Function ComputeColumnChunkRange

cpp/src/parquet/file_reader.cc:176–213

Compute the section of the file that should be read for the given row group and column chunk.

/// Compute the section of the file that should be read for the given
/// row group and column chunk.
::arrow::io::ReadRange ComputeColumnChunkRange(FileMetaData* file_metadata,
                                               int64_t source_size, int row_group_index,
                                               int column_index) {
  std::unique_ptr<RowGroupMetaData> row_group_metadata =
      file_metadata->RowGroup(row_group_index);
  std::unique_ptr<ColumnChunkMetaData> column_metadata =
      row_group_metadata->ColumnChunk(column_index);

  int64_t col_start = column_metadata->data_page_offset();
  if (column_metadata->has_dictionary_page() &&
      column_metadata->dictionary_page_offset() > 0 &&
      col_start > column_metadata->dictionary_page_offset()) {
    col_start = column_metadata->dictionary_page_offset();
  }

  int64_t col_length = column_metadata->total_compressed_size();
  int64_t col_end;
  if (col_start < 0 || col_length < 0) {
    throw ParquetException("Invalid column metadata (corrupt file?)");
  }

  if (AddWithOverflow(col_start, col_length, &col_end) || col_end > source_size) {
    throw ParquetException("Invalid column metadata (corrupt file?)");
  }

  // PARQUET-816 workaround for old files created by older parquet-mr
  const ApplicationVersion& version = file_metadata->writer_version();
  if (version.VersionLt(ApplicationVersion::PARQUET_816_FIXED_VERSION())) {
    // The Parquet MR writer had a bug in 1.2.8 and below where it didn't include the
    // dictionary page header size in total_compressed_size and total_uncompressed_size
    // (see IMPALA-694). We add padding to compensate.
    int64_t bytes_remaining = source_size - col_end;
    int64_t padding = std::min<int64_t>(kMaxDictHeaderSize, bytes_remaining);
    col_length += padding;
  }

  return {col_start, col_length};
}

}  // namespace