED.
59
68
  - Transitive use and dynamic code (`getattr` with a computed name, `importlib`, star imports) aren't analysed. They give UNKNOWN or NOT IMPORTED, never a false "not called".
60
69
 
61
70
  ## Links
@@ -1,6 +1,6 @@
1
1
  [project]
2
2
  name = "patchsight"
3
- version = "0.3.0"
3
+ version = "0.4.0"
4
4
  description = "Know which vulnerable Python dependencies your code actually calls: symbol-level triage, no LLM at scan time."
5
5
  readme = "docs/pypi.md"
6
6
  license = "Apache-2.0"
@@ -1,6 +1,6 @@
1
1
  [project]
2
2
  name = "patchsight"
3
- version = "0.3.0"
3
+ version = "0.4.0"
4
4
  description = "Know which vulnerable Python dependencies your code actually calls: symbol-level triage, no LLM at scan time."
5
5
  readme = "docs/pypi.md"
6
6
  license = "Apache-2.0"
@@ -102,7 +102,9 @@ def scan(
102
102
  Path | None,
103
103
  typer.Option("--db", help="Symbol database (default: the one 'db update' downloaded)."),
104
104
  ] = None,
105
- output: Annotated[str, typer.Option("--format", help="text or json.")] = "text",
105
+ output: Annotated[
106
+ str, typer.Option("--format", help="text, json, or sarif (for code scanning).")
107
+ ] = "text",
106
108
  fail_on: Annotated[
107
109
  str, typer.Option(help="Exit 1 when an alert is: called, unknown (or called), or never.")
108
110
  ] = "never",
@@ -131,6 +133,16 @@ def scan(
131
133
  shown = lockfile.as_posix()
132
134
  if output == "json":
133
135
  typer.echo(render_json(alerts, dependencies, shown, meta))
136
+ elif output == "sarif":
137
+ import os
138
+
139
+ from patchsight.scan.sarif import render_sarif
140
+
141
+ # Code locations are relative to the scanned directory; SARIF wants them relative to
142
+ # where the scan runs (the repository root, in CI).
143
+ root = path if path.is_dir() else path.parent
144
+ base = os.path.relpath(root.resolve(), Path.cwd().resolve())
145
+ typer.echo(render_sarif(alerts, dependencies, shown, meta, base))
134
146
  else:
135
147
  typer.echo(render_text(alerts, dependencies, shown, meta))
136
148
  failing = {"called": {"CALLED"}, "unknown": {"CALLED", "UNKNOWN"}}.get(fail_on, set())
@@ -295,6 +295,10 @@ class SymbolDatabase:
295
295
  ).fetchone()
296
296
  return str(row[0]) if row else None
297
297
 
298
+ def all_records(self) -> list[Record]:
299
+ """Every record, ordered by vulnerability and package."""
300
+ return [r for v in self.vulnerabilities() for r in self.records_for(v)]
301
+
298
302
  def confidence_counts(self) -> dict[str, int]:
299
303
  """Records by confidence tier."""
300
304
  rows = self._connection.execute(
@@ -619,6 +619,13 @@ def confidence_report_command(
619
619
  primary: Annotated[str, typer.Option(help="Extractor run tag (replay bundle).")],
620
620
  secondary: Annotated[str, typer.Option(help="Second-opinion run tag (replay bundle).")],
621
621
  part: Annotated[str, typer.Option("--split", help="Split part: dev, test or all.")] = "dev",
622
+ variant: Annotated[
623
+ str | None,
624
+ typer.Option(
625
+ help="Experiment variant: read entry points from the reports ending in -VARIANT, "
626
+ "and write the report under that suffix."
627
+ ),
628
+ ] = None,
622
629
  root: Root = Path(),
623
630
  ) -> None:
624
631
  """Compare confidence rules, in the benchmark and safety views, from committed files."""
@@ -635,8 +642,9 @@ def confidence_report_command(
635
642
  for tag in (primary, secondary):
636
643
  run = replay.load(root / DUNLAP_BUNDLE.parent / f"extraction-{tag}.json")
637
644
  # Validated entry points come from the run's committed report (any part that has them).
645
+ suffix = f"{tag}-{variant}" if variant else tag
638
646
  entry: dict[str, frozenset[str]] = {}
639
- for source in sorted(reports_dir.glob(f"extraction-dunlap-*-{tag}.json")):
647
+ for source in sorted(reports_dir.glob(f"extraction-dunlap-*-{suffix}.json")):
640
648
  for outcome in extraction.ExtractionRun.model_validate_json(
641
649
  source.read_text("utf-8")
642
650
  ).outcomes:
@@ -678,6 +686,11 @@ def confidence_report_command(
678
686
  f"- Extractor: `{runs[primary][0]}` (`{primary}`)",
679
687
  f"- Second opinion: `{runs[secondary][0]}` (`{secondary}`)",
680
688
  f"- Split `{frozen.split.name}:{part}` (SHA-256 `{frozen.sha256}`), {len(first)} reports",
689
+ *(
690
+ [f"- Variant `{variant}`: entry points from the `-{variant}` extraction reports"]
691
+ if variant
692
+ else []
693
+ ),
681
694
  "",
682
695
  "Two views. The **benchmark view** compares function names exactly and ignores entry "
683
696
  "points, so it stays comparable with Dunlap et al. The **safety view** scores what the "
@@ -727,12 +740,13 @@ def confidence_report_command(
727
740
  reasons = ", ".join(f"{k} {v}" for k, v in r.low_reasons.items())
728
741
  lines.append(f"- `{r.rule}`{', public' if r.require_public else ''}: {reasons}")
729
742
  lines.append("")
730
- target = reports_dir / f"confidence-dunlap-{part}-{primary}-vs-{secondary}"
743
+ name = f"confidence-dunlap-{part}-{primary}-vs-{secondary}" + (f"-{variant}" if variant else "")
744
+ target = reports_dir / name
731
745
  target.with_name(target.name + ".md").write_text("\n".join(lines), "utf-8")
732
746
  target.with_name(target.name + ".json").write_text(
733
747
  json.dumps([r.model_dump(mode="json") for r in results], indent=2) + "\n", "utf-8"
734
748
  )
735
- typer.echo("\n".join(lines[10:]))
749
+ typer.echo("\n".join(lines[10 + bool(variant) :]))
736
750
 
