Fix compilation, change row_number() expr_fn to 0 args

[alamb]

Which issue does this PR close?

N/A

Rationale for this change

Fix logical conflict between #12030 and #12000

What changes are included in this PR?

1. Update include
2. fix signature of row_number expr_fn to not take arguments

Are these changes tested?

Yes

Are there any user-facing changes?

[alamb]

Note row_number() does not take an argument -- it was erroniously changed here #12030 and it happens we didn't have other tests for it

> select row_number() OVER ();
+-----------------------------------------------------------------------+
| ROW_NUMBER() ROWS BETWEEN UNBOUNDED PRECEDING AND UNBOUNDED FOLLOWING |
+-----------------------------------------------------------------------+
| 1                                                                     |
+-----------------------------------------------------------------------+
1 row(s) fetched.
Elapsed 0.042 seconds.

[jayzhan211]

Given there is signature Any(0) but the test in round_trip still works.

I guess we should add data_types_with_window_udf somewhere

[alamb]

I guess we should add data_types_with_window_udf somewhere

I don't understand this comment 🤔

[alamb]

Thank you for the quick review @jayzhan211 -- I am going to merge this in to get the CI green again. I am happy to make additional changes as a follow on PR if other reviewers have suggestions

[jayzhan211]

If we did the signature check, this test should fail. Because the signature is Any(0), which should only accept zero arguments

The reason why it is not is because we didn't do signature check for window udf.

We have similar functions for scalar and udaf.

datafusion/datafusion/expr/src/type_coercion/functions.rs

Lines 42 to 96 in cb1e3f0

	pub fn data_types_with_scalar_udf(
	current_types: &[DataType],
	func: &ScalarUDF,
	) -> Result<Vec<DataType>> {
	let signature = func.signature();
	if current_types.is_empty() {
	if signature.type_signature.supports_zero_argument() {
	return Ok(vec![]);
	} else {
	return plan_err!("{} does not support zero arguments.", func.name());
	}
	}
	let valid_types =
	get_valid_types_with_scalar_udf(&signature.type_signature, current_types, func)?;
	if valid_types
	.iter()
	.any(|data_type| data_type == current_types)
	{
	return Ok(current_types.to_vec());
	}
	try_coerce_types(valid_types, current_types, &signature.type_signature)
	}
	pub fn data_types_with_aggregate_udf(
	current_types: &[DataType],
	func: &AggregateUDF,
	) -> Result<Vec<DataType>> {
	let signature = func.signature();
	if current_types.is_empty() {
	if signature.type_signature.supports_zero_argument() {
	return Ok(vec![]);
	} else {
	return plan_err!("{} does not support zero arguments.", func.name());
	}
	}
	let valid_types = get_valid_types_with_aggregate_udf(
	&signature.type_signature,
	current_types,
	func,
	)?;
	if valid_types
	.iter()
	.any(|data_type| data_type == current_types)
	{
	return Ok(current_types.to_vec());
	}
	try_coerce_types(valid_types, current_types, &signature.type_signature)
	}

data_types_with_scalar_udf and data_types_with_aggregate_udf.

After adding a new data_types_with_window_udf function, we should get the failed msg for incorrect non-empty args for row_number

Maybe add it somewhere here

datafusion/datafusion/expr/src/expr_schema.rs

Lines 169 to 196 in cb1e3f0

	Expr::WindowFunction(WindowFunction { fun, args, .. }) => {
	let data_types = args
	.iter()
	.map(|e| e.get_type(schema))
	.collect::<Result<Vec<_>>>()?;
	let nullability = args
	.iter()
	.map(|e| e.nullable(schema))
	.collect::<Result<Vec<_>>>()?;
	match fun {
	WindowFunctionDefinition::AggregateUDF(udf) => {
	let new_types = data_types_with_aggregate_udf(&data_types, udf)
	.map_err(|err| {
	plan_datafusion_err!(
	"{} {}",
	err,
	utils::generate_signature_error_msg(
	fun.name(),
	fun.signature().clone(),
	&data_types
	)
	)
	})?;
	Ok(fun.return_type(&new_types, &nullability)?)
	}
	_ => fun.return_type(&data_types, &nullability),
	}
	}

[alamb]

I believe this was done in #12045