[BUG] File written by ORC writer cannot be read in Pandas and Spark

[ttnghia]

For the attached parquet file, if we read it then write it out into an .orc file using cudf ORC writer then the output file cannot be read in Pandas or Spark.

For more details, the given file contains a map with content like this:

+-----------+
|        _c0|
+-----------+
|       null|
|         {}|
|{B -> null}|
|{B -> null}|
|{B -> null}|
|    {A -> }|
|       null|
|{B -> null}|
|{B -> null}|
|    {A -> }|
|       null|
|    {A -> }|
|    {A -> }|
|{B -> null}|
|    {A -> }|
|         {}|
|{B -> null}|
|{B -> null}|
|       null|
|{B -> null}|
+-----------+

And code to read/write file in cudf:

TEST_F(OrcReaderTest, Write)
{
  cudf::io::parquet_reader_options read_opts = cudf::io::parquet_reader_options::builder(
    cudf::io::source_info{"/home/nghiat/Devel/tmp/ORC_DATA/tmp/708769/GPU_pq/input.parquet"});
  auto input = cudf::io::read_parquet(read_opts);

  cudf::io::chunked_orc_writer_options write_opts =
    cudf::io::chunked_orc_writer_options::builder(
      cudf::io::sink_info{"/home/nghiat/Devel/tmp/ORC_DATA/tmp/708769/GPU_pq/output.orc"})
      .compression(cudf::io::compression_type::SNAPPY);
  cudf::io::orc_chunked_writer(write_opts).write(*input.tbl);
}

Related: NVIDIA/spark-rapids#10806.