737
751
 
738
752
  @app.command("parity-report")
@@ -1036,6 +1050,69 @@ def update_db_command(
1036
1050
  typer.echo(f"LLM merge: {merged}")
1037
1051
 
1038
1052
 
1053
+ @app.command("merge-llm")
1054
+ def merge_llm_command(
1055
+ database: Annotated[Path, typer.Argument(help="The database to merge into.")],
1056
+ version: Annotated[str, typer.Option(help="Version of the result, e.g. 2026.11.17.1.")],
1057
+ llm: Annotated[
1058
+ list[Path], typer.Option(help="Extractor outcomes (pypi-extract JSONL); repeatable.")
1059
+ ],
1060
+ second: Annotated[
1061
+ list[Path] | None, typer.Option(help="Second-opinion outcomes; repeatable.")
1062
+ ] = None,
1063
+ root: Root = Path(),
1064
+ out: Annotated[Path, typer.Option(help="Output path, without extension.")] = Path(
1065
+ "data/build/patchsight-merged"
1066
+ ),
1067
+ ) -> None:
1068
+ """Merge new LLM extractions into an existing database (the backfill, shard by shard)."""
1069
+ from patchsight.pipeline import llm_records, pypi_extract
1070
+
1071
+ source = db.SymbolDatabase(root / database)
1072
+ try:
1073
+ records, aliases = source.all_records(), source.alias_map()
1074
+ builds, meta = list(source.build_ledger().values()), dict(source.meta)
1075
+ finally:
1076
+ source.close()
1077
+ if not builds:
1078
+ raise typer.BadParameter("the database has no builds ledger; build it with build-db")
1079
+ primaries: dict[str, pypi_extract.PyPIExtraction] = {}
1080
+ for path in llm:
1081
+ primaries.update(llm_records.load(root / path))
1082
+ secondaries: dict[str, pypi_extract.PyPIExtraction] = {}
1083
+ for path in second or []:
1084
+ secondaries.update(llm_records.load(root / path))
1085
+ merged, outcomes = llm_records.apply(records, primaries, secondaries)
1086
+ report = llm_records.merge_report(
1087
+ records, merged, primaries, secondaries, outcomes, meta.get("db_version", "?"), version
1088
+ )
1089
+ commit = subprocess.run( # noqa: S603 - fixed argv, no shell
1090
+ ["git", "-C", str(root), "rev-parse", "HEAD"], # noqa: S607
1091
+ capture_output=True, text=True, check=False,
1092
+ ).stdout.strip() # fmt: skip
1093
+ meta.pop("updated_from", None) # the source's own origin; this build is a merge
1094
+ meta.update(
1095
+ {
1096
+ "db_version": version,
1097
+ "built_at": datetime.now(UTC).isoformat(),
1098
+ "builder": "llm-v1",
1099
+ "code_commit": commit,
1100
+ "merged_into": meta.get("db_version", "?"),
1101
+ "records_by_tier": json.dumps(dict(Counter(r.confidence for r in merged))),
1102
+ }
1103
+ )
1104
+ meta.pop("schema_version", None) # db.write sets it
1105
+ target = root / out
1106
+ db.write(target.with_name(target.name + ".sqlite"), merged, aliases, meta, builds=builds)
1107
+ target.with_name(target.name + "-merge.json").write_text(
1108
+ report.model_dump_json(indent=2) + "\n", "utf-8"
1109
+ )
1110
+ typer.echo(
1111
+ f"db {version} merged into {meta['merged_into']}: {len(merged)} records; "
1112
+ f"{dict(Counter(r.confidence for r in merged))}; {outcomes}"
1113
+ )
1114
+
1115
+
1039
1116
  @app.command("gate-db")
1040
1117
  def gate_db_command(
1041
1118
  candidate: Annotated[Path, typer.Argument(help="The candidate database.")],
@@ -1077,6 +1154,11 @@ CC BY 4.0.
1077
1154
  lines to function names with tree-sitter, and, where shown, had language models choose the \
1078
1155
  vulnerable ones. It stores those names with file paths, line ranges and commit SHAs. No \
1079
1156
  third-party source code is included.
1157
+
1158
+ **Licence:** the PatchSight symbol database is licensed under \
1159
+ [CC BY 4.0](https://creativecommons.org/licenses/by/4.0/). If you share it, or data derived from \
1160
+ it, credit "PatchSight symbol database" with the release, keep the attribution above, and say \
1161
+ what you changed.
1080
1162
  """
1081
1163
 
