How to project Schema or SchemaDescriptor by ProjectionMask?

[RinChanNOWWW]

Suppose we have a parquet file whose schema is (i Int, s Struct(a Int, b Struct(c int, d int))).

We can construct a ProjectionMask { vec![true, false, false, true] } for select i, s.b.d from t, and the Schema of the output RecordBatch will be (i Int, s Struct(b Struct(d int))).

My question is, how can I precompute the output Schema by the original Schema and the ProjectionMask?

I only found

impl Schema {
    pub fn project(&self, indices: &[usize]) -> Result<Schema, ArrowError> {
      let new_fields = indices
          .iter()
          .map(|i| {
              self.fields.get(*i).cloned().ok_or_else(|| {
                  ArrowError::SchemaError(format!(
                      "project index {} out of bounds, max field {}",
                      i,
                      self.fields().len()
                  ))
              })
          })
          .collect::<Result<Vec<_>, _>>()?;
      Ok(Self::new_with_metadata(new_fields, self.metadata.clone()))
  }
}

which only can project root fields.

[tustvold]

I don't believe we support this directly on arrow Schema yet, but there is https://docs.rs/parquet/latest/parquet/arrow/fn.parquet_to_arrow_schema_by_columns.html

[RinChanNOWWW]

LGTM, thank you!!