Method ResolveArray

be/src/exec/parquet/parquet-metadata-utils.cc:912–947

There are three types of array encodings: 1. One-level encoding A bare repeated field. This is interpreted as a required array of required items. Example: repeated item; 2. Two-level encoding A group containing a single repeated field. This is interpreted as a array of required items ( is either optional or required). Example: grou

// encoding, determined by 'array_encoding'. '*node' is set to the repeated field for all
// three encodings (unless '*pos_field' or '*missing_field' are set to true).
Status ParquetSchemaResolver::ResolveArray(ArrayEncoding array_encoding,
    const SchemaPath& path, int idx, SchemaNode** node, bool* pos_field,
    bool* missing_field) const {
  if (array_encoding == ONE_LEVEL) {
    if (!(*node)->is_repeated()) {
      ErrorMsg msg(TErrorCode::PARQUET_UNRECOGNIZED_SCHEMA, filename_,
          PrintSubPath(tbl_desc_, path, idx), "array", (*node)->DebugString());
      return Status::Expected(msg);
    }
  } else {
    // In the multi-level case, we always expect the outer group to contain a single
    // repeated field
    if ((*node)->children.size() != 1 || !(*node)->children[0].is_repeated()) {
      ErrorMsg msg(TErrorCode::PARQUET_UNRECOGNIZED_SCHEMA, filename_,
          PrintSubPath(tbl_desc_, path, idx), "array", (*node)->DebugString());
      return Status::Expected(msg);
    }
    // Set *node to the repeated field
    *node = &(*node)->children[0];
  }
  DCHECK((*node)->is_repeated());

  if (idx + 1 < path.size()) {
    if (path[idx + 1] == SchemaPathConstants::ARRAY_POS) {
      // The next index in 'path' is the artifical position field.
      DCHECK_EQ(path.size(), idx + 2) << "position field cannot have children!";
      *pos_field = true;
      *node = NULL;
      return Status::OK();
    } else {
      // The next value in 'path' should be the item index
      DCHECK_EQ(path[idx + 1], SchemaPathConstants::ARRAY_ITEM);
    }
  }
  return Status::OK();
}

// According to the parquet spec, map columns are represented like:
// <map-repetition> group <name> (MAP) {