Merge pull request #7275 from rapidsai/branch-0.18

[gpuCI] Auto-merge branch-0.18 to branch-0.19 [skip ci]
rapidsai · Feb 1, 2021 · d9ed2b1 · d9ed2b1
2 parents 9718f5c + 0ee8004
commit d9ed2b1
Show file tree

Hide file tree

Showing 3 changed files with 14 additions and 1 deletion.
diff --git a/cpp/src/io/parquet/reader_impl.cu b/cpp/src/io/parquet/reader_impl.cu
@@ -669,7 +669,11 @@ class aggregate_metadata {
       for (const auto &use_name : local_use_names) {
         for (size_t schema_idx = 1; schema_idx < pfm.schema.size(); schema_idx++) {
           auto const &schema = pfm.schema[schema_idx];
-          if (use_name == schema.name) { output_column_schemas.push_back(schema_idx); }
+          // We select only top level columns by name. Selecting nested columns by name is not
+          // supported. Top level columns are identified by their parent being the root (idx == 0)
+          if (use_name == schema.name and schema.parent_idx == 0) {
+            output_column_schemas.push_back(schema_idx);
+          }
         }
       }
     }

diff --git a/python/cudf/cudf/tests/data/parquet/nested_column_map.parquet b/python/cudf/cudf/tests/data/parquet/nested_column_map.parquet
diff --git a/python/cudf/cudf/tests/test_parquet.py b/python/cudf/cudf/tests/test_parquet.py
@@ -602,6 +602,15 @@ def test_parquet_reader_mixedcompression(datadir):
     assert_eq(expect, got)
 
 
+def test_parquet_reader_select_columns(datadir):
+    fname = datadir / "nested_column_map.parquet"
+
+    expect = cudf.read_parquet(fname).to_pandas()[["value"]]
+    got = cudf.read_parquet(fname, columns=["value"])
+
+    assert_eq(expect, got)
+
+
 def test_parquet_reader_invalids(tmpdir):
     test_pdf = make_pdf(nrows=1000, nvalids=1000 // 4, dtype=np.int64)