1082
1164
 
@@ -1087,6 +1169,7 @@ def release_notes_command(
1087
1169
  "releases/latest.json"
1088
1170
  ),
1089
1171
  update: Annotated[Path | None, typer.Option(help="The update-db report, if any.")] = None,
1172
+ merge: Annotated[Path | None, typer.Option(help="The merge-llm report, if any.")] = None,
1090
1173
  gate: Annotated[Path | None, typer.Option(help="The gate-db report, if any.")] = None,
1091
1174
  root: Root = Path(),
1092
1175
  out: Annotated[Path, typer.Option(help="Notes file to write.")] = Path(
@@ -1095,6 +1178,7 @@ def release_notes_command(
1095
1178
  ) -> None:
1096
1179
  """Write release notes for a database release, with provenance and attribution."""
1097
1180
  from patchsight.pipeline.gate import GateReport
1181
+ from patchsight.pipeline.llm_records import MergeReport
1098
1182
  from patchsight.pipeline.update_db import UpdateReport
1099
1183
 
1100
1184
  opened = db.SymbolDatabase(root / database)
@@ -1105,7 +1189,13 @@ def release_notes_command(
1105
1189
  opened.close()
1106
1190
  published = dbcache.Pointer.model_validate_json((root / pointer).read_text("utf-8"))
1107
1191
  version = meta.get("db_version", "?")
1108
- origin = f", updated from {meta['updated_from']}" if "updated_from" in meta else ""
1192
+ origin = (
1193
+ f", updated from {meta['updated_from']}"
1194
+ if "updated_from" in meta
1195
+ else f", {meta['merged_into']} with new model picks"
1196
+ if "merged_into" in meta
1197
+ else ""
1198
+ )
1109
1199
  lines = [
1110
1200
  f"Symbol database **{version}**{origin}.",
1111
1201
  "",
@@ -1113,6 +1203,35 @@ def release_notes_command(
1113
1203
  "patchsight db update # downloads this release and verifies its SHA-256",
1114
1204
  "```",
1115
1205
  "",
1206
+ ]
1207
+ if merge is not None:
1208
+ merged = MergeReport.model_validate_json((root / merge).read_text("utf-8"))
1209
+ second = (
1210
+ "; second opinion " + ", ".join(f"`{m}`" for m in merged.second_opinion_models)
1211
+ if merged.second_opinion_models
1212
+ else ""
1213
+ )
1214
+ lines += [
1215
+ f"## Changes since {merged.previous_version}",
1216
+ "",
1217
+ "- Merges model picks from "
1218
+ + ", ".join(f"`{m}`" for m in merged.models)
1219
+ + " (prompt "
1220
+ + ", ".join(f"`{p}`" for p in merged.prompt_versions)
1221
+ + f"){second}.",
1222
+ f"- {merged.after.vulnerabilities:,} vulnerabilities ({merged.after.records:,} "
1223
+ "records) now carry model picks with a rationale and cited hunks, up from "
1224
+ f"{merged.before.vulnerabilities:,} ({merged.before.records:,} records).",
1225
+ "- No record is high confidence yet, so matching still uses every changed "
1226
+ "definition. `patchsight explain <id>` shows the picks."
1227
+ if not merged.tiers.get("high")
1228
+ else f"- {merged.tiers['high']:,} "
1229
+ + ("record is" if merged.tiers["high"] == 1 else "records are")
1230
+ + " high confidence. `patchsight explain <id>` shows the picks and why each tier "
1231
+ "was chosen.",
1232
+ "",
1233
+ ]
1234
+ lines += [
1116
1235
  "## Contents",
1117
1236
  "",
1118
1237
