Make RowType fields available in SPI?

[zhenxiao]

Currently, RowType is in presto-main, not in presto-spi
To better handle nested fields in Parquet(or ORC), need to get fields for a RowType, and RowField. Is it possible to put RowType in presto-spi? Or, is there other ways to make Parquet or ORC aware of RowType and RowFields?
@martint @dain @cberner @electrum @erichwang

[cberner]

Could you provide more details about why it's needed? The current implementation doesn't need it, and it can support reading and writing them

[zhenxiao]

When we are trying to implement Nested Schema Evolution, and Nested Column Pruning in #5547, we need to get RowFields of a RowType, and use each RowField name to look up the corresponding parquet type in parquet file schema. There is no RowType information in SPI, then Parquet files could not get RowFields.

[cberner]

You should be able to get that information by calling getTypeSignature().getParameters() which is on Type in the SPI

[zhenxiao]

Thank you, @cberner Correct me if I get misunderstanding. if I have a RowType struct:
RowType <Integer ColumnA, Double ColumnB>

My understanding is getTypeSignature().getParameters() will return <Integer, Double>, will it return ColumnA and ColumnB?

[cberner]

yes, it should return a list of the fields which can be accessed as NamedTypeSignature I believe

[zhenxiao]

oh, get it. Thank you

[cberner]

Np. Please re-open if you end up needing this