diff --git a/Objects/typeobject.c b/Objects/typeobject.c
index a107715808f..6a9bd701dfb 100644
--- a/Objects/typeobject.c
+++ b/Objects/typeobject.c
@@ -1021,6 +1021,38 @@ type_call(PyTypeObject *type, PyObject *args, PyObject *kwds)
     return obj;
 }
 
+PyObject *
+PyType_FromSpec_Alloc(PyTypeObject *type, Py_ssize_t nitems)
+{
+    PyObject *obj;
+    const size_t size = _Py_SIZE_ROUND_UP(
+            _PyObject_VAR_SIZE(type, nitems+1) + sizeof(traverseproc),
+            SIZEOF_VOID_P);
+    /* note that we need to add one, for the sentinel and space for the
+       provided tp-traverse: See bpo-40217 for more details */
+
+    if (PyType_IS_GC(type))
+        obj = _PyObject_GC_Malloc(size);
+    else
+        obj = (PyObject *)PyObject_MALLOC(size);
+
+    if (obj == NULL)
+        return PyErr_NoMemory();
+
+    obj = obj;
+
+    memset(obj, '\0', size);
+
+    if (type->tp_itemsize == 0)
+        (void)PyObject_INIT(obj, type);
+    else
+        (void) PyObject_INIT_VAR((PyVarObject *)obj, type, nitems);
+
+    if (PyType_IS_GC(type))
+        _PyObject_GC_TRACK(obj);
+    return obj;
+}
+
 PyObject *
 PyType_GenericAlloc(PyTypeObject *type, Py_ssize_t nitems)
 {
@@ -2853,6 +2885,36 @@ static const short slotoffsets[] = {
 #include "typeslots.inc"
 };
 
+static int
+PyType_FromSpec_tp_traverse(PyObject *self, visitproc visit, void *arg)
+{
+    PyTypeObject *parent = Py_TYPE(self);
+
+    // Only a instance of a type that is directly created by
+    // PyType_FromSpec (not subclasses) must visit its parent.
+    if (parent->tp_traverse == PyType_FromSpec_tp_traverse) {
+        Py_VISIT(parent);
+    }
+
+    // Search for the original type that was created using PyType_FromSpec
+    PyTypeObject *base;
+    base = parent;
+    while (base->tp_traverse != PyType_FromSpec_tp_traverse) {
+        base = base->tp_base;
+        assert(base);
+    }
+
+    // Extract the user defined traverse function that we placed at the end
+    // of the type and call it.
+    size_t size = Py_SIZE(base);
+    size_t _offset = _PyObject_VAR_SIZE(&PyType_Type, size+1);
+    traverseproc fun = *(traverseproc*)((char*)base + _offset);
+    if (fun == NULL) {
+        return 0;
+    }
+    return fun(self, visit, arg);
+}
+
 PyObject *
 PyType_FromSpecWithBases(PyType_Spec *spec, PyObject *bases)
 {
@@ -2886,7 +2948,7 @@ PyType_FromSpecWithBases(PyType_Spec *spec, PyObject *bases)
         }
     }
 
-    res = (PyHeapTypeObject*)PyType_GenericAlloc(&PyType_Type, nmembers);
+    res = (PyHeapTypeObject*)PyType_FromSpec_Alloc(&PyType_Type, nmembers);
     if (res == NULL)
         return NULL;
     res_start = (char*)res;
@@ -2991,6 +3053,26 @@ PyType_FromSpecWithBases(PyType_Spec *spec, PyObject *bases)
             memcpy(PyHeapType_GET_MEMBERS(res), slot->pfunc, len);
             type->tp_members = PyHeapType_GET_MEMBERS(res);
         }
+        else if (slot->slot == Py_tp_traverse) {
+
+           /* Types created by PyType_FromSpec own a strong reference to their
+            * type, but this was added in Python 3.8. The tp_traverse function
+            * needs to call Py_VISIT on the type but all existing traverse
+            * functions cannot be updated (especially the ones from existing user
+            * functions) so we need to provide a tp_traverse that manually calls
+            * Py_VISIT(Py_TYPE(self)) and then call the provided tp_traverse. In
+            * this way, user functions do not need to be updated, preserve
+            * backwards compatibility.
+            *
+            * We store the user-provided traverse function at the end of the type
+            * (we have allocated space for it) so we can call it from our
+            * PyType_FromSpec_tp_traverse wrapper. */
+
+            type->tp_traverse = PyType_FromSpec_tp_traverse;
+            size_t _offset = _PyObject_VAR_SIZE(&PyType_Type, nmembers+1);
+            traverseproc *user_traverse = (traverseproc*)((char*)type + _offset);
+            *user_traverse = slot->pfunc;
+        }
         else {
             /* Copy other slots directly */
             *(void**)(res_start + slotoffsets[slot->slot]) = slot->pfunc;