fix(captures): validate packet read scope before paging

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/captures/TsharkCaptureProvider.ts

@@ -369,7 +369,10 @@ export class TsharkCaptureProvider implements PacketCapturePort {
         resourcePath = "path" in input ? input.path : command;
         // TShark resolves `-e` aliases before emitting canonical field names;
         // the command arguments, not output names, own field selection.
-        const matchingPackets = parsePacketOutput(output.stdout.text);
+        const matchingPackets = parsePacketOutput(
+          output.stdout.text,
+          input.packet_limit,
+        );
         const start = Math.min(input.match_offset, matchingPackets.length);
         const end =
           input.match_limit === null

src/captures/parsePacketOutput.ts

@@ -6,7 +6,7 @@ import { packetFieldsSchema } from "../domain/captures/packetCapture.js";
 const producerSchema = z.array(
   z.object({ _source: z.object({ layers: packetFieldsSchema }) }),
 );
-export const parsePacketOutput = (text: string) => {
+export const parsePacketOutput = (text: string, packetLimit: number | null) => {
   const decoded: unknown = JSON.parse(text);
   const packets = producerSchema
     .parse(decoded)
@@ -19,6 +19,11 @@ export const parsePacketOutput = (text: string) => {
         "inspect_packet_capture",
         "TShark returned unordered frames.",
       );
+    if (packetLimit !== null && frame > BigInt(packetLimit))
+      throw new AnalysisOutputError(
+        "inspect_packet_capture",
+        "TShark returned a frame beyond the selected input packet limit.",
+      );
     previous = frame;
   }
   return packets;

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);

tests/boundary/captures/captureProvider.test.ts

@@ -247,6 +247,26 @@ it("preserves TShark canonical output for a caller-selected field alias", async
   expect(report.tool.arguments).toContain("bootp.type");
 });
 
+it("refuses producer frames outside the selected input read scope before paging", async () => {
+  const f = await fixture("many");
+  const request = { path: f.path, fields: ["ip.addr"], match_limit: 1 };
+  expect(
+    await f.service.inspect({ ...request, packet_limit: 10 }),
+  ).toMatchObject({
+    ok: false,
+    error: { _tag: "AnalysisOutputError", operation: "inspect_packet_capture" },
+  });
+  const admitted = await f.service.inspect({ ...request, packet_limit: 90 });
+  if (!admitted.ok) throw admitted.error;
+  const report = packetCaptureSchema.parse(admitted.value.normalized_result);
+  expect(report.output_window).toEqual({
+    matching_packets: 90,
+    returned_packets: 1,
+    has_more: true,
+  });
+  expect(report.packets[0]?.["frame.number"]).toEqual(["1"]);
+});
+
 it("returns canonical alias output through the public packet-capture CLI", async () => {
   const f = await fixture("alias-canonical");
   const cli = Cli.create("capture-alias-test", { sync: false });

tests/boundary/captures/packetOutput.test.ts

@@ -17,7 +17,7 @@ it("preserves repeated field values and missing fields while discarding export i
     },
     { _source: { layers: { "frame.number": ["7"] } } },
   ]);
-  expect(parsePacketOutput(output)).toEqual([
+  expect(parsePacketOutput(output, 7)).toEqual([
     { "frame.number": ["2"], "ip.addr": ["192.0.2.1", "192.0.2.2"] },
     { "frame.number": ["7"] },
   ]);
@@ -49,7 +49,7 @@ it.each([
 ])("rejects malformed producer fields: $layers", ({ layers, issue }) => {
   let failure: unknown;
   try {
-    parsePacketOutput(JSON.stringify([{ _source: { layers } }]));
+    parsePacketOutput(JSON.stringify([{ _source: { layers } }]), null);
   } catch (error) {
     failure = error;
   }
@@ -72,6 +72,7 @@ it.each([
           _source: { layers: { "frame.number": frame } },
         })),
       ),
+      null,
     );
   } catch (error) {
     failure = error;
@@ -88,7 +89,18 @@ it("preserves canonical field names emitted for a selected alias", () => {
       },
     },
   ]);
-  expect(parsePacketOutput(output)).toEqual([
+  expect(parsePacketOutput(output, null)).toEqual([
     { "frame.number": ["1"], "dhcp.type": ["1"] },
   ]);
 });
+
+it("validates every frame against the input read limit while allowing filtered gaps", () => {
+  const output = JSON.stringify(
+    [2, 7].map((frame) => ({
+      _source: { layers: { "frame.number": [String(frame)] } },
+    })),
+  );
+  expect(parsePacketOutput(output, 7)).toHaveLength(2);
+  expect(() => parsePacketOutput(output, 6)).toThrow(AnalysisOutputError);
+  expect(parsePacketOutput(output, null)).toHaveLength(2);
+});