  f"- **{vulnerabilities:,} PyPI vulnerabilities**, {total:,} records: "
@@ -143,6 +143,7 @@ def run(
143
143
  counter,
144
144
  config.budget.max_hunk_lines,
145
145
  limit,
146
+ config.validation.dotted_entry_points,
146
147
  )
147
148
  outcomes.append(outcome)
148
149
  truth[report.id] = set(report.symbols)
@@ -118,6 +118,7 @@ def extract(
118
118
  counter: TokenCounter,
119
119
  max_hunk_lines: int,
120
120
  max_candidates: int | None = None,
121
+ dotted_entry_points: bool = False,
121
122
  ) -> ExtractionOutcome:
122
123
  prompt = render(
123
124
  template, advisory, inputs, budget_tokens, counter, max_hunk_lines, max_candidates
@@ -156,7 +157,7 @@ def extract(
156
157
  continue
157
158
  responses.append(response)
158
159
  try:
159
- result = validate(response.content, prompt, known_names)
160
+ result = validate(response.content, prompt, known_names, dotted_entry_points)
160
161
  break
161
162
  except ValidationFailedError as failure:
162
163
  if attempt == 0:
@@ -4,7 +4,10 @@
4
4
  * Every candidate gets exactly one judgement. Unknown IDs are counted as invented.
5
5
  * Every vulnerable judgement cites at least one hunk, and every cited hunk must be shown
6
6
  under that candidate.
7
- * Entry points survive only if they name a definition known from the fetched source.
7
+ * Entry points survive only if they name a definition known from the fetched source:
8
+ exactly, or, when ``dotted`` is on, by matching exactly one definition on a dotted boundary
9
+ (``SetValue`` names ``Paved.SetValue``; ``yaml.load`` names ``load``). An ambiguous name is
10
+ dropped. A resolved entry point is kept as the definition's own name.
8
11
  """
9
12
 
10
13
  import json
@@ -13,6 +16,7 @@ from collections.abc import Set
13
16
 
14
17
  from pydantic import BaseModel, ConfigDict, ValidationError
15
18
 
19
+ from patchsight.eval.safety import lenient_match
16
20
  from patchsight.pipeline.extract.prompt import RenderedPrompt
17
21
  from patchsight.pipeline.extract.schema import Extraction
18
22
 
@@ -34,7 +38,19 @@ class ValidatedExtraction(BaseModel):
34
38
  entry_points_dropped: tuple[str, ...]
35
39
 
36
40
 
37
- def validate(content: str, prompt: RenderedPrompt, known_names: Set[str]) -> ValidatedExtraction:
41
+ def resolve_entry_point(name: str, allowed: Set[str], dotted: bool) -> str | None:
42
+ """The definition an entry point names, or None when it names none (or several)."""
43
+ if name in allowed:
44
+ return name
45
+ if not dotted:
46
+ return None
47
+ matches = [d for d in allowed if lenient_match(name, d)]
48
+ return matches[0] if len(matches) == 1 else None
49
+
50
+
51
+ def validate(
52
+ content: str, prompt: RenderedPrompt, known_names: Set[str], dotted_entry_points: bool = False
53
+ ) -> ValidatedExtraction:
38
54
  try:
39
55
  extraction = Extraction.model_validate(json.loads(content))
40
56
  except (json.JSONDecodeError, ValidationError) as error:
@@ -69,8 +85,12 @@ def validate(content: str, prompt: RenderedPrompt, known_names: Set[str]) -> Val
69
85
  raise ValidationFailedError(errors, len(invented))
70
86
 
