chore: merge main into codex/audit-python-bytecode

src/artifacts/pyinstaller/MarshalSubset.test.ts

@@ -20,6 +20,11 @@ it("reads a Python 3.9 reference-flagged PYZ table", () => {
   ]);
 });
 
+it("reads marshal one-byte string tags as Latin-1", () => {
+  for (const hex of ["6101000000e9", "4101000000e9", "7a01e9", "5a01e9"])
+    expect(readMarshalSubset(bytes(hex))).toEqual({ ok: true, value: "é" });
+});
+
 it("decodes CPython marshal Unicode with surrogatepass", () => {
   expect(
     readMarshalSubset(

src/artifacts/pyinstaller/MarshalSubset.ts

@@ -168,14 +168,8 @@ const readAsciiBytes = (
 ): Result<string, PyInstallerFailure> => {
   const bytes = readBytes(cursor, length);
   if (!bytes.ok) return bytes;
-  if (bytes.value.some((byte) => byte > 0x7f))
+  // These tags use CPython's one-byte Unicode constructor, not an ASCII check.
-    return err(
+  return ok(bytes.value.toString("latin1"));
-      pyInstallerFailure(
-        "format",
-        "PYZ marshal ASCII string contains a non-ASCII byte",
-      ),
-    );
-  return ok(bytes.value.toString("ascii"));
 };
 
 const readUtf8 = (

src/domain/analysisView/analysisViewResult.ts

@@ -559,7 +559,6 @@ export const erlangBeamSummaryViewResultSchema = z.strictObject({
 });
 export const erlangBeamTableViewResultSchema = z.strictObject({
   ...viewResultBase,
-  coverage: nullableCollectionViewCoverageSchema,
   kind: z.literal("beam"),
   parent_operation: z.literal("inspect_erlang_beam"),
   view: erlangBeamViewRequestSchema,
@@ -625,7 +624,6 @@ export const rustBinarySummaryViewResultSchema = z.strictObject({
 
 export const rustBinaryPageViewResultSchema = z.strictObject({
   ...viewResultBase,
-  coverage: nullableCollectionViewCoverageSchema,
   ...rustBinaryViewContextSchema.shape,
   kind: z.literal("rust"),
   parent_operation: z.literal("inspect_rust_binary"),
@@ -1507,7 +1505,6 @@ export const analysisViewResultSchema = z.discriminatedUnion("kind", [
   z.strictObject({
     ...viewResultBase,
     ...goBinaryViewContextSchema.shape,
-    coverage: nullableCollectionViewCoverageSchema,
     kind: z.literal("go"),
     parent_operation: z.literal("inspect_go_binary"),
     view: goBinaryViewRequestSchema,

