https://github.com/python/cpython/commit/a4f28a52b4b54c34100ee0891b9a15640ed3a7b2 commit: a4f28a52b4b54c34100ee0891b9a15640ed3a7b2 branch: main author: John <[email protected]> committer: vstinner <[email protected]> date: 2026-10-01T21:39:34Z summary:
gh-158316: Avoid second buffer fetch on bytearray_decode() (#158485) Co-authored-by: Victor Stinner <[email protected]> files: A Misc/NEWS.d/next/Core_and_Builtins/2026-09-30-10-12-41.gh-issue-158316.Yv3kQe.rst M Lib/test/test_bytes.py M Objects/bytearrayobject.c diff --git a/Lib/test/test_bytes.py b/Lib/test/test_bytes.py index b55863256cc37c..ea0a54dd5e1d35 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 @@ -2195,6 +2196,30 @@ def delslice(): self.assertRaises(BufferError, delslice) self.assertEqual(b, orig) + def test_decode_resize_forbidden(self): + # The storage is pinned while it is decoded, so an error handler + # cannot resize the bytearray. + b = bytearray(b'ab\xffcd') + errors = 'test.bytearray_decode_resize' + def handler(exc): + self.assertRaises(BufferError, b.clear) + self.assertRaises(BufferError, b.append, 0) + return ('?', exc.end) + self.addCleanup(_codecs_unregister_error, errors) + codecs.register_error(errors, handler) + for encoding in 'utf-8', 'utf-8-sig': + with self.subTest(encoding=encoding): + self.assertEqual(b.decode(encoding, errors), 'ab?cd') + self.assertEqual(b, b'ab\xffcd') + + def test_decode_subclass_buffer(self): + # decode() decodes the buffer that the object exports. + class B(bytearray): + def __buffer__(self, flags): + return memoryview(b'other') + self.assertEqual(B(b'mine').decode(), 'other') + self.assertEqual(B(b'mine').decode('latin-1'), 'other') + @test.support.cpython_only def test_obsolete_write_lock(self): _testcapi = import_helper.import_module('_testcapi') diff --git a/Misc/NEWS.d/next/Core_and_Builtins/2026-09-30-10-12-41.gh-issue-158316.Yv3kQe.rst b/Misc/NEWS.d/next/Core_and_Builtins/2026-09-30-10-12-41.gh-issue-158316.Yv3kQe.rst new file mode 100644 index 00000000000000..c95bab19af393a --- /dev/null +++ b/Misc/NEWS.d/next/Core_and_Builtins/2026-09-30-10-12-41.gh-issue-158316.Yv3kQe.rst @@ -0,0 +1,2 @@ +On Free Threading, speed up :meth:`bytearray.decode` by decoding the underlying +storage directly instead of going through the buffer protocol. diff --git a/Objects/bytearrayobject.c b/Objects/bytearrayobject.c index caf8bad88a280d..496d04a1704d46 100644 --- a/Objects/bytearrayobject.c +++ b/Objects/bytearrayobject.c @@ -2599,7 +2599,19 @@ bytearray_decode_impl(PyByteArrayObject *self, const char *encoding, { if (encoding == NULL) encoding = PyUnicode_GetDefaultEncoding(); - return PyUnicode_FromEncodedObject((PyObject*)self, encoding, errors); + if (Py_TYPE(self)->tp_as_buffer->bf_getbuffer != bytearray_getbuffer) { + /* A subclass may export a different buffer. */ + return PyUnicode_FromEncodedObject((PyObject*)self, encoding, errors); + } + + /* Decode the storage directly instead of exporting a buffer, which + would re-acquire the critical section we already hold. Increase + exports to prevent the storage from changing during the decode. */ + self->ob_exports++; + PyObject *res = PyUnicode_Decode(PyByteArray_AS_STRING(self), + Py_SIZE(self), encoding, errors); + self->ob_exports--; + return res; } PyDoc_STRVAR(alloc_doc, _______________________________________________ Python-checkins mailing list -- [email protected] To unsubscribe send an email to [email protected] https://mail.python.org/mailman3//lists/python-checkins.python.org Member address: [email protected]
