gh-144569: Avoid creating temporary objects in BINARY_SLICE for bytes

Author: koxudaxi

Include/internal/pycore_bytesobject.h

@@ -18,6 +18,7 @@ extern PyObject* _PyBytes_FormatEx(
  * specializing interpreter. Unlike PyBytes_Concat(), this returns a new
  * reference rather than modifying its first argument in place. */
 extern PyObject* _PyBytes_Concat(PyObject *a, PyObject *b);
+PyAPI_FUNC(PyObject *) _PyBytes_BinarySlice(PyObject *, PyObject *, PyObject *);
 
 extern PyObject* _PyBytes_FromHex(
     PyObject *string,

Lib/test/test_bytes.py

@@ -1168,6 +1168,58 @@ def test_getitem_error(self):
         with self.assertRaisesRegex(TypeError, msg):
             b['a']
 
+    def test_binary_slice(self):
+        def binary_slice(data, start, stop):
+            return data[start:stop]
+
+        data = b'0123456789'
+        indices = (None, 0, 1, 5, 10, 20, -1, -5, -10, -20,
+                   sys.maxsize, -sys.maxsize - 1, 10**100, -10**100)
+        for start in indices:
+            for stop in indices:
+                with self.subTest(start=start, stop=stop):
+                    self.assertEqual(binary_slice(data, start, stop),
+                                     data[slice(start, stop)])
+
+        self.assertIs(binary_slice(data, None, None), data)
+
+        calls = []
+
+        class Index:
+            def __init__(self, name, value):
+                self.name = name
+                self.value = value
+
+            def __index__(self):
+                calls.append(self.name)
+                return self.value
+
+        self.assertEqual(binary_slice(data, Index('start', 2),
+                                     Index('stop', 5)), b'234')
+        self.assertEqual(calls, ['start', 'stop'])
+
+        calls.clear()
+
+        class BadIndex:
+            def __index__(self):
+                calls.append('start')
+                raise ValueError('bad index')
+
+        with self.assertRaisesRegex(ValueError, 'bad index'):
+            binary_slice(data, BadIndex(), Index('stop', 5))
+        self.assertEqual(calls, ['start'])
+
+        msg = "slice indices must be integers or have an __index__ method"
+        with self.assertRaisesRegex(TypeError, msg):
+            binary_slice(data, 1.5, 5)
+
+        class SliceOverride(bytes):
+            def __getitem__(self, key):
+                return key
+
+        key = binary_slice(SliceOverride(data), 2, 5)
+        self.assertEqual(key, slice(2, 5))
+
     def test_buffer_is_readonly(self):
         fd = os.open(__file__, os.O_RDONLY)
         with open(fd, "rb", buffering=0) as f:

Misc/NEWS.d/next/Core_and_Builtins/2026-08-24-15-30-00.gh-issue-144569.Bn7qKx.rst

@@ -0,0 +1,2 @@
+Optimize ``BINARY_SLICE`` for :class:`bytes` by avoiding temporary
+:class:`slice` object creation.

Modules/_testinternalcapi/test_cases.c.h

@@ -1729,6 +1729,19 @@ bytes_hash(PyObject *self)
     return hash;
 }
 
+static PyObject *
+bytes_slice(PyObject *op, Py_ssize_t start, Py_ssize_t length)
+{
+    if (length <= 0) {
+        return Py_GetConstant(Py_CONSTANT_EMPTY_BYTES);
+    }
+    if (start == 0 && length == PyBytes_GET_SIZE(op) &&
+        PyBytes_CheckExact(op)) {
+        return Py_NewRef(op);
+    }
+    return PyBytes_FromStringAndSize(PyBytes_AS_STRING(op) + start, length);
+}
+
 static PyObject*
 bytes_subscript(PyObject *op, PyObject* item)
 {
@@ -1759,18 +1772,11 @@ bytes_subscript(PyObject *op, PyObject* item)
         slicelength = PySlice_AdjustIndices(PyBytes_GET_SIZE(self), &start,
                                             &stop, step);
 
-        if (slicelength <= 0) {
-            return Py_GetConstant(Py_CONSTANT_EMPTY_BYTES);
-        }
-        else if (start == 0 && step == 1 &&
-                 slicelength == PyBytes_GET_SIZE(self) &&
-                 PyBytes_CheckExact(self)) {
-            return Py_NewRef(self);
+        if (step == 1) {
+            return bytes_slice(op, start, slicelength);
         }
-        else if (step == 1) {
-            return PyBytes_FromStringAndSize(
-                PyBytes_AS_STRING(self) + start,
-                slicelength);
+        else if (slicelength <= 0) {
+            return Py_GetConstant(Py_CONSTANT_EMPTY_BYTES);
         }
         else {
             source_buf = PyBytes_AS_STRING(self);
@@ -1795,6 +1801,18 @@ bytes_subscript(PyObject *op, PyObject* item)
     }
 }
 
+PyObject *
+_PyBytes_BinarySlice(PyObject *container, PyObject *start_o, PyObject *stop_o)
+{
+    assert(PyBytes_CheckExact(container));
+    Py_ssize_t len = PyBytes_GET_SIZE(container);
+    Py_ssize_t istart, istop;
+    if (!_PyEval_UnpackIndices(start_o, stop_o, len, &istart, &istop)) {
+        return NULL;
+    }
+    return bytes_slice(container, istart, istop - istart);
+}
+
 static int
 bytes_buffer_getbuffer(PyObject *op, Py_buffer *view, int flags)
 {

Objects/bytesobject.c

@@ -1096,6 +1096,9 @@ dummy_func(
             else if (PyUnicode_CheckExact(container_o)) {
                 res_o = _PyUnicode_BinarySlice(container_o, start_o, stop_o);
             }
+            else if (PyBytes_CheckExact(container_o)) {
+                res_o = _PyBytes_BinarySlice(container_o, start_o, stop_o);
+            }
             else {
                 PyObject *slice = PySlice_New(start_o, stop_o, NULL);
                 if (slice == NULL) {

Python/bytecodes.c

@@ -2469,11 +2469,12 @@ dummy_func(void) {
     }
 
     op(_BINARY_SLICE, (container, start, stop -- res)) {
-        // Slicing a string/list/tuple always returns the same type.
+        // Slicing a string/list/tuple/bytes always returns the same type.
         PyTypeObject *type = sym_get_type(container);
         if (type == &PyUnicode_Type ||
             type == &PyList_Type ||
-            type == &PyTuple_Type)
+            type == &PyTuple_Type ||
+            type == &PyBytes_Type)
         {
             res = sym_new_type(ctx, type);
         }