tustvold commented on code in PR #2649:
URL: https://github.com/apache/arrow-rs/pull/2649#discussion_r963429788
##########
arrow/src/compute/kernels/cast.rs:
##########
@@ -1456,21 +1621,62 @@ where
}
/// Convert Array into a PrimitiveArray of type, and apply numeric cast
-fn cast_numeric_arrays<FROM, TO>(from: &ArrayRef) -> Result<ArrayRef>
+fn cast_numeric_arrays<FROM, TO>(
+ from: &ArrayRef,
+ cast_options: &CastOptions,
+) -> Result<ArrayRef>
where
FROM: ArrowNumericType,
TO: ArrowNumericType,
FROM::Native: num::NumCast,
TO::Native: num::NumCast,
{
- Ok(Arc::new(numeric_cast::<FROM, TO>(
- from.as_any()
- .downcast_ref::<PrimitiveArray<FROM>>()
- .unwrap(),
- )))
+ if cast_options.safe {
+ // If the value can't be casted to the `TO::Native`, return null
+ Ok(Arc::new(numeric_cast::<FROM, TO>(
+ from.as_any()
+ .downcast_ref::<PrimitiveArray<FROM>>()
+ .unwrap(),
+ )))
+ } else {
+ // If the value can't be casted to the `TO::Native`, return error
+ Ok(Arc::new(numeric_cast_with_error::<FROM, TO>(
+ from.as_any()
+ .downcast_ref::<PrimitiveArray<FROM>>()
+ .unwrap(),
+ )?))
+ }
}
-/// Natural cast between numeric types
+// Natural cast between numeric types
+// If the value of T can't be casted to R, will throw error
+fn numeric_cast_with_error<T, R>(from: &PrimitiveArray<T>) ->
Result<PrimitiveArray<R>>
+where
+ T: ArrowNumericType,
+ R: ArrowNumericType,
+ T::Native: num::NumCast,
+ R::Native: num::NumCast,
+{
+ let iter = from
+ .iter()
+ .map(|v| match v {
+ None => Ok(None),
+ Some(value) => match num::cast::cast::<T::Native,
R::Native>(value) {
+ None => Err(ArrowError::CastError(format!(
+ "Can't cast value {:?} to type {}",
+ value,
+ R::DATA_TYPE
+ ))),
+ Some(v) => Ok(Some(v)),
+ },
+ })
+ .collect::<Result<Vec<Option<R::Native>>>>()?;
+
+ Ok(unsafe { PrimitiveArray::<R>::from_trusted_len_iter(iter) })
Review Comment:
I don't think the code above is correct, in particular you are taking an
iterator of Option and collecting into a Buffer. I would not expect it to even
compile...
I had thought PrimitiveArray had a try_from_trysted_len_iter but it does
not, so ignore me. The real performance will likely be from not using iterators
at all, but one step at a time and I don't imagine the cast kernels are a major
bottleneck atm
--
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]