71
87
  allowed = set(known_names) | {c.name for c in prompt.candidates}
72
- kept = tuple(sorted({e for e in extraction.public_entry_points if e in allowed}))
73
- dropped = tuple(sorted({e for e in extraction.public_entry_points if e not in allowed}))
88
+ resolved = {
89
+ e: resolve_entry_point(e, allowed, dotted_entry_points)
90
+ for e in extraction.public_entry_points
91
+ }
92
+ kept = tuple(sorted({d for d in resolved.values() if d is not None}))
93
+ dropped = tuple(sorted({e for e, d in resolved.items() if d is None}))
74
94
  judged = {item.candidate_id: item for item in extraction.items}
75
95
  return ValidatedExtraction(
76
96
  vulnerable_ids=tuple(c.id for c in prompt.candidates if judged[c.id].is_vulnerable),
@@ -13,12 +13,15 @@ workflow then opens an issue with this report instead.
13
13
  ``MAX_LOST_SHARE`` of the previous ones, or ``MIN_LOST_ALLOWED``, whichever is larger).
14
14
  A bigger drop means a broken snapshot or builder, not real withdrawals.
15
15
  * ``high_kept``: a high-confidence record whose extraction inputs did not change stays high.
16
+ * ``high_withheld``: no record is high confidence while ADR 0008 withholds the tier, however
17
+ the database was built.
16
18
  """
17
19
 
18
20
  from pydantic import BaseModel, ConfigDict
19
21
 
20
22
  from patchsight.db import SymbolDatabase
21
23
  from patchsight.pipeline.build_db import MAX_SYMBOLS
24
+ from patchsight.pipeline.llm_records import PUBLISH_HIGH
22
25
 
23
26
  MAX_LOST_SHARE = 0.02
24
27
  MIN_LOST_ALLOWED = 10
@@ -126,6 +129,23 @@ def run(candidate: SymbolDatabase, previous: SymbolDatabase) -> GateReport:
126
129
  + (f"; demoted: {_sample(demoted)}" if demoted else ""),
127
130
  )
128
131
  )
132
+
133
+ high = [f"{r.vulnerability}/{r.package}" for rs in records.values() for r in rs
134
+ if r.confidence == "high"] # fmt: skip
135
+ checks.append(
136
+ Check(
137
+ name="high_withheld",
138
+ passed=PUBLISH_HIGH or not high,
139
+ detail=f"{len(high)} high-confidence records"
140
+ + (
141
+ ""
142
+ if PUBLISH_HIGH
143
+ else " (ADR 0008 withholds the tier)"
144
+ if not high
145
+ else f" while ADR 0008 withholds the tier: {_sample(high)}"
146
+ ),
147
+ )
148
+ )
129
149
  return GateReport(
130
150
  candidate_version=candidate.meta.get("db_version", "?"),
131
151
  previous_version=previous.meta.get("db_version", "?"),
@@ -42,6 +42,14 @@ class BudgetConfig(BaseModel):
42
42
  output_tokens_per_candidate: int | None = None
43
43
 
44
44
 
45
+ class ValidationConfig(BaseModel):
46
+ model_config = ConfigDict(frozen=True)
47
+
48
+ # Accept an entry point that matches no definition exactly but exactly one on a dotted
49
+ # boundary (``SetValue`` for ``Paved.SetValue``), kept as that definition's name.
50
+ dotted_entry_points: bool = False
51
+
52
+
45
53
  class LLMConfig(BaseModel):
46
54
  model_config = ConfigDict(frozen=True)
47
55
 
@@ -49,6 +57,7 @@ class LLMConfig(BaseModel):
49
57
  models: dict[str, ModelConfig]
50
58
  budget: BudgetConfig
51
59
  tokenizer: TokenizerConfig
60
+ validation: ValidationConfig = ValidationConfig()
52
61
 
53
62
  def prompt_budget(self, role: str) -> int:
54
63
  """Largest prompt, in exact tokens, that still fits the repair round under the caps.
@@ -11,16 +11,32 @@ extractor has a valid answer made from the same inputs:
11
11
 
12
12
  An extractor answer that failed validation, or was made from different inputs (its
13
13
  ``input_hash`` no longer matches the advisory), leaves the fallback record unchanged.
14
+
15
+ **High confidence is withheld** (ADR 0008): the rule missed its safety target in its declared
16
+ test evaluation, so a record the rule would make high is kept low, with every changed
17
+ definition, and its tier reason says so. Lifting this takes a new ADR, not a flag.
18
+
19
+ A record that is already high confidence is never re-merged: it no longer stores every
20
+ changed definition, so a lower tier could not be rebuilt from it. ``merge`` can therefore run
21
+ again over a published database (``merge-llm``) as new extractions arrive.
14
22
  """
15
23
 
