Fix construction of nested structs with EMPTY child

cpp/src/structs/utilities.cpp

@@ -309,6 +309,11 @@ void superimpose_parent_nulls(bitmask_type const* parent_null_mask,
                               rmm::cuda_stream_view stream,
                               rmm::mr::device_memory_resource* mr)
 {
+  if (child.type().id() == cudf::type_id::EMPTY) {
+    // EMPTY columns should not have a null mask,
+    // so don't superimpose null mask on empty columns.
+    return;
+  }
   if (!child.nullable()) {
     // Child currently has no null mask. Copy parent's null mask.
     child.set_null_mask(cudf::detail::copy_bitmask(parent_null_mask, 0, child.size(), stream, mr));

cpp/tests/structs/structs_column_tests.cpp

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2020-2021, NVIDIA CORPORATION.
+ * Copyright (c) 2020-2022, NVIDIA CORPORATION.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
@@ -623,4 +623,21 @@ TYPED_TEST(TypedStructColumnWrapperTest, CopyColumnFromView)
                                       cudf::column(list_of_structs_column->view()));
 }
 
+TEST_F(StructColumnWrapperTest, TestStructsColumnWithEmptyChild)
+{
+  // structs_column_views should not superimpose their null mask onto any EMPTY children,
+  // because EMPTY columns cannot have a null mask. This test ensures that
+  // we can construct a structs column with a parent null mask and an EMPTY
+  // child and then view it.
+  auto empty_col =
+    std::make_unique<cudf::column>(cudf::data_type(cudf::type_id::EMPTY), 3, rmm::device_buffer{});
+  int num_rows{empty_col->size()};
+  vector_of_columns cols;
+  cols.push_back(std::move(empty_col));
+  auto mask_vec    = std::vector<bool>{true, false false};
+  auto mask        = cudf::test::detail::make_null_mask(mask_vec.begin(), mask_vec.end());
+  auto structs_col = cudf::make_structs_column(num_rows, std::move(cols), 2, std::move(mask));
+  EXPECT_NO_THROW(structs_col->view());
+}
+
 CUDF_TEST_PROGRAM_MAIN()

python/cudf/cudf/_lib/types.pyx

@@ -93,6 +93,10 @@ SUPPORTED_NUMPY_TO_LIBCUDF_TYPES = {
 }
 
 LIBCUDF_TO_SUPPORTED_NUMPY_TYPES = {
+    # There's no equivalent to EMPTY in cudf.  We translate EMPTY
+    # columns from libcudf to ``int8`` columns of all nulls in Python.
+    # ``int8`` is chosen because it uses the least amount of memory.
+    TypeId.EMPTY: np.dtype("int8"),
     TypeId.INT8: np.dtype("int8"),
     TypeId.INT16: np.dtype("int16"),
     TypeId.INT32: np.dtype("int32"),

python/cudf/cudf/tests/test_struct.py

@@ -322,3 +322,10 @@ def test_struct_int_values():
     assert isinstance(actual_series[0]["b"], int)
     assert isinstance(actual_series[1]["b"], type(None))
     assert isinstance(actual_series[2]["b"], int)
+
+
+def test_nested_struct_from_pandas_empty():
+    pdf = pd.Series([[{"c": {"x": None}}], [{"c": None}]])
+    gdf = cudf.from_pandas(pdf)
+
+    assert_eq(pdf, gdf)