Issue #1621: Fix undefined behaviour from signed overflow in datetime module hashes, array and list iterations, and get_integer (stringlib/string_format.h)

2 files changed, 9 insertions, 11 deletions

diff --git a/Objects/listobject.c b/Objects/listobject.c
index cada70857c..1c516e962b 100644
--- a/Objects/listobject.c
+++ b/Objects/listobject.c
@@ -2434,7 +2434,7 @@ list_subscript(PyListObject* self, PyObject* item)
             src = self->ob_item;
             dest = ((PyListObject *)result)->ob_item;
             for (cur = start, i = 0; i < slicelength;
-                 cur += step, i++) {
+                 cur += (size_t)step, i++) {
                 it = src[cur];
                 Py_INCREF(it);
                 dest[i] = it;
@@ -2525,7 +2525,7 @@ list_ass_subscript(PyListObject* self, PyObject* item, PyObject* value)
                     self->ob_item + cur + 1,
                     lim * sizeof(PyObject *));
             }
-            cur = start + slicelength*step;
+            cur = start + (size_t)slicelength * step;
             if (cur < (size_t)Py_SIZE(self)) {
                 memmove(self->ob_item + cur - slicelength,
                     self->ob_item + cur,
@@ -2589,7 +2589,7 @@ list_ass_subscript(PyListObject* self, PyObject* item, PyObject* value)
             selfitems = self->ob_item;
             seqitems = PySequence_Fast_ITEMS(seq);
             for (cur = start, i = 0; i < slicelength;
-                 cur += step, i++) {
+                 cur += (size_t)step, i++) {
                 garbage[i] = selfitems[cur];
                 ins = seqitems[i];
                 Py_INCREF(ins);
diff --git a/Objects/stringlib/string_format.h b/Objects/stringlib/string_format.h
index 6c7adcb01e..d992b6f53c 100644
--- a/Objects/stringlib/string_format.h
+++ b/Objects/stringlib/string_format.h
@@ -209,19 +209,17 @@ get_integer(const SubString *str)
         if (digitval < 0)
             return -1;
         /*
-           This trick was copied from old Unicode format code.  It's cute,
-           but would really suck on an old machine with a slow divide
-           implementation.  Fortunately, in the normal case we do not
-           expect too many digits.
+           Detect possible overflow before it happens:
+
+              accumulator * 10 + digitval > PY_SSIZE_T_MAX if and only if
+              accumulator > (PY_SSIZE_T_MAX - digitval) / 10.
         */
-        oldaccumulator = accumulator;
-        accumulator *= 10;
-        if ((accumulator+10)/10 != oldaccumulator+1) {
+        if (accumulator > (PY_SSIZE_T_MAX - digitval) / 10) {
             PyErr_Format(PyExc_ValueError,
                          "Too many decimal digits in format string");
             return -1;
         }
-        accumulator += digitval;
+        accumulator = accumulator * 10 + digitval;
     }
     return accumulator;
 }