python
diff --git a/Include/cpython/listobject.h b/Include/cpython/listobject.h
@@ -29,11 +29,7 @@ typedef struct {
 
 static inline Py_ssize_t PyList_GET_SIZE(PyObject *op) {
     PyListObject *list = _PyList_CAST(op);
-#ifdef Py_GIL_DISABLED
-    return _Py_atomic_load_ssize_relaxed(&(_PyVarObject_CAST(list)->ob_size));
-#else
     return Py_SIZE(list);
-#endif
 }
 #define PyList_GET_SIZE(op) PyList_GET_SIZE(_PyObject_CAST(op))
 

diff --git a/Include/object.h b/Include/object.h
@@ -240,7 +240,11 @@ _Py_SIZE_impl(PyObject *ob)
 {
     assert(Py_TYPE(ob) != &PyLong_Type);
     assert(Py_TYPE(ob) != &PyBool_Type);
-    return  _PyVarObject_CAST(ob)->ob_size;
+#ifdef Py_GIL_DISABLED
+    return _Py_atomic_load_ssize_relaxed(&(_PyVarObject_CAST(ob)->ob_size));
+#else
+    return _PyVarObject_CAST(ob)->ob_size;
+#endif
 }
 
 static inline int

diff --git a/Objects/bytearrayobject.c b/Objects/bytearrayobject.c
@@ -9,6 +9,7 @@
 #include "pycore_object.h"        // _PyObject_GC_UNTRACK()
 #include "pycore_strhex.h"        // _Py_strhex_with_sep()
 #include "pycore_long.h"          // _PyLong_FromUnsignedChar()
+#include "pycore_list.h"          // _PyList_GetItemRef
 #include "pycore_pyatomic_ft_wrappers.h"
 #include "bytesobject.h"
 

diff --git a/Objects/stringlib/join.h b/Objects/stringlib/join.h
@@ -60,7 +60,16 @@ STRINGLIB(bytes_join)(PyObject *sep, PyObject *iterable)
      */
     for (i = 0, nbufs = 0; i < seqlen; i++) {
         Py_ssize_t itemlen;
+#ifdef Py_GIL_DISABLED
+        if (PyList_Check(seq)) {
+            item = _PyList_GetItemRef((PyListObject *)seq, i);
+        }
+        else {
+            item = PyTuple_GET_ITEM(seq, i);
+        }
+#else
         item = PySequence_Fast_GET_ITEM(seq, i);
+#endif
         if (PyBytes_CheckExact(item)) {
             /* Fast path. */
             buffers[i].obj = Py_NewRef(item);