src/domain/analysisView/artifactAvailabilityView.test.ts

@@ -1,125 +0,0 @@
-import { expect, it } from "vitest";
-import { createEvidence } from "../evidence.js";
-import type { ErlangBeam } from "../erlang/beam.js";
-import type { ModelArtifact } from "../model/modelArtifact.js";
-import {
-  projectAnalysisView,
-  type AnalysisViewParent,
-} from "./analysisView.js";
-
-const parent = (
-  operation: string,
-  result: ErlangBeam | ModelArtifact,
-): AnalysisViewParent => {
-  const evidence = createEvidence(
-    { path: result.artifact.path, sha256: "a".repeat(64), format: "file" },
-    { id: "fixture", name: "Fixture reader", version: "1" },
-    { operation, parameters: {}, result },
-  );
-  return {
-    predicateType: evidence.predicate_type,
-    evidenceId: evidence.evidence_id,
-    operation,
-    subject: evidence.subject,
-    normalizedResult: result,
-    limitations: [],
-  };
-};
-
-const beam: ErlangBeam = {
-  artifact: {
-    path: "/artifacts/module.beam",
-    sha256: "a".repeat(64),
-    bytes: 12,
-  },
-  form: { location: { offset: 0, bytes: 12 }, size: 4 },
-  chunks: [],
-  atoms: null,
-  atom_encoding: null,
-  exports: null,
-  imports: null,
-  lambdas: null,
-  code: null,
-  module: null,
-  coverage: { status: "complete", examined_chunks: 0, atoms: 0 },
-  limitations: [],
-};
-
-it.each(["atoms", "exports", "imports", "lambdas"] as const)(
-  "distinguishes unavailable BEAM %s from a decoded empty table",
-  (collection) => {
-    for (const available of [false, true]) {
-      const result = projectAnalysisView(
-        parent("inspect_erlang_beam", {
-          ...beam,
-          [collection]: available ? [] : null,
-        }),
-        { kind: "beam", collection, offset: 7, limit: 1 },
-      );
-      if (!result.ok) throw result.error;
-      expect(result.value.coverage).toEqual({
-        status: available ? "empty" : "unavailable",
-        examined: 0,
-        total: available ? 0 : null,
-        next_offset: null,
-        exhausted: available,
-      });
-      expect(result.value).toMatchObject({ page: { collection, items: [] } });
-      if (!available)
-        expect(result.value.unknowns.join(" ")).toContain(
-          `No ${collection} table was decoded`,
-        );
-    }
-  },
-);
-
-const model: ModelArtifact = {
-  artifact: {
-    path: "/artifacts/model.gguf",
-    file_size: 32,
-    header_sha256: "a".repeat(64),
-    header_length: 32,
-    sha256: "a".repeat(64),
-    bytes_read: 32,
-    read_scope: "full-file",
-  },
-  format: "gguf",
-  version: 3,
-  byte_order: "little",
-  alignment: 32,
-  tensor_data_offset: 32,
-  metadata: [],
-  tensors: [],
-  totals: {
-    confidence: "derived",
-    parameter_count: "0",
-    tensor_bytes: 0,
-    by_type: [],
-  },
-  coverage: {
-    header: "complete",
-    tensor_data: "not-interpreted",
-    array_values: "preview",
-    tensor_ranges: "partial",
-    storage_sizes: "partial",
-  },
-  limitations: [
-    "Array metadata retains only the requested preview; storage sizes remain unknown.",
-  ],
-};
-
-it.each(["metadata", "tensors", "types"] as const)(
-  "retains producer limitations on model %s pages with no envelope warnings",
-  (collection) => {
-    const input = parent("inspect_model_artifact", model);
-    const result = projectAnalysisView(input, {
-      kind: "model-artifact",
-      collection,
-      offset: 0,
-      limit: 1,
-    });
-    if (!result.ok) throw result.error;
-    expect(result.value.limitations).toEqual(model.limitations);
-    expect(input.limitations).toEqual([]);
-  },
-);

src/domain/analysisView/erlangBeamView.ts

@@ -13,7 +13,6 @@ import {
   boundedViewMessages,
   completeWithinViewCoverage,
   pageViewCoverage,
-  unavailableViewCoverage,
 } from "./analysisViewCoverage.js";
 import { selectViewRows } from "./selectViewRows.js";
 
@@ -121,7 +120,7 @@ type Row = NonNullable<ErlangBeam[Table]>[number];
 type BeamView = z.infer<typeof erlangBeamViewRequestSchema>;
 
