fix(v8): validate complete payload scan coverage

src/application/v8CodeCache/V8CodeCacheService.test.ts

@@ -0,0 +1,68 @@
+import { createHash } from "node:crypto";
+import { expect, it } from "vitest";
+
+import { AnalysisOutputError } from "../../domain/analysisErrorCore.js";
+import { parseEvidence } from "../../domain/evidence.js";
+import { ok } from "../../domain/result.js";
+import {
+  V8_CODE_CACHE_HEADER_BYTES,
+  V8_CODE_CACHE_MAGIC_HIGH,
+  type V8CodeCache,
+} from "../../domain/v8/v8CodeCache.js";
+import { parseV8CodeCache } from "../../v8/V8CodeCacheParser.js";
+import { V8CodeCacheService } from "./V8CodeCacheService.js";
+
+const reportFor = (payloadBytes: number): V8CodeCache => {
+  const bytes = Buffer.alloc(V8_CODE_CACHE_HEADER_BYTES + payloadBytes, 0x61);
+  bytes.fill(0, 0, V8_CODE_CACHE_HEADER_BYTES);
+  bytes.writeUInt32LE(V8_CODE_CACHE_MAGIC_HIGH * 0x10000, 0);
+  bytes.writeUInt32LE(payloadBytes, 20);
+  return parseV8CodeCache(
+    bytes,
+    {
+      path: "/artifacts/authored.jsc",
+      sha256: createHash("sha256").update(bytes).digest("hex"),
+      bytes: bytes.length,
+    },
+    { min_string_bytes: 4, max_strings: 8 },
+  );
+};
+
+const inspect = (report: V8CodeCache) =>
+  new V8CodeCacheService({
+    identity: { id: "cache-fixture", name: "Cache fixture", version: "1" },
+    close: async () => {},
+    inspect: async () => ok(report),
+  }).inspect({ path: report.artifact.path });
+
+it.each([0, 9])(
+  "rejects complete coverage claiming %s scanned bytes for an eight-byte payload",
+  async (scanned_bytes) => {
+    const report = reportFor(8);
+    const result = await inspect({
+      ...report,
+      coverage: { ...report.coverage, scanned_bytes },
+    });
+    expect(result.ok).toBe(false);
+    if (!result.ok) expect(result.error).toBeInstanceOf(AnalysisOutputError);
+  },
+);
+
+it.each([8, 5000])(
+  "retains complete scan facts for a %s-byte payload and its bounded string prefix",
+  async (payloadBytes) => {
+    const report = reportFor(payloadBytes);
+    const result = await inspect(report);
+    if (!result.ok) throw result.error;
+    expect(parseEvidence(result.value).normalized_result).toMatchObject({
+      coverage: { status: "complete", scanned_bytes: payloadBytes, strings: 1 },
+      embedded_strings: [
+        {
+          bytes: payloadBytes,
+          retained_bytes: Math.min(payloadBytes, 4096),
+          truncated: payloadBytes > 4096,
+        },
+      ],
+    });
+  },
+);

src/artifacts/pyinstaller/MarshalSubset.test.ts

@@ -20,11 +20,6 @@ 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

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

src/domain/v8/v8CodeCache.ts

@@ -196,6 +196,12 @@ export const v8CodeCacheSchema = z
         message:
           "Observation coverage disagrees with the listed embedded strings.",
       });
+    if (report.coverage.scanned_bytes !== report.payload.bytes)
+      context.addIssue({
+        code: "custom",
+        path: ["coverage", "scanned_bytes"],
+        message: "Complete coverage must examine every declared payload byte.",
+      });
     report.embedded_strings.forEach((observation, index) => {
       if (observation.retained_bytes > observation.bytes)
         context.addIssue({

src/python/PythonPycParser.test.ts

@@ -407,30 +407,6 @@ 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"))]),
@@ -1558,36 +1534,6 @@ 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>();
-  /** Completed Python hashability checks per code object. */
+  /** Python hashability per code object. False while that check is in progress. */
   readonly #codeHashable = new WeakMap<CodeNode, boolean>();
   /** Tuples whose items are still being read; CPython cannot hash them yet. */
   readonly #readingTuples = new WeakSet<MarshalValue>();
@@ -436,19 +436,13 @@ class MarshalReader {
         };
       case TYPE_INTERNED:
       case TYPE_UNICODE:
-        return this.string(
+        return this.utf8(this.nonNegativeInt32("Marshal string length"), false);
-          this.nonNegativeInt32("Marshal string length"),
-          "utf8",
-        );
       case TYPE_ASCII:
       case TYPE_ASCII_INTERNED:
-        return this.string(
+        return this.utf8(this.nonNegativeInt32("Marshal string length"), true);
-          this.nonNegativeInt32("Marshal string length"),
-          "latin1",
-        );
       case TYPE_SHORT_ASCII:
       case TYPE_SHORT_ASCII_INTERNED:
-        return this.string(this.u8(), "latin1");
+        return this.utf8(this.u8(), true);
       case TYPE_TUPLE:
         return this.sequenceValue(
           "tuple",
@@ -1315,70 +1309,13 @@ class MarshalReader {
   #isCodeHashable(node: CodeNode): boolean {
     const known = this.#codeHashable.get(node);
     if (known !== undefined) return known;
-    const frames: { value: MarshalValue; finish: boolean }[] = [
+    this.#codeHashable.set(node, false);
-      { value: { kind: "code", value: node }, finish: false },
+    const material =
-    ];
+      this.#codeEquality.get(node) ??
-    const gray = new Set<MarshalValue | CodeNode>();
+      malformed("Code object is missing equality material.");
-    const complete = new Set<MarshalValue>();
+    const hashable = this.#equalityKey(material.constsTuple, "hash") !== null;
-    const reject = (): false => {
+    this.#codeHashable.set(node, hashable);
-      this.#codeHashable.set(node, false);
+    return hashable;
-      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 {
@@ -1555,20 +1492,17 @@ class MarshalReader {
     );
   }
 
-  string(length: number, encoding: "latin1" | "utf8"): MarshalValue {
+  utf8(length: number, ascii: boolean): MarshalValue {
     const bytes = this.blob(length);
-    if (encoding === "latin1") {
+    if (ascii) {
-      // CPython reads the ASCII-tagged forms as one-byte Unicode, even when
+      if (bytes.some((byte) => byte > 0x7f))
-      // high bytes appear. Identity must agree with the UTF-8 string form.
+        malformed("Marshal ASCII string contains a non-ASCII byte.");
-      const value = bytes.toString("latin1");
       return {
         kind: "str",
-        value,
+        value: bytes.toString("latin1"),
         previewEnd: Math.min(length, PYTHON_PYC_CONST_PREVIEW_LIMIT),
         bytes: length,
-        raw: bytes.some((byte) => byte > 0x7f)
+        raw: bytes,
-          ? Buffer.from(value, "utf8")
-          : bytes,
       };
     }
     const decoded = decodeMarshalUtf8(bytes, PYTHON_PYC_CONST_PREVIEW_LIMIT);