16
24
  from collections import Counter
17
25
  from collections.abc import Iterable, Mapping, Sequence
18
26
  from pathlib import Path
19
27
 
28
+ from pydantic import BaseModel, ConfigDict
29
+
20
30
  from patchsight.db import Record, Symbol, SymbolEvidence
21
31
  from patchsight.eval.confidence import VALID, Answer, Decision, decide, is_public
22
32
  from patchsight.pipeline.pypi_extract import PyPIExtraction
23
33
 
34
+ # ADR 0008: in the declared test evaluation (ADR 0004), 7.0% of high-confidence reports missed
35
+ # the vulnerable symbol, against a target of 2% or less. Until a revised rule meets the target
36
+ # on data not used to choose it, published databases hold no high-confidence records.
37
+ PUBLISH_HIGH = False
38
+ WITHHELD = "high confidence withheld (ADR 0008)"
39
+
24
40
 
25
41
  def load(path: Path) -> dict[str, PyPIExtraction]:
26
42
  """A pypi-extract JSONL file, keyed by vulnerability."""
@@ -60,8 +76,11 @@ def merge(
60
76
  fallback: Sequence[Record],
61
77
  primary: PyPIExtraction | None,
62
78
  secondary: PyPIExtraction | None,
79
+ publish_high: bool = PUBLISH_HIGH,
63
80
  ) -> tuple[list[Record], str]:
64
81
  """The records for one vulnerability, and what happened (for the build report)."""
82
+ if any(r.confidence == "high" for r in fallback):
83
+ return list(fallback), "kept: already high confidence"
65
84
  if primary is None:
66
85
  return list(fallback), "fallback: no extraction"
67
86
  if any(r.input_hash != primary.input_hash for r in fallback):
@@ -73,6 +92,8 @@ def merge(
73
92
  decision: Decision = decide(
74
93
  answer(primary), answer(secondary) if secondary else None, language="python"
75
94
  )
95
+ if decision.tier == "high" and not publish_high:
96
+ decision = Decision(tier="low", reason=f"{WITHHELD}: {decision.reason}", agreed=frozenset())
76
97
  valid_second = secondary is not None and secondary.outcome.status in VALID
77
98
  entry_points = set(primary.entry_points)
78
99
  if valid_second and secondary is not None:
@@ -115,6 +136,7 @@ def apply(
115
136
  records: Iterable[Record],
116
137
  primaries: Mapping[str, PyPIExtraction],
117
138
  secondaries: Mapping[str, PyPIExtraction],
139
+ publish_high: bool = PUBLISH_HIGH,
118
140
  ) -> tuple[list[Record], dict[str, int]]:
119
141
  """Merge extraction into fallback records. Returns the records and outcome counts."""
120
142
  by_vulnerability: dict[str, list[Record]] = {}
@@ -127,7 +149,61 @@ def apply(
127
149
  by_vulnerability[vulnerability],
128
150
  primaries.get(vulnerability),
129
151
  secondaries.get(vulnerability),
152
+ publish_high,
130
153
  )
131
154
  merged.extend(result)
132
155
  outcomes[outcome] += 1
133
156
  return merged, dict(sorted(outcomes.items()))
157
+
158
+
159
+ class Picks(BaseModel):
160
+ """How much of a database carries model picks (records whose source is the LLM)."""
161
+
162
+ model_config = ConfigDict(frozen=True)
163
+
164
+ vulnerabilities: int
165
+ records: int
166
+
167
+
168
+ def picks(records: Iterable[Record]) -> Picks:
169
+ chosen = [r for r in records if r.source == "llm"]
170
+ return Picks(vulnerabilities=len({r.vulnerability for r in chosen}), records=len(chosen))
171
+
172
+
173
+ class MergeReport(BaseModel):
174
+ """What one ``merge-llm`` run changed, for the release notes."""
175
+
176
+ model_config = ConfigDict(frozen=True)
177
+
178
+ previous_version: str
179
+ db_version: str
180
+ models: tuple[str, ...] # extractor models whose answers were offered
181
+ second_opinion_models: tuple[str, ...]
182
+ prompt_versions: tuple[str, ...]
183
+ before: Picks
184
+ after: Picks
185
+ tiers: dict[str, int] # records by confidence, after the merge
186
+ outcomes: dict[str, int] # per vulnerability, as returned by apply
187
+
188
+
189
+ def merge_report(