 /** Select already parsed rows; original atom/lambda indexes are never repurposed. */
-const selectRows = (source: readonly Row[], view: BeamView) => {
+const selectRows = (rows: readonly Row[] | null, view: BeamView) => {
   const names = view.names === undefined ? undefined : new Set(view.names);
   const indices =
     view.indices === undefined ? undefined : new Set(view.indices);
@@ -139,6 +138,7 @@ const selectRows = (source: readonly Row[], view: BeamView) => {
     ),
     source_index,
   });
+  const source = rows ?? [];
   if (names === undefined && indices === undefined && modules === undefined)
     return {
       rows: source
@@ -223,17 +223,7 @@ export const projectErlangBeamView = (
       summary,
       coverage: completeWithinViewCoverage(chunks.length, chunks.length),
     });
-  const rows = parsed.value[view.collection];
+  const selected = selectRows(parsed.value[view.collection], view);
-  if (rows === null)
-    return ok({
-      ...shared,
-      kind: "beam" as const,
-      view,
-      beam: summary,
-      page: { collection: view.collection, items: [] },
-      coverage: unavailableViewCoverage(),
-    });
-  const selected = selectRows(rows, view);
   return ok({
     ...shared,
     kind: "beam" as const,

src/domain/analysisView/goBinaryView.test.ts

@@ -238,53 +238,3 @@ it("refuses string_literals view when unrequested and projects filtered/truncate
     ],
   });
 });
-
-it.each(["dependencies", "settings"] as const)(
-  "distinguishes unavailable Go %s from decoded empty records",
-  (collection) => {
-    for (const available of [false, true]) {
-      const input = parent();
-      const result = projectAnalysisView(
-        {
-          ...input,
-          normalizedResult: {
-            ...report,
-            build_info: available
-              ? {
-                  header_offset: 0,
-                  encoding: "inline",
-                  go_version: null,
-                  go_version_bytes_base64: "",
-                  module_text: null,
-                  module_bytes_base64: "",
-                  version_location: { offset: 32, bytes: 0 },
-                  module_location: { offset: 32, bytes: 0 },
-                  module: {
-                    path: null,
-                    main: null,
-                    dependencies: [],
-                    settings: [],
-                    unparsed_lines: [],
-                    unparsed_line_bytes_base64: [],
-                    complete: true,
-                  },
-                }
-              : null,
-          },
-        },
-        { kind: "go", collection, offset: 3, limit: 1 },
-      );
-      if (!result.ok) throw result.error;
-      expect(result.value.coverage).toEqual({
-        status: available ? "empty" : "unavailable",
-        examined: 0,
-        total: available ? 0 : null,
-        next_offset: null,
-        exhausted: available,
-      });
-      expect(result.value).toMatchObject({
-        source_coverage: { build_info_available: available },
-      });
-    }
-  },
-);

src/domain/analysisView/goBinaryView.ts

@@ -20,7 +20,6 @@ import {
   boundedViewMessages,
   completeWithinViewCoverage,
   pageViewCoverage,
-  unavailableViewCoverage,
 } from "./analysisViewCoverage.js";
 import { analysisViewInputError } from "./analysisViewErrors.js";
 import { createAnalysisViewResultReader } from "./analysisViewParent.js";
@@ -323,18 +322,10 @@ export const projectGoBinaryView = (
       ),
     });
   }
-  if (info === null)
-    return ok({
-      kind: "go" as const,
-      view,
-      ...shared,
-      page: { collection: view.collection, rows: [] },
-      coverage: unavailableViewCoverage(),
-    });
   if (view.collection === "dependencies") {
     const paths = view.paths === undefined ? undefined : new Set(view.paths);
     const selected = selectViewRows(
-      info.module.dependencies,
+      info?.module.dependencies ?? [],
       view,
       (row, index) =>
         (indices?.has(index) ?? true) && (paths?.has(row.path) ?? true),
@@ -362,7 +353,7 @@ export const projectGoBinaryView = (
   }
   const keys = view.keys === undefined ? undefined : new Set(view.keys);
   const selected = selectViewRows(
-    info.module.settings,
+    info?.module.settings ?? [],
     view,
     (row, index) =>
       (indices?.has(index) ?? true) && (keys?.has(row.key) ?? true),

