GH-96793: Change `FOR_ITER` to not pop the iterator on exhaustion. (GH-96801)

Change FOR_ITER to have the same stack effect regardless of whether it branches or not.
Performance is unchanged as FOR_ITER (and specialized forms jump over the cleanup code).

1 files changed, 13 insertions, 3 deletions

diff --git a/Python/ceval.c b/Python/ceval.c
index 35ce767..ac995d7 100644
--- a/Python/ceval.c
+++ b/Python/ceval.c
@@ -1298,6 +1298,14 @@ handle_eval_breaker:
             DISPATCH();
         }
 
+        TARGET(END_FOR) {
+            PyObject *value = POP();
+            Py_DECREF(value);
+            value = POP();
+            Py_DECREF(value);
+            DISPATCH();
+        }
+
         TARGET(UNARY_POSITIVE) {
             PyObject *value = TOP();
             PyObject *res = PyNumber_Positive(value);
@@ -3844,9 +3852,11 @@ handle_eval_breaker:
                 _PyErr_Clear(tstate);
             }
             /* iterator ended normally */
+            assert(_Py_OPCODE(next_instr[INLINE_CACHE_ENTRIES_FOR_ITER + oparg] == END_FOR));
             STACK_SHRINK(1);
             Py_DECREF(iter);
-            JUMPBY(INLINE_CACHE_ENTRIES_FOR_ITER + oparg);
+            /* Skip END_FOR */
+            JUMPBY(INLINE_CACHE_ENTRIES_FOR_ITER + oparg + 1);
             DISPATCH();
         }
 
@@ -3884,7 +3894,7 @@ handle_eval_breaker:
             }
             STACK_SHRINK(1);
             Py_DECREF(it);
-            JUMPBY(INLINE_CACHE_ENTRIES_FOR_ITER + oparg);
+            JUMPBY(INLINE_CACHE_ENTRIES_FOR_ITER + oparg + 1);
             DISPATCH();
         }
 
@@ -3898,7 +3908,7 @@ handle_eval_breaker:
             if (r->index >= r->len) {
                 STACK_SHRINK(1);
                 Py_DECREF(r);
-                JUMPBY(INLINE_CACHE_ENTRIES_FOR_ITER + oparg);
+                JUMPBY(INLINE_CACHE_ENTRIES_FOR_ITER + oparg + 1);
                 DISPATCH();
             }
             long value = (long)(r->start +