190
+ before: Sequence[Record],
191
+ after: Sequence[Record],
192
+ primaries: Mapping[str, PyPIExtraction],
193
+ secondaries: Mapping[str, PyPIExtraction],
194
+ outcomes: Mapping[str, int],
195
+ previous_version: str,
196
+ db_version: str,
197
+ ) -> MergeReport:
198
+ offered = [*primaries.values(), *secondaries.values()]
199
+ return MergeReport(
200
+ previous_version=previous_version,
201
+ db_version=db_version,
202
+ models=tuple(sorted({e.outcome.model for e in primaries.values()})),
203
+ second_opinion_models=tuple(sorted({e.outcome.model for e in secondaries.values()})),
204
+ prompt_versions=tuple(sorted({e.outcome.prompt_version for e in offered})),
205
+ before=picks(before),
206
+ after=picks(after),
207
+ tiers=dict(sorted(Counter(r.confidence for r in after).items())),
208
+ outcomes=dict(sorted(outcomes.items())),
209
+ )
@@ -237,7 +237,7 @@ def extract_vulnerability(
237
237
  outcome = extract(
238
238
  llm, config.models[role], template, advisory_text(group, by_id), inputs, known,
239
239
  config.prompt_budget(role), counter, config.budget.max_hunk_lines,
240
- config.candidate_limit(role),
240
+ config.candidate_limit(role), config.validation.dotted_entry_points,
241
241
  ) # fmt: skip
242
242
  except ResponseNotCachedError:
243
243
  return "not_cached" # cache-only replay of a run made elsewhere: never a model call
@@ -0,0 +1,177 @@
1
+ """SARIF 2.1.0 output, for GitHub code scanning and other SARIF readers. No LLM.
2
+
3
+ Every alert becomes one result, so nothing is hidden. The level ranks it:
4
+
5
+ * CALLED: ``error``, located at the first reference in your code;
6
+ * UNKNOWN: ``warning``, located at the dependency's line in the lockfile;
7
+ * IMPORTED_NOT_CALLED and NOT_IMPORTED: ``note``, also at the lockfile line.
8
+
9
+ Each vulnerability is one rule, identified by the alert's advisory ID. A partial fingerprint
10
+ over the advisory and the package keeps an alert's identity when line numbers or the pinned
11
+ version change, so code scanning tracks it across scans.
12
+ """
13
+
14
+ import hashlib
15
+ import json
16
+ import posixpath
17
+ from collections.abc import Mapping, Sequence
18
+ from pathlib import Path, PurePosixPath
19
+ from typing import Any
20
+
21
+ from patchsight import __version__
22
+ from patchsight.scan.lockfiles import Dependency
23
+ from patchsight.scan.report import FOOTNOTE, stale_warning
24
+ from patchsight.scan.verdicts import Alert, Verdict, order
25
+
26
+ SCHEMA = "https://json.schemastore.org/sarif-2.1.0.json"
27
+ INFORMATION_URI = "https://github.com/professor3333/PatchSight"
28
+ LEVELS: dict[Verdict, str] = {
29
+ "CALLED": "error",
30
+ "UNKNOWN": "warning",
31
+ "IMPORTED_NOT_CALLED": "note",
32
+ "NOT_IMPORTED": "note",
33
+ }
34
+ LABELS: dict[Verdict, str] = {
35
+ "CALLED": "Called",
36
+ "UNKNOWN": "Unknown",
37
+ "IMPORTED_NOT_CALLED": "Imported, not called",
38
+ "NOT_IMPORTED": "Not imported",
39
+ }
40
+ FINGERPRINT = "patchsight/v1"
41
+ MAX_RELATED = 20
42
+
43
+
44
+ def uri(path: str, base: str = "") -> str:
45
+ """A result path as a SARIF URI: relative paths stay relative, joined under ``base``."""
46
+ joined = posixpath.normpath(posixpath.join(base, path)) if base else posixpath.normpath(path)
47
+ pure = PurePosixPath(joined)
48
+ return Path(joined).as_uri() if pure.is_absolute() else joined
49
+
50
+
51
+ def _location(path: str, line: int, base: str = "", message: str | None = None) -> dict[str, Any]:
52
+ location: dict[str, Any] = {
53
+ "physicalLocation": {
54
+ "artifactLocation": {"uri": uri(path, base)},
55
+ "region": {"startLine": max(1, line)},
56
+ }
57
+ }
58
+ if message:
59
+ location["message"] = {"text": message}
60
+ return location
61
+
62
+
63
+ def _message(alert: Alert) -> str:
64