src/domain/analysisView/modelArtifactView.ts

@@ -217,7 +217,7 @@ export const projectModelArtifactView = (
     parent_digest: parent.evidenceId.slice(3),
     artifact: { path: report.artifact.path, sha256: report.artifact.sha256 },
     limitations: boundedViewMessages(
-      [...parent.limitations, ...report.limitations],
+      parent.limitations,
       parent.evidenceId,
       "limitations",
     ),

src/domain/analysisView/rustBinaryView.test.ts

@@ -206,40 +206,3 @@ it("pages matching original indices and omits dependency vectors unless selected
     next_offset: null,
   });
 });
-
-it.each(["packages", "edges"] as const)(
-  "distinguishes unavailable Rust %s from decoded empty records",
-  (collection) => {
-    for (const available of [false, true]) {
-      const metadata = report.auditable_metadata!;
-      const result = projectAnalysisView(
-        parent({
-          ...report,
-          auditable_metadata: available
-            ? {
-                ...metadata,
-                packages: [],
-                root_package_index: null,
-                dependency_edges: [],
-                cyclic_dependency_indices: [],
-              }
-            : null,
-        }),
-        { kind: "rust", collection, offset: 3, limit: 1 },
-      );
-      if (!result.ok) throw result.error;
-      expect(result.value.coverage).toEqual({
-        status: available ? "empty" : "unavailable",
-        examined: 0,
-        total: available ? 0 : null,
-        next_offset: null,
-        exhausted: available,
-      });
-      expect(result.value).toMatchObject({
-        source_coverage: {
-          embedded_records: available ? "complete" : "unknown",
-        },
-      });
-    }
-  },
-);

src/domain/analysisView/rustBinaryView.ts

@@ -22,7 +22,6 @@ import {
   boundedViewMessages,
   completeWithinViewCoverage,
   pageViewCoverage,
-  unavailableViewCoverage,
 } from "./analysisViewCoverage.js";
 import { analysisViewInputError } from "./analysisViewErrors.js";
 import { selectViewRows } from "./selectViewRows.js";
@@ -191,15 +190,6 @@ export const projectRustBinaryView = (
       summary,
       coverage: completeWithinViewCoverage(1, 1),
     });
-  if (metadata === null)
-    return ok({
-      kind: "rust" as const,
-      view,
-      ...shared,
-      rust: summary,
-      page: { collection: view.collection, rows: [] },
-      coverage: unavailableViewCoverage(),
-    });
   const indices =
     view.indices === undefined ? undefined : new Set(view.indices);
   const fields =
@@ -209,7 +199,7 @@ export const projectRustBinaryView = (
     const versions =
       view.versions === undefined ? undefined : new Set(view.versions);
     const selected = selectViewRows(
-      metadata.packages,
+      metadata?.packages ?? [],
       view,
       (row, index) =>
         (indices?.has(index) ?? true) &&
@@ -248,7 +238,7 @@ export const projectRustBinaryView = (
   const to =
     view.to_indices === undefined ? undefined : new Set(view.to_indices);
   const selected = selectViewRows(
-    metadata.dependency_edges,
+    metadata?.dependency_edges ?? [],
     view,
     (row, index) =>
       (indices?.has(index) ?? true) &&

src/python/PythonPycParser.test.ts

@@ -407,6 +407,30 @@ it("charges long digits to the string budget and keeps large previews bounded",
   );
 });
 
