make array_union/array_except/array_intersect handle empty/null arrays rightly

[Veeupup]

Which issue does this PR close?

Closes#8181.

for array_union:

array_union([], []) = []
array_union([], null) = []
array_union(null, []) = []
array_union(null, null) = null

for array_except:

array_except([], []) = []
array_except([], null) = []
array_except(null, []) = null
array_except(null, null) = null

for array_intersect:

array_intersect([], []) = []
array_intersect([], null) = []
array_intersect(null, []) = []
array_intersect(null, null) = null

Rationale for this change

What changes are included in this PR?

Are these changes tested?

Are there any user-facing changes?

[Veeupup]

@jayzhan211 maybe you can check the behavior for these functions now cc @alamb

[jayzhan211]

I think we need to fix here.

https://github.com/apache/arrow-datafusion/blob/393e48f98872c696a90fce033fa584533d2326fa/datafusion/sql/src/expr/value.rs#L156-L158

if data_types.is_empty(){// ListArray[NullArray(0)], the same as `make_array()`.let array = new_null_array(&DataType::Null,0);let array = Arc::new(array_into_list_array(array));Ok(lit(ScalarValue::List(array)))}elseif data_types.len() > 1{

[jayzhan211]

I think we need to fix here.

https://github.com/apache/arrow-datafusion/blob/393e48f98872c696a90fce033fa584533d2326fa/datafusion/sql/src/expr/value.rs#L156-L158

if data_types.is_empty(){// ListArray[NullArray(0)], the same as `make_array()`.let array = new_null_array(&DataType::Null,0);let array = Arc::new(array_into_list_array(array));Ok(lit(ScalarValue::List(array)))}elseif data_types.len() > 1{

Update: I found we can easily reuse MakeArray for sql_array_literal

Ok(Expr::ScalarFunction(ScalarFunction::new(BuiltinScalarFunction::MakeArray,
values,)))

[Veeupup]

@jayzhan211 Thanks for your constructive comments！I suppose we make it in a right way : )

[jayzhan211]

We can move select_array_no_common_type and select_array_non_literal_type to slt

[alamb]

Thank you very much @Veeupup and @jayzhan211 - this looks really nice and exactly correct 🙏

[alamb]

yes!

[alamb]

👍

[jayzhan211]

@Veeupup Why do we need Null checking here, which function failed without this?

[Veeupup]

when it comes to handle multi arrays functions,their input types maybe like null, List(Int32).... Then, but we should handle null type in function body rather than return error here.