gh-158316: Avoid second buffer fetch on bytearray_decode

[johng]

• Issue: Unnecessary lock in bytearray.decode #158316

[vstinner]

Please unregister the error handler when done:

diff --git a/Lib/test/test_bytes.py b/Lib/test/test_bytes.py
index b923f31178c..97460f1fc57 100644
--- a/Lib/test/test_bytes.py
+++ b/Lib/test/test_bytes.py
@@ -18,6 +18,7 @@
 import textwrap
 import threading
 import unittest
+from _codecs import _unregister_error as _codecs_unregister_error
 
 import test.support
 from test import support
@@ -2203,6 +2204,7 @@ def handler(exc):
             self.assertRaises(BufferError, b.clear)
             self.assertRaises(BufferError, b.append, 0)
             return ('?', exc.end)
+        self.addCleanup(_codecs_unregister_error, 'test.bytearray_decode_resize')
         codecs.register_error('test.bytearray_decode_resize', handler)
         for encoding in 'utf-8', 'utf-8-sig':
             with self.subTest(encoding=encoding):

[vstinner]

Note: I wanted to suggest replacing Py_SIZE() with PyByteArray_GET_SIZE() but I see that it uses an atomic operation on Free Threading, which is not needed in a critical section. Hum. PyByteArray_GET_SIZE() usage in this file is not really consistent anyway.

[vstinner]

Let's be confident in the implementation: checking the bytearray for modification only once at the end should be enough :-)

Suggested change

	self.assertEqual(b, b'ab\xffcd')
	self.assertEqual(b, b'ab\xffcd')

[vstinner]

LGTM

[vstinner]

Merged, thanks for the optimization.