From 6bc39ae06a9003cb8c99f24420c487c9f1a8f822 Mon Sep 17 00:00:00 2001
From: Jonathan Baldie <jon@jonbaldie.com>
Date: Mon, 5 Oct 2026 15:23:07 +0100
Subject: [PATCH] gh-158858: Stop re-copying the tarfile stream buffer for
 every block

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---
 Lib/tarfile.py                                | 10 +++++++---
 Lib/test/test_tarfile.py                      | 20 +++++++++++++++++++
 ...-10-05-16-05-00.gh-issue-158858.Tb4rWq.rst |  2 ++
 3 files changed, 29 insertions(+), 3 deletions(-)
 create mode 100644 Misc/NEWS.d/next/Library/2026-10-05-16-05-00.gh-issue-158858.Tb4rWq.rst

diff --git a/Lib/tarfile.py b/Lib/tarfile.py
index a4f9ce3311f6dad..d3fe3341d751ab9 100644
--- a/Lib/tarfile.py
+++ b/Lib/tarfile.py
@@ -478,9 +478,13 @@ def __write(self, s):
            is ready to be written.
         """
         self.buf += s
-        while len(self.buf) > self.bufsize:
-            self.fileobj.write(self.buf[:self.bufsize])
-            self.buf = self.buf[self.bufsize:]
+        if len(self.buf) > self.bufsize:
+            # Write whole blocks from buf and keep the rest. Slicing off
+            # each block would copy the whole remaining buffer every time.
+            end = (len(self.buf) - 1) // self.bufsize * self.bufsize
+            for i in range(0, end, self.bufsize):
+                self.fileobj.write(self.buf[i:i + self.bufsize])
+            self.buf = self.buf[end:]
 
     def close(self):
         """Close the _Stream object. No operation should be
diff --git a/Lib/test/test_tarfile.py b/Lib/test/test_tarfile.py
index 55f052d0504d67b..b3898d38b5c83dc 100644
--- a/Lib/test/test_tarfile.py
+++ b/Lib/test/test_tarfile.py
@@ -2977,6 +2977,26 @@ def test_pax_limits(self):
 
 class MiscTest(unittest.TestCase):
 
+    def test_stream_write_large_chunks(self):
+        # A chunk spanning many blocks is still written one block at a time.
+        data = bytes(range(256)) * 1024
+        writes = []
+        class Recorder(io.BytesIO):
+            def write(self, b):
+                writes.append(b)
+                return super().write(b)
+        fobj = Recorder()
+        with tarfile.open(fileobj=fobj, mode='w|',
+                          copybufsize=len(data)) as tar:
+            tarinfo = tarfile.TarInfo('data')
+            tarinfo.size = len(data)
+            tar.addfile(tarinfo, io.BytesIO(data))
+        self.assertEqual({(type(b), len(b)) for b in writes},
+                         {(bytes, tarfile.RECORDSIZE)})
+        fobj.seek(0)
+        with tarfile.open(fileobj=fobj, mode='r|') as tar:
+            self.assertEqual(tar.extractfile(tar.next()).read(), data)
+
     def test_char_fields(self):
         self.assertEqual(tarfile.stn("foo", 8, "ascii", "strict"),
                          b"foo\0\0\0\0\0")
diff --git a/Misc/NEWS.d/next/Library/2026-10-05-16-05-00.gh-issue-158858.Tb4rWq.rst b/Misc/NEWS.d/next/Library/2026-10-05-16-05-00.gh-issue-158858.Tb4rWq.rst
new file mode 100644
index 000000000000000..85d013aa39d8508
--- /dev/null
+++ b/Misc/NEWS.d/next/Library/2026-10-05-16-05-00.gh-issue-158858.Tb4rWq.rst
@@ -0,0 +1,2 @@
+Speed up writing :mod:`tarfile` streams (``w|`` modes) with a large
+*copybufsize*: each chunk is no longer re-copied once per output block.