+it("reads marshal one-byte string tags as Latin-1 with Unicode identity", () => {
+  for (const tag of [0x61, 0x41, 0x7a, 0x5a]) {
+    const encoded = Buffer.concat([
+      Buffer.from([tag]),
+      tag === 0x61 || tag === 0x41 ? u32(1) : Buffer.from([1]),
+      Buffer.from([0xe9]),
+    ]);
+    const sameUnicode = marshalUnicode(Buffer.from("é", "utf8"));
+    const set = Buffer.concat([
+      Buffer.from([0x3c]),
+      u32(2),
+      encoded,
+      sameUnicode,
+    ]);
+    const report = root(parse(withConsts([encoded, set])));
+    expect(report.consts[0]).toMatchObject({
+      kind: "str",
+      value: "é",
+      bytes: 1,
+    });
+    expect(report.consts[1]).toEqual({ kind: "set", length: 1 });
+  }
+});
+
 it("decodes marshal strings with CPython surrogatepass rules", () => {
   const report = parse(
     withConsts([marshalUnicode(Buffer.from("efbbbf73eda080efbbbf", "hex"))]),
@@ -1534,6 +1558,36 @@ it("hashes a long TYPE_REF tuple chain without recursing", () => {
   expect(pythonPycSchema.safeParse(report).success).toBe(true);
 });
 
+it("hashes shallow TYPE_REF code chains without using the JavaScript stack", () => {
+  const links = 3000;
+  const codes = Array.from({ length: links }, (_, index) =>
+    marshalCode("py311", {
+      flagged: true,
+      name: "f",
+      consts: index === 0 ? [Buffer.from([0x4e])] : [marshalRef(index - 1)],
+    }),
+  );
+  const bytes = Buffer.concat([
+    pythonPycHeader(3531),
+    marshalCode("py311", {
+      constsObject: Buffer.concat([
+        Buffer.from([0x28]),
+        u32(links + 1),
+        ...codes,
+        Buffer.from([0x3c]),
+        u32(1),
+        marshalRef(links - 1),
+      ]),
+    }),
+  ]);
+  const report = parseWith(bytes, { max_objects: 65_536 });
+  expect(root(report).consts.at(-1)).toEqual({ kind: "set", length: 1 });
+  expect(report.coverage).toMatchObject({
+    status: "complete",
+    code_objects: links + 1,
+  });
+});
+
 it("rejects code CPython treats as corrupt instead of complete coverage", () => {
   const file = (
     layout: "py37" | "py38" | "py311",

