From df57e42457f1cbda6896e7b2c57e5e22480be142 Mon Sep 17 00:00:00 2001
From: HipHoff <matt.d.hoffman@gmail.com>
Date: Wed, 28 Oct 2026 12:15:32 -0700
Subject: [PATCH] gh-160842: Read every frame of a multi-frame Zstandard
 zipfile member

ZipExtFile._read1() treated the end of the first zstd frame as the end
of the member.  When the decompressor reaches the end of a frame and
compressed input remains, continue with a new ZstdDecompressor.  Bytes
that are not a valid frame raise ZstdError, and a truncated last frame
still fails the CRC check.

Co-authored-by: Claude <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_016uX6RoHqnajMDyVi3LrJjs
---
 Lib/test/test_zipfile/test_core.py            | 64 +++++++++++++++++++
 Lib/zipfile/__init__.py                       | 22 ++++++-
 ...-10-28-17-05-12.gh-issue-160842.Zs7mFr.rst |  3 +
 3 files changed, 86 insertions(+), 3 deletions(-)
 create mode 100644 Misc/NEWS.d/next/Library/2026-10-28-17-05-12.gh-issue-160842.Zs7mFr.rst

diff --git a/Lib/test/test_zipfile/test_core.py b/Lib/test/test_zipfile/test_core.py
index ab1995e91286e68..5c9a6fc980d3c95 100644
--- a/Lib/test/test_zipfile/test_core.py
+++ b/Lib/test/test_zipfile/test_core.py
@@ -4935,6 +4935,70 @@ class ZstdBoundedDecompressTests(AbstractBoundedDecompressTests,
     compression = zipfile.ZIP_ZSTANDARD
 
 
+@requires_zstd()
+class ZstdMultipleFramesTests(unittest.TestCase):
+    # A zstd stream may consist of several concatenated frames (RFC 8878,
+    # section 3.1), as written by multi-threaded or streaming compressors.
+    part1 = b'first frame ' * 1000
+    part2 = b'second frame ' * 1000
+
+    def make_zip(self, member_data, content):
+        # Store member_data, then relabel the member as Zstandard-compressed
+        # with the CRC and size of the decompressed content.
+        buf = io.BytesIO()
+        with zipfile.ZipFile(buf, 'w') as zf:
+            zf.writestr(zipfile.ZipInfo('a'), member_data)
+        data = bytearray(buf.getvalue())
+        crc = zipfile.crc32(content)
+        cd = data.rindex(zipfile.stringCentralDir)
+        # Offsets of the method, CRC-32 and uncompressed size fields in the
+        # local file header and in the central directory record.
+        for method, crc_pos, size_pos in ((8, 14, 22),
+                                          (cd + 10, cd + 16, cd + 24)):
+            struct.pack_into('<H', data, method, zipfile.ZIP_ZSTANDARD)
+            struct.pack_into('<L', data, crc_pos, crc)
+            struct.pack_into('<L', data, size_pos, len(content))
+        return io.BytesIO(data)
+
+    def test_read_multiple_frames(self):
+        from compression import zstd
+        content = self.part1 + self.part2
+        member_data = zstd.compress(self.part1) + zstd.compress(self.part2)
+        with zipfile.ZipFile(self.make_zip(member_data, content)) as zf:
+            self.assertIsNone(zf.testzip())
+            self.assertEqual(zf.read('a'), content)
+            for size in (1, 7, 1000, len(content) + 1):
+                with self.subTest(size=size), zf.open('a') as f:
+                    chunks = iter(lambda: f.read(size), b'')
+                    self.assertEqual(b''.join(chunks), content)
+
+    def test_read_many_small_frames(self):
+        from compression import zstd
+        parts = [bytes([i]) * i for i in range(1, 50)]
+        content = b''.join(parts)
+        member_data = b''.join(zstd.compress(p) for p in parts)
+        with zipfile.ZipFile(self.make_zip(member_data, content)) as zf:
+            self.assertEqual(zf.read('a'), content)
+
+    def test_garbage_between_frames(self):
+        from compression import zstd
+        content = self.part1 + self.part2
+        member_data = (zstd.compress(self.part1) + b'garbage!' +
+                       zstd.compress(self.part2))
+        with zipfile.ZipFile(self.make_zip(member_data, content)) as zf:
+            with self.assertRaises(zstd.ZstdError):
+                zf.read('a')
+
+    def test_truncated_last_frame(self):
+        from compression import zstd
+        content = self.part1 + self.part2
+        member_data = (zstd.compress(self.part1) +
+                       zstd.compress(self.part2)[:-4])
+        with zipfile.ZipFile(self.make_zip(member_data, content)) as zf:
+            with self.assertRaises(zipfile.BadZipFile):
+                zf.read('a')
+
+
 class MonkeypatchedDecompressorTests(unittest.TestCase):
     # Some third-party projects monkey-patch _get_decompressor() to add
     # additional compression schemes. This can break at any time as the
diff --git a/Lib/zipfile/__init__.py b/Lib/zipfile/__init__.py
index 5963f84bdad2996..341dc3137ef9f0b 100644
--- a/Lib/zipfile/__init__.py
+++ b/Lib/zipfile/__init__.py
@@ -1197,6 +1197,15 @@ def _read1(self, n):
                 data += self._read2(n - len(data))
         elif self._compress_type == ZIP_STORED:
             data = self._read2(n)
+        elif (self._compress_type == ZIP_ZSTANDARD and
+              self._decompressor.eof):
+            # A zstd stream may consist of several frames (RFC 8878,
+            # section 3.1).  Decode the next one with a new decompressor;
+            # bytes that are not a valid frame raise ZstdError.
+            data = self._decompressor.unused_data
+            self._decompressor = _get_decompressor(ZIP_ZSTANDARD)
+            if not data:
+                data = self._read2(n)
         else:
             # bzip2/lzma/zstd: a bounded decompress() call may leave input
             # buffered inside the decompressor; drain that before reading more.
@@ -1233,9 +1242,16 @@ def _read1(self, n):
                     remove=(3, 21),
                     stacklevel=4)
                 data = self._decompressor.decompress(data)
-            self._eof = (self._decompressor.eof or
-                         self._compress_left <= 0 and
-                         getattr(self._decompressor, "needs_input", True))
+            if self._compress_type == ZIP_ZSTANDARD:
+                # The end of a zstd frame is not the end of the member.
+                self._eof = (self._compress_left <= 0 and
+                             (self._decompressor.needs_input or
+                              self._decompressor.eof and
+                              not self._decompressor.unused_data))
+            else:
+                self._eof = (self._decompressor.eof or
+                             self._compress_left <= 0 and
+                             getattr(self._decompressor, "needs_input", True))
 
         data = data[:self._left]
         self._left -= len(data)
diff --git a/Misc/NEWS.d/next/Library/2026-10-28-17-05-12.gh-issue-160842.Zs7mFr.rst b/Misc/NEWS.d/next/Library/2026-10-28-17-05-12.gh-issue-160842.Zs7mFr.rst
new file mode 100644
index 000000000000000..0fef76048347266
--- /dev/null
+++ b/Misc/NEWS.d/next/Library/2026-10-28-17-05-12.gh-issue-160842.Zs7mFr.rst
@@ -0,0 +1,3 @@
+:mod:`zipfile` now reads Zstandard-compressed members that consist of
+several frames, instead of returning only the first frame and raising
+:exc:`~zipfile.BadZipFile` on the CRC check.