How to support a row-type function parameter with an indeterminate number of columns？

[duanmeng]

I am trying to add a "to_json" function in pr #6287, which supports 3 input types, ARRAY, MAP, ROW. It seems to only support a fixed number of columns, such as .argumentType("row(bigint, varchar)"). But we cannot determine how many columns there are especially the type of each column at definition time. I cannot find a way to support this scenario although I can use .argumentType("T").

[mbasmanova]

CC: @kagamiori

This is not supported yet. Curious, why to_json function supports only ARRAY, MAP and ROW. I'd expect it to support all types.

[duanmeng]

Spark only support array, map and struct (aka row in velox), it checks the input types, and Velox's JsonCastOperator checks and only support array, map and row as well.

SparkSQL
https://github.com/apache/spark/blob/master/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/expressions/jsonExpressions.scala#L733

  override def checkInputDataTypes(): TypeCheckResult = inputSchema match {
    case struct: StructType =>
      ...
    case map: MapType =>
      ...
    case array: ArrayType =>
      ...
    case _ => TypeCheckResult.TypeCheckFailure

Velox
https://github.com/facebookincubator/velox/blob/main/velox/functions/prestosql/types/JsonType.cpp#L788

bool JsonCastOperator::isSupportedToType(const TypePtr& other) const {
  ...
  switch (other->kind()) {
    case TypeKind::ARRAY:
      return isSupportedToType(other->childAt(0));
    case TypeKind::ROW:
      ...
      return true;
    case TypeKind::MAP:
     ...
    default:
      return false;