[FEA] GroupBy on an arbitrary nested array(with struct/array inside)

[viadea]

I wish we can support GroupBy on an arbitrary nested array(with struct/array inside).
For example, the column could be very complex like inside the struct, there is array, struct nested as well.

The sample query:

SELECT a_nested_struct_col FROM table
GROUP BY request.a_nested_struct_col  limit 3

The not-supported-message could be like:

!Expression <KnownFloatingPointNormalized> knownfloatingpointnormalized(if (... (StructType(StructField(a,DoubleType,true), StructField(b,StringType,true),  StructField(c,LongType,true), StructField(d,ArrayType(LongType,true),true)) is not supported); expression KnownFloatingPointNormalized knownfloatingpointnormalized(...

The real use case has a very complex array column which has nested struct/array inside.
If you need the real example, please ping me offline.

[ttnghia]

A recently merged cudf PR enables GroupBy with array:

• Support nested types as groupby keys in libcudf rapidsai/cudf#11792

However, supporting arbitrary nested array needs to wait for:

• Proposal for lexicographic comparators for lists of structs rapidsai/cudf#11222
• [FEA] Enable structs of lists in the new row lexicographic comparator rapidsai/cudf#11672

While waiting for thes issues, if necessary then we can support non-nested array GroupBy first.