DDE issue:set dictionary element flag

mindspore/core/ir/anf.cc

@@ -730,13 +730,21 @@ void SetSequenceElementsUseFlagsRecursively(const AbstractBasePtr &abs, bool new
 
   SetSequenceElementsUseFlags(abs, new_flag);
 
-  // Check its elements if it's sequence node.
+  // Check its elements if it's a sequence node.
   auto sequence_abs = dyn_cast<abstract::AbstractSequence>(abs);
-  if (sequence_abs == nullptr) {
+  if (sequence_abs != nullptr) {
+    for (auto &element : sequence_abs->elements()) {
+      SetSequenceElementsUseFlagsRecursively(element, new_flag);
+    }
     return;
   }
-  for (auto &element : sequence_abs->elements()) {
-    SetSequenceElementsUseFlagsRecursively(element, new_flag);
+
+  // Check its elements if it's a dictionary node.
+  auto dictionary_abs = dyn_cast<abstract::AbstractDictionary>(abs);
+  if (dictionary_abs != nullptr) {
+    for (auto &element : dictionary_abs->elements()) {
+      SetSequenceElementsUseFlagsRecursively(element.second, new_flag);
+    }
   }
 }
 }  // namespace mindspore

tests/st/control/test_list_append.py

@@ -18,6 +18,7 @@ import mindspore.context as context
 from mindspore import Tensor, dtype
 from mindspore.nn import Cell
 import mindspore.ops.operations as P
+import numpy as np
 
 
 @pytest.mark.skip(reason='Not support list  as parameter in while function yet')
@@ -83,3 +84,40 @@ def test_for_list():
     x2 = Tensor([[0., 0.], [0., 0.], [0., 0.], [0., 0.], [0., 0.]], dtype.float32)
     net = Net()
     print(net(x1, x2))
+
+
+@pytest.mark.level0
+@pytest.mark.platform_x86_gpu_training
+@pytest.mark.platform_arm_ascend_training
+@pytest.mark.platform_x86_ascend_training
+@pytest.mark.env_onecard
+def test_dictionary_list():
+    """
+    Feature: dictionary list.
+    Description: Infer list in dictionary.
+    Expectation: Null.
+    """
+
+    class D3rNet(Cell):
+        def __init__(self):
+            super().__init__()
+            self.a = Tensor(np.random.randn(300, 9).astype(np.float32))
+            self.b = Tensor(np.random.randn(300).astype(np.float32))
+            self.c = Tensor(np.ones([300]).astype(np.int32))
+
+        def construct(self, a, b, c):
+            a_o = a * self.a
+            b_o = b * self.b
+            c_o = c * self.c
+            pts = [[a_o, b_o, c_o]]
+            bbox = []
+            for i in pts:
+                bbox.append({"ptx:": i})
+            return bbox
+
+    a = Tensor(np.random.randn(300, 9).astype(np.float32))
+    b = Tensor(np.random.randn(300).astype(np.float32))
+    c = Tensor(np.ones([300]).astype(np.int32))
+    net = D3rNet()
+    bbox = net(a, b, c)
+    print(bbox)