Error running crypto functions on Dictionary arrays such as md5

[alamb]

Describe the bug

A regression appears to have been introduced in

• Add support for Utf8View to crypto functions #13407

T

To Reproduce

This used to work in DataFusion 42.0.0 but doesn't in DataFusion 43.0.0:

DataFusion CLI v43.0.0
> create table t as values (arrow_cast('Foo', 'Dictionary(Int32, LargeUtf8)'));
0 row(s) fetched.
Elapsed 0.013 seconds.

> select md5(column1) from t;
Arrow error: Compute error: Internal Error: Cannot cast Utf8View to StringArray of expected type

Expected behavior

Query should complete

Additional context

I found this while writing tests in #13443 -- Feel free to take those tests to work on this issue

Those tests fail like this:

External error: query failed: DataFusion error: Arrow error: Invalid argument error: column types must match schema types, expected LargeUtf8 but found Utf8 at column index 0
[SQL] select md5(ascii_1) from test_basic_operator;
at test_files/string/./string_query.slt.part:1381
at test_files/string/large_string.slt:93

[Omega359]

I think this may have been a pre-existing issue. The core issue I think is that apparently md5 is required to return utf8:

datafusion/datafusion/functions/src/crypto/basic.rs

Line 190 in 6d8313e

// md5 requires special handling because of its unique utf8 return type

but the return_type function says it can return other types:

fn return_type(&self, arg_types: &[DataType]) -> Result<DataType> {
    use DataType::*;
    Ok(match &arg_types[0] {
        LargeUtf8 | LargeBinary => LargeUtf8,
        Utf8View | Utf8 | Binary => Utf8,
        Null => Null,
        Dictionary(_, t) => match **t {
            LargeUtf8 | LargeBinary => LargeUtf8,
            Utf8 | Binary => Utf8,
            Null => Null,
            _ => {
                return plan_err!(
                    "the md5 can only accept strings but got {:?}",
                    **t
                );
            }
        },
        other => {
            return plan_err!(
                "The md5 function can only accept strings. Got {other}"
            );
        }
    })
}

Should be an easy fix to the return_type fn.

[Omega359]

take

[jayzhan211]

I think adding casting from binary to string in TypeSignature::String and apply it to md5 is able to fix this too