fix(ntfs): preserve header facts alongside invalid fixup sequences

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.
-  return ok(bytes.value.toString("latin1"));
+  if (bytes.value.some((byte) => byte > 0x7f))
+    return err(
+      pyInstallerFailure(
+        "format",
+        "PYZ marshal ASCII string contains a non-ASCII byte",
+      ),
+    );
+  return ok(bytes.value.toString("ascii"));
 };
 
 const readUtf8 = (

src/domain/windows/ntfsMft.test.ts

@@ -127,10 +127,17 @@ describe("MFT header facts", () => {
     });
   });
 
-  it.each([0, 49, 64, 72])(
-    "reports an invalid first attribute offset %i",
-    (offset) => {
-      const record = recordOf();
+  it.each([
+    [0, 1],
+    [49, 1],
+    [64, 1],
+    [72, 1],
+    [48, 1],
+    [48, 0],
+  ])(
+    "reports invalid first attribute offset %i with USA sequence %i",
+    (offset, sequenceNumber) => {
+      const record = recordOf({ sequenceNumber });
       new DataView(record.buffer).setUint16(20, offset, true);
       const parsed = parseMftImage(record);
       if (!parsed.ok) throw new Error(parsed.error.message);
@@ -139,20 +146,29 @@ describe("MFT header facts", () => {
     },
   );
 
-  it("reports a declared record number that contradicts its stream position", () => {
-    const bytes = streamOf(recordOf(), recordOf());
-    new DataView(bytes.buffer).setUint32(1024 + 44, 0, true);
-    const parsed = parseMftImage(bytes);
-    if (!parsed.ok) throw new Error(parsed.error.message);
-    expect(parsed.value.records[1]).toMatchObject({
-      index: 1,
-      declaredRecordNumber: 0,
-      recordNumberMatchesSlot: false,
-    });
-    expect(parsed.value.limitations.join(" ")).toContain(
-      "disagrees with their stream slot",
-    );
-  });
+  it.each([1, 0, 0xffff])(
+    "reports a contradictory declared record number with USA sequence %i",
+    (sequenceNumber) => {
+      const bytes = streamOf(recordOf(), recordOf({ sequenceNumber }));
+      new DataView(bytes.buffer).setUint32(1024 + 44, 0, true);
+      const parsed = parseMftImage(bytes);
+      if (!parsed.ok) throw new Error(parsed.error.message);
+      expect(parsed.value.records[1]).toMatchObject({
+        index: 1,
+        declaredRecordNumber: 0,
+        recordNumberMatchesSlot: false,
+        fixups: {
+          malformed: sequenceNumber === 0 || sequenceNumber === 0xffff,
+        },
+      });
+      expect(parsed.value.metadata.fixup_unverified_record_count).toBe(
+        sequenceNumber === 1 ? 0 : 1,
+      );
+      expect(parsed.value.limitations.join(" ")).toContain(
+        "disagrees with their stream slot",
+      );
+    },
+  );
 
   it("keeps the inline header window separate from the parsed record count", () => {
     const parsed = parseMftImage(

src/domain/windows/ntfsMft.ts

@@ -373,8 +373,13 @@ const readRecord = (
     true,
   );
 
+  const usaLayoutValid = updateSequenceArrayFits(
+    usaOffset,
+    fixups.entries,
+    recordBytes / FIXUP_SECTOR_BYTES,
+  );
   const declaredRecordNumber =
-    !fixups.malformed && usaOffset >= RECORD_HEADER_WITH_NUMBER_BYTES
+    usaLayoutValid && usaOffset >= RECORD_HEADER_WITH_NUMBER_BYTES
       ? view.getUint32(RECORD_HEADER.recordNumber, true)
       : null;
 
@@ -396,7 +401,7 @@ const readRecord = (
       firstAttributeOffset % 8 === 0 &&
       firstAttributeOffset >= NTFS_MFT_MIN_RECORD_BYTES &&
       firstAttributeOffset + 4 <= usedBytes &&
-      (fixups.malformed ||
+      (!usaLayoutValid ||
         firstAttributeOffset >= usaOffset + fixups.entries * 2),
     baseRecordReference:
       baseRecordReference === 0n

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(
-          this.nonNegativeInt32("Marshal string length"),
-          "utf8",
-        );
+        return this.utf8(this.nonNegativeInt32("Marshal string length"), false);
       case TYPE_ASCII:
       case TYPE_ASCII_INTERNED:
-        return this.string(
-          this.nonNegativeInt32("Marshal string length"),
-          "latin1",
-        );
+        return this.utf8(this.nonNegativeInt32("Marshal string length"), true);
       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 }[] = [
-      { value: { kind: "code", value: node }, finish: false },
-    ];
-    const gray = new Set<MarshalValue | CodeNode>();
-    const complete = new Set<MarshalValue>();
-    const reject = (): false => {
-      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;
+    this.#codeHashable.set(node, false);
+    const material =
+      this.#codeEquality.get(node) ??
+      malformed("Code object is missing equality material.");
+    const hashable = this.#equalityKey(material.constsTuple, "hash") !== null;
+    this.#codeHashable.set(node, hashable);
+    return hashable;
   }
 
   #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") {
-      // CPython reads the ASCII-tagged forms as one-byte Unicode, even when
-      // high bytes appear. Identity must agree with the UTF-8 string form.
-      const value = bytes.toString("latin1");
+    if (ascii) {
+      if (bytes.some((byte) => byte > 0x7f))
+        malformed("Marshal ASCII string contains a non-ASCII byte.");
       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)
-          ? Buffer.from(value, "utf8")
-          : bytes,
+        raw: bytes,
       };
     }
     const decoded = decodeMarshalUtf8(bytes, PYTHON_PYC_CONST_PREVIEW_LIMIT);