amend

test/test_tensordict.py

@@ -752,6 +752,7 @@ def test_getitem_nested(self):
         assert sub_tensordict.shape == torch.Size([4, 5])
         assert sub_sub_tensordict.shape == torch.Size([4, 5, 6])
 
+    @set_lazy_legacy(True)
     def test_inferred_view_size(self):
         td = TensorDict({"a": torch.randn(3, 4)}, [3, 4])
         assert td.view(-1).view(-1, 4) is td
@@ -2539,12 +2540,13 @@ def test_inferred_view_size(self, td_name, device):
                 dim_size if dim_idx != i else -1
                 for dim_idx, dim_size in enumerate(td.shape)
             ]
-            if td_name in ("td_params",):
-                assert td.view(-1).view(*new_shape)._param_td is td._param_td
-                assert td.view(*new_shape)._param_td is td._param_td
-            else:
-                assert td.view(*new_shape) is td
-                assert td.view(-1).view(*new_shape) is td
+            if lazy_legacy():
+                if td_name in ("td_params",):
+                    assert td.view(-1).view(*new_shape)._param_td is td._param_td
+                    assert td.view(*new_shape)._param_td is td._param_td
+                else:
+                    assert td.view(*new_shape) is td
+                    assert td.view(-1).view(*new_shape) is td
 
     def test_items_values_keys(self, td_name, device):
         torch.manual_seed(1)
@@ -3102,6 +3104,10 @@ def test_nestedtensor_stack(self, td_name, device, dim, key):
         # cloning is type-preserving: we can do that operation
         td_stack.clone()
 
+    # This test fails on lazy tensordicts when lazy-legacy is False
+    # Deprecating lazy modules will make this decorator useless (the test should
+    # still run ok).
+    @set_lazy_legacy(True)
     def test_non_tensor_data(self, td_name, device):
         td = getattr(self, td_name)(device)
         # check lock
@@ -3132,6 +3138,10 @@ def test_non_tensor_data(self, td_name, device):
             assert isinstance(td.get(("this", "other", "tensor")), NonTensorData)
             assert td.get_non_tensor(("this", "other", "tensor")) == "success"
 
+    # This test fails on lazy tensordicts when lazy-legacy is False
+    # Deprecating lazy modules will make this decorator useless (the test should
+    # still run ok).
+    @set_lazy_legacy(True)
     def test_non_tensor_data_flatten_keys(self, td_name, device):
         td = getattr(self, td_name)(device)
         with td.unlock_():
@@ -3149,6 +3159,10 @@ def test_non_tensor_data_flatten_keys(self, td_name, device):
         assert (td_flat.get("this.tensor") == 0).all()
         assert td_flat.get_non_tensor("this.will") == "succeed"
 
+    # This test fails on lazy tensordicts when lazy-legacy is False
+    # Deprecating lazy modules will make this decorator useless (the test should
+    # still run ok).
+    @set_lazy_legacy(True)
     def test_non_tensor_data_pickle(self, td_name, device, tmpdir):
         td = getattr(self, td_name)(device)
         with td.unlock_():
@@ -3484,6 +3498,7 @@ def test_select_exception(self, td_name, device, strict):
             assert td2 is not td
             assert len(list(td2.keys())) == 0
 
+    @set_lazy_legacy(True)
     def test_set_lazy_legacy(self, td_name, device):
         if td_name in (
             "sub_td",
@@ -3493,16 +3508,19 @@ def test_set_lazy_legacy(self, td_name, device):
             "unsqueezed_td",
             "permute_td",
             "transpose_td",
-            "nested_stacked_td",
-            "stacked_td",
         ):
             raiser = pytest.raises(RuntimeError)
+            raiser_view = raiser
+        elif "stack" in td_name:
+            raiser = contextlib.nullcontext()
+            raiser_view = pytest.raises(RuntimeError)
         else:
             raiser = contextlib.nullcontext()
+            raiser_view = raiser
 
-        def test_not_id(td, td_name=td_name, raiser=raiser):
+        def test_not_id(td, td_name=td_name, raiser=raiser, raiser_view=raiser_view):
             # view
-            with raiser:
+            with raiser_view:
                 td_view = td.view(-1).view(td.shape)
                 if td_name in ("td_params",):
                     assert isinstance(td_view, TensorDict)
@@ -4039,7 +4057,24 @@ def test_stack_onto(self, td_name, device, tmpdir):
             else:
                 td1.apply_(lambda x: x.zero_() + 1)
 
-        td_out = td.unsqueeze(1).expand(td.shape[0], 2, *td.shape[1:]).clone()
+        is_lazy = td_name in (
+            "sub_td",
+            "sub_td2",
+            "permute_td",
+            "unsqueezed_td",
+            "squeezed_td",
+            "td_h5",
+        ) and not lazy_legacy()
+        error_dec = (
+            pytest.raises(RuntimeError, match="Make it dense")
+            if is_lazy
+            else contextlib.nullcontext()
+        )
+        with error_dec:
+            td_out = td.unsqueeze(1)
+        if is_lazy:
+            return
+        td_out = td_out.expand(td.shape[0], 2, *td.shape[1:]).clone()
         td_stack = LazyStackedTensorDict.lazy_stack([td0, td1], 1)
         if td_name == "td_params":
             with pytest.raises(RuntimeError, match="out.batch_size and stacked"):