kumarUjjawal commented on code in PR #20093:
URL: https://github.com/apache/datafusion/pull/20093#discussion_r2749669561
##########
datafusion/functions/src/math/nans.rs:
##########
@@ -84,36 +88,123 @@ impl ScalarUDFImpl for IsNanFunc {
}
fn invoke_with_args(&self, args: ScalarFunctionArgs) ->
Result<ColumnarValue> {
- // Handle NULL input
- if args.args[0].data_type().is_null() {
- return Ok(ColumnarValue::Scalar(ScalarValue::Boolean(None)));
- }
+ let [arg] = take_function_args(self.name(), args.args)?;
+
+ match arg {
+ ColumnarValue::Scalar(scalar) => {
+ if scalar.is_null() {
+ return
Ok(ColumnarValue::Scalar(ScalarValue::Boolean(None)));
+ }
+
+ let result = match scalar {
+ ScalarValue::Float64(Some(v)) => Some(v.is_nan()),
+ ScalarValue::Float32(Some(v)) => Some(v.is_nan()),
+ ScalarValue::Float16(Some(v)) => Some(v.is_nan()),
- let args = ColumnarValue::values_to_arrays(&args.args)?;
-
- let arr: ArrayRef = match args[0].data_type() {
- DataType::Float64 => Arc::new(BooleanArray::from_unary(
- args[0].as_primitive::<Float64Type>(),
- f64::is_nan,
- )) as ArrayRef,
-
- DataType::Float32 => Arc::new(BooleanArray::from_unary(
- args[0].as_primitive::<Float32Type>(),
- f32::is_nan,
- )) as ArrayRef,
-
- DataType::Float16 => Arc::new(BooleanArray::from_unary(
- args[0].as_primitive::<Float16Type>(),
- |x| x.is_nan(),
- )) as ArrayRef,
- other => {
- return exec_err!(
- "Unsupported data type {other:?} for function {}",
- self.name()
- );
+ // Non-float numeric inputs are never NaN
+ ScalarValue::Int8(_)
+ | ScalarValue::Int16(_)
+ | ScalarValue::Int32(_)
+ | ScalarValue::Int64(_)
+ | ScalarValue::UInt8(_)
+ | ScalarValue::UInt16(_)
+ | ScalarValue::UInt32(_)
+ | ScalarValue::UInt64(_)
+ | ScalarValue::Decimal32(_, _, _)
+ | ScalarValue::Decimal64(_, _, _)
+ | ScalarValue::Decimal128(_, _, _)
+ | ScalarValue::Decimal256(_, _, _) => Some(false),
+
+ other => {
+ return exec_err!(
+ "Unsupported data type {other:?} for function {}",
+ self.name()
+ );
+ }
+ };
+
+ Ok(ColumnarValue::Scalar(ScalarValue::Boolean(result)))
}
- };
- Ok(ColumnarValue::Array(arr))
+ ColumnarValue::Array(array) => {
+ // NOTE: BooleanArray::from_unary preserves nulls.
+ let arr: ArrayRef = match array.data_type() {
+ Null => Arc::new(BooleanArray::new_null(array.len())) as
ArrayRef,
+
+ Float64 => Arc::new(BooleanArray::from_unary(
+ array.as_primitive::<Float64Type>(),
+ f64::is_nan,
+ )) as ArrayRef,
+ Float32 => Arc::new(BooleanArray::from_unary(
+ array.as_primitive::<Float32Type>(),
+ f32::is_nan,
+ )) as ArrayRef,
+ Float16 => Arc::new(BooleanArray::from_unary(
+ array.as_primitive::<Float16Type>(),
+ |x| x.is_nan(),
+ )) as ArrayRef,
+
+ // Non-float numeric arrays are never NaN
+ Decimal32(_, _) => Arc::new(BooleanArray::from_unary(
Review Comment:
Scalar NULL would get broadcast to all rows, so for a Decimal* array it
would incorrectly return NULL even for non-null inputs. For decimal arrays we
need false for non-null rows and NULL for null rows, so returning an array
which preserves nulls) is required; scalar NULL is only correct for
DataType::Null inputs ?
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
