How to downcast dyn Array using as_any and data_type?

[twitu]

I'm reading a chunk from a file, then iterating the column array but I need to convert the array to a more specific array type. I've looked at the documentation but haven't found a way to do it.

    let f = File::open("array_of_arrays.parquet").unwrap();
    let fr = FileReader::try_new(&f, None, None, None, None).unwrap();

    for chunk in fr.into_iter() {
        if let Ok(cols) = chunk {
            for array in cols.arrays().iter() {
                 // QUESTION: array: &Arc<dyn Array>
                 // QUESTION: cast array to some array type say ListArray
            }
        }
    }

Documentation for data_type mentions using as_any and data_type to this. But it's not very clear how to use this to downcast the array.

Any help is deeply appreciated.

[jorgecarleitao]

Hey! So, because Array is a trait object, we need to match the datatype as you mention. We tried to describe this in the guide. Something like this:

use arrow2::datatypes::{PhysicalType, PrimitiveType};

match array.data_type().to_physical_type() {
     PhysicalType::Primitive(PrimitiveType::Float32) => {
            let array = array.as_any().downcast_ref::<PrimitiveArray<f32>>().unwrap(); // guaranteed
            // let array = f32-specific operator
            let array = array.clone();
            Ok(Box::new(array))
     }
     PhysicalType::Primitive(PrimitiveType::Float64) => {
            let array = array.as_any().downcast_ref::<PrimitiveArray<f64>>().unwrap(); // guaranteed
            // let array = f64-specific operator
            let array = array.clone();
            Ok(Box::new(array))
     }
     PhysicalType::List => {
            let array = array.as_any().downcast_ref::<ListArray<i32>>().unwrap(); // guaranteed
            // let array = list-specific operator
            let array = array.clone();
            Ok(Box::new(array))
     }
}

in other words, data_type is the logical type, to_physical_type converts it to its corresponding physical type, which we match to get the arrays' physical type.

Hope this helps.

[twitu]

Thanks for sharing this. I saw it in the documentation but it didn't click until now. Using this I've been able to successfully de-serialize parquet files back to structs.