src/python/PythonPycParser.ts

@@ -332,7 +332,7 @@ class MarshalReader {
   readonly #codeKeys = new WeakMap<CodeNode, string>();
   readonly #codeIdentities = new WeakMap<CodeNode, number>();
   readonly #codeEquality = new WeakMap<CodeNode, CodeEquality>();
-  /** Python hashability per code object. False while that check is in progress. */
+  /** Completed Python hashability checks per code object. */
   readonly #codeHashable = new WeakMap<CodeNode, boolean>();
   /** Tuples whose items are still being read; CPython cannot hash them yet. */
   readonly #readingTuples = new WeakSet<MarshalValue>();
@@ -436,13 +436,19 @@ class MarshalReader {
         };
       case TYPE_INTERNED:
       case TYPE_UNICODE:
-        return this.utf8(this.nonNegativeInt32("Marshal string length"), false);
+        return this.string(
+          this.nonNegativeInt32("Marshal string length"),
+          "utf8",
+        );
       case TYPE_ASCII:
       case TYPE_ASCII_INTERNED:
-        return this.utf8(this.nonNegativeInt32("Marshal string length"), true);
+        return this.string(
+          this.nonNegativeInt32("Marshal string length"),
+          "latin1",
+        );
       case TYPE_SHORT_ASCII:
       case TYPE_SHORT_ASCII_INTERNED:
-        return this.utf8(this.u8(), true);
+        return this.string(this.u8(), "latin1");
       case TYPE_TUPLE:
         return this.sequenceValue(
           "tuple",
@@ -1309,13 +1315,70 @@ class MarshalReader {
   #isCodeHashable(node: CodeNode): boolean {
     const known = this.#codeHashable.get(node);
     if (known !== undefined) return known;
-    this.#codeHashable.set(node, false);
+    const frames: { value: MarshalValue; finish: boolean }[] = [
-    const material =
+      { value: { kind: "code", value: node }, finish: false },
-      this.#codeEquality.get(node) ??
+    ];
-      malformed("Code object is missing equality material.");
+    const gray = new Set<MarshalValue | CodeNode>();
-    const hashable = this.#equalityKey(material.constsTuple, "hash") !== null;
+    const complete = new Set<MarshalValue>();
-    this.#codeHashable.set(node, hashable);
+    const reject = (): false => {
-    return hashable;
+      this.#codeHashable.set(node, false);
+      return false;
+    };
+    while (frames.length > 0) {
+      this.#takeKeyStep();
+      const frame = frames.pop()!;
+      const value = frame.value;
+      const identity = value.kind === "code" ? value.value : value;
+      if (frame.finish) {
+        gray.delete(identity);
+        if (value.kind === "code") {
+          const material =
+            this.#codeEquality.get(value.value) ??
+            malformed("Code object is missing equality material.");
+          // Child code checks are now complete, so the existing tuple key
+          // walker preserves its equality and expansion budgets without recursion.
+          const hashable =
+            this.#equalityKey(material.constsTuple, "hash") !== null;
+          this.#codeHashable.set(value.value, hashable);
+          if (!hashable) return reject();
+        } else complete.add(value);
+        continue;
+      }
+      if (value.kind === "code") {
+        const cached = this.#codeHashable.get(value.value);
+        if (cached === false) return reject();
+        if (cached === true) continue;
+      } else {
+        const cached = this.#keyCache.get(value);
+        if (cached === null) return reject();
+        if (cached !== undefined || complete.has(value)) continue;
+      }
+      if (
+        gray.has(identity) ||
+        this.#readingTuples.has(value) ||
+        value.kind === "list" ||
+        value.kind === "set" ||
+        value.kind === "dict" ||
+        value.kind === "null"
+      )
+        return reject();
+      const children =
+        value.kind === "code"
+          ? [
+              (
+                this.#codeEquality.get(value.value) ??
+                malformed("Code object is missing equality material.")
+              ).constsTuple,
+            ]
+          : value.kind === "tuple" || value.kind === "frozenset"
+            ? value.items
+            : [];
+      gray.add(identity);
+      frames.push({ value, finish: true });
+      for (let index = children.length - 1; index >= 0; index -= 1)
+        frames.push({ value: children[index]!, finish: false });
+    }
+    return true;
   }
 
   #codeIdentity(node: CodeNode): number {
@@ -1492,17 +1555,20 @@ class MarshalReader {
     );
   }
 
-  utf8(length: number, ascii: boolean): MarshalValue {
+  string(length: number, encoding: "latin1" | "utf8"): MarshalValue {
     const bytes = this.blob(length);
-    if (ascii) {
+    if (encoding === "latin1") {
-      if (bytes.some((byte) => byte > 0x7f))
+      // CPython reads the ASCII-tagged forms as one-byte Unicode, even when
-        malformed("Marshal ASCII string contains a non-ASCII byte.");
+      // high bytes appear. Identity must agree with the UTF-8 string form.
+      const value = bytes.toString("latin1");
       return {
         kind: "str",
-        value: bytes.toString("latin1"),
+        value,
         previewEnd: Math.min(length, PYTHON_PYC_CONST_PREVIEW_LIMIT),
         bytes: length,
-        raw: bytes,
+        raw: bytes.some((byte) => byte > 0x7f)
+          ? Buffer.from(value, "utf8")
+          : bytes,
       };
     }
     const decoded = decodeMarshalUtf8(bytes, PYTHON_PYC_CONST_PREVIEW_LIMIT);

tests/boundary/mcp/goBinaryViews.test.ts

@@ -368,7 +368,7 @@ mcpTest(
       const page = goPage(parseEvidence(empty.structuredContent));
       expect(page).toMatchObject({
         page: { rows: [] },
-        coverage: { status: "unavailable", total: null, exhausted: false },
+        coverage: { total: 0 },
         source_coverage: {
           build_info_available: false,
           module_parse_complete: null,