[REVIEW] Fix an error in IO with GzipFile type

python/cudf/cudf/tests/test_json.py

@@ -1,6 +1,7 @@
 # Copyright (c) 2018-2022, NVIDIA CORPORATION.
 
 import copy
+import gzip
 import itertools
 import os
 from io import BytesIO, StringIO
@@ -943,3 +944,14 @@ def test_order_nested_json_reader(tag, data):
     )
 
     assert_eq(expected, target, check_dtype=True)
+
+
+def test_json_round_trip_gzip():
+    df = cudf.DataFrame({"a": [1, 2, 3], "b": ["abc", "def", "ghi"]})
+    bytes = BytesIO()
+    with gzip.open(bytes, mode="wb") as fo:
+        df.to_json(fo, orient="records", lines=True)
+    bytes.seek(0)
+    with gzip.open(bytes, mode="rb") as fo:
+        written_df = cudf.read_json(fo, orient="records", lines=True)
+    assert_eq(written_df, df)

python/cudf/cudf/utils/ioutils.py

@@ -1953,7 +1953,15 @@ def _fsspec_data_transfer(
 
     # Calculate total file size
     if file_like:
-        file_size = path_or_fob.size
+        try:
+            file_size = path_or_fob.size
+        except AttributeError:
+            # Find file size of there is no `size`
+            # attribute
+            old_file_position = path_or_fob.tell()
+            path_or_fob.seek(0, os.SEEK_END)
+            file_size = path_or_fob.tell()
+            path_or_fob.seek(old_file_position, os.SEEK_SET)
     file_size = file_size or fs.size(path_or_fob)
 
     # Check if a direct read makes the most sense