Jefffrey commented on code in PR #9959:
URL: https://github.com/apache/arrow-rs/pull/9959#discussion_r3295741297
##########
arrow-array/src/array/run_array.rs:
##########
@@ -781,6 +789,37 @@ where
RunArrayIter::new(self)
}
}
+/// An array that can be downcast to a [`RunArray`] of any run end type and
any value type.
+///
+/// This can be used to efficiently implement kernels for all possible run end
+/// types without needing to create specialized implementations for each key
type.
+pub trait AnyRunEndArray: Array {
+ /// Returns the run ends of this array as a primitive array.
+ fn run_ends(&self) -> ArrayRef;
+
+ /// Returns the values of this array.
+ fn values(&self) -> &Arc<dyn Array>;
+
+ /// Returns a new run-end encoded array with the given values, preserving
the
+ /// existing run ends.
+ fn with_values(&self, values: ArrayRef) -> ArrayRef;
+}
Review Comment:
> looking at the example you provided, shouldn't `let array_sliced