+ also = f" (also {', '.join(alert.aliases)})" if alert.aliases else ""
65
+ text = (
66
+ f"{LABELS[alert.verdict]}: {alert.package} {alert.version} is affected by "
67
+ f"{alert.advisory}{also}. {alert.reason}"
68
+ )
69
+ if alert.matched:
70
+ text += f" Matched: {', '.join(alert.matched)}."
71
+ if alert.confidence:
72
+ text += f" Symbol data: {alert.confidence} confidence."
73
+ return text
74
+
75
+
76
+ def render_sarif(
77
+ alerts: Sequence[Alert],
78
+ dependencies: Sequence[Dependency],
79
+ lockfile: str,
80
+ db_meta: Mapping[str, str],
81
+ code_base: str = "",
82
+ ) -> str:
83
+ """One SARIF run. ``code_base`` is the scanned directory, which code paths are relative to."""
84
+ ordered = order(alerts)
85
+ pins = {(d.name, d.version): d for d in dependencies if d.version}
86
+ rules: list[dict[str, Any]] = []
87
+ index: dict[str, int] = {}
88
+ results: list[dict[str, Any]] = []
89
+ for alert in ordered:
90
+ if alert.advisory not in index:
91
+ index[alert.advisory] = len(rules)
92
+ rules.append(
93
+ {
94
+ "id": alert.advisory,
95
+ "shortDescription": {"text": f"Vulnerable dependency: {alert.advisory}"},
96
+ "helpUri": f"https://osv.dev/vulnerability/{alert.advisory}",
97
+ "help": {
98
+ "text": f"{alert.advisory} affects {alert.package}. Upgrade to a fixed "
99
+ "version, or check whether your code reaches the vulnerable code."
100
+ },
101
+ "properties": {"tags": ["security", "dependency"]},
102
+ }
103
+ )
104
+ dependency = pins.get((alert.package, alert.version))
105
+ manifest = _location(
106
+ dependency.source if dependency else lockfile,
107
+ dependency.line if dependency else 1,
108
+ message=f"{alert.package}=={alert.version}",
109
+ )
110
+ if alert.verdict == "CALLED" and alert.evidence:
111
+ first, *rest = alert.evidence
112
+ primary = _location(first.path, first.line, code_base)
113
+ related = [
114
+ _location(e.path, e.line, code_base, "another reference")
115
+ for e in rest[: MAX_RELATED - 1]
116
+ ] + [manifest]
117
+ else:
118
+ primary, related = manifest, []
119
+ result: dict[str, Any] = {
120
+ "ruleId": alert.advisory,
121
+ "ruleIndex": index[alert.advisory],
122
+ "level": LEVELS[alert.verdict],
123
+ "message": {"text": _message(alert)},
124
+ "locations": [primary],
125
+ "partialFingerprints": {
126
+ FINGERPRINT: hashlib.sha256(
127
+ f"{alert.advisory}|{alert.package}".encode()
128
+ ).hexdigest()[:32]
129
+ },
130
+ "properties": {
131
+ "verdict": alert.verdict,
132
+ "package": alert.package,
133
+ "version": alert.version,
134
+ "aliases": list(alert.aliases),
135
+ "matched": list(alert.matched),
136
+ "confidence": alert.confidence,
137
+ },
138
+ }
139
+ if related:
140
+ for number, location in enumerate(related, 1):
141
+ location["id"] = number
142
+ result["relatedLocations"] = related
143
+ results.append(result)
144
+ notifications = [
145
+ {"level": "warning", "message": {"text": warning}}
146
+ for warning in [stale_warning(dict(db_meta))]
147
+ if warning
148
+ ]
149
+ run: dict[str, Any] = {
150
+ "tool": {
151
+ "driver": {
152
+ "name": "PatchSight",
153
+ "semanticVersion": __version__,
154
+ "informationUri": INFORMATION_URI,
155
+ "rules": rules,
156
+ }
157
+ },
158
+ "invocations": [{"executionSuccessful": True, "toolExecutionNotifications": notifications}],
159
+ "results": results,
160
+ "properties": {
161
+ "database": {
162
+ k: db_meta[k] for k in ("db_version", "builder", "built_at") if k in db_meta
163
+ },
164
+ "unpinned": [
165
+ {
166
+ "name": d.raw_name,
167
+ "reason": d.unpinned_reason,
168
+ "source": d.source,
169
+ "line": d.line,
170
+ }
171
+ for d in dependencies
172
+ if not d.version
173
+ ],
174
+ "note": FOOTNOTE,
175
+ },
176
+ }
177
+ return json.dumps({"$schema": SCHEMA, "version": "2.1.0", "runs": [run]}, indent=2)
File without changes