Write "upgrade guide" for DataFusion 44.0.0

[andygrove]

Is your feature request related to a problem or challenge?

DataFusion 44.0.0 has breaking changes that require downstream projects to make code changes. Let's document these as we upgrade our own subprojects (Python, Comet, Ballista) to make life easier for other downstream projects.

Describe the solution you'd like

No response

Describe alternatives you've considered

No response

Additional context

No response

[andygrove]

I plan on adding notes to this issue as I encounter issues while upgrading Comet to use latest DF. First couple of issues:

• down_cast_any_ref was removed from the public API. Perhaps we should consider adding it back (and mark it as deprecated)?
• PhysicalExpr trait no longer has dyn_hash method and it is now necessary to implement DynEq. This may cause confusion unless we explain that deriving Hash, PartialEq, Eq for structs will automatically implement these traits.

[andygrove]

next issue:

60 |             signature: Signature::coercible(vec![DataType::Float64], Volatility::Immutable),
   |                                                  ^^^^^^^^^^^^^^^^^ expected `Arc<dyn LogicalType>`, found `DataType`

Solution was to change DataType::Float64 to Arc::new(NativeType::Float64)

[andygrove]

We can't just derive PartialEq for expressions that contain other expressions due to rust-lang/rust#78808:

error[E0507]: cannot move out of `other.child` which is behind a shared reference
  --> spark-expr/src/list.rs:49:5
   |
47 | #[derive(Debug, Hash, PartialEq, Eq)]
   |                       --------- in this derive macro expansion

We have to implement PartialEq manually instead.

[andygrove]

GroupValues and new_group_values were removed from the public API without being deprecated first. It does not impact Comet; I'm just pointing this out.

[andygrove]

public function find_df_window_func was removed in #13201. It would have been better to mark it deprecated and have it return None.

[andygrove]

error[E0308]: mismatched types
     |
1929 |         datafusion::physical_plan::windows::create_window_expr(
     |         ------------------------------------------------------ arguments to this function are incorrect
...
1934 |             sort_exprs,
     |             ^^^^^^^^^^ expected `&LexOrdering`, found `&[PhysicalSortExpr]`

fixed by changing sort_exprs to &LexOrdering::new(sort_exprs.to_vec()),

[andygrove]

I am now running into:

Error from DataFusion: return_type_from_exprs shoud be called instead.
This was likely caused by a bug in DataFusion's code and we would welcome that you file an bug report in our issue tracker.

[Omega359]

Error from DataFusion: return_type_from_exprs shoud be called instead.

The only place that has that typo is date_part which was changed in #13466 @jayzhan211

[andygrove]

ScalarUDFImpl.return_type is part of DataFusion's public API but returns an internal error if called, which doesn't seem correct. Should we deprecate this function and change the error type?

    fn return_type(&self, _arg_types: &[DataType]) -> Result<DataType> {
        internal_err!("return_type_from_exprs shoud be called instead")
    }

edit: it only returns this error in DatePartFunc

[andygrove]

I created a PR to deprecate return_type and revert the breaking change - #13717