Handle zero-column files. Closes #361. (#363)

fastparquet/api.py

@@ -153,8 +153,10 @@ def _set_attrs(self):
                 self.group_files.setdefault(i, set()).add(chunk.file_path)
         self.schema = schema.SchemaHelper(self._schema)
         self.selfmade = self.created_by.split(' ', 1)[0] == "fastparquet-python"
-        self.file_scheme = get_file_scheme([rg.columns[0].file_path
-                                           for rg in self.row_groups])
+        files = [rg.columns[0].file_path
+                 for rg in self.row_groups
+                 if rg.columns]
+        self.file_scheme = get_file_scheme(files)
         self._read_partitions()
         self._dtypes()
 
@@ -215,7 +217,7 @@ def _read_partitions(self):
                                 for key, v in cats.items()])
 
     def row_group_filename(self, rg):
-        if rg.columns[0].file_path:
+        if rg.columns and rg.columns[0].file_path:
             base = self.fn.replace('_metadata', '').rstrip('/')
             if base:
                 return join_path(base, rg.columns[0].file_path)

fastparquet/test/test_read.py

@@ -335,3 +335,18 @@ def test_multi_index_category(tempdir):
     assert dg.index.levels[0].dtype == '<M8[ns]'
     assert dg.index.levels[1].name == 'b'
     assert dg.equals(df)
+
+def test_no_columns(tempdir):
+    # https://github.com/dask/fastparquet/issues/361
+    # Create a non-empty DataFrame, then select no columns. That way we get
+    # _some_ rows, _no_ columns.
+    #
+    # df = pd.DataFrame({"A": [1, 2]})[[]]
+    # fastparquet.write("test-data/no_columns.parquet", df)
+    pf = fastparquet.ParquetFile(os.path.join(TEST_DATA, "no_columns.parquet"))
+    assert pf.count == 2
+    assert pf.columns == []
+    result = pf.to_pandas()
+    expected = pd.DataFrame({"A": [1, 2]})[[]]
+    assert len(result) == 2
+    pd.testing.assert_frame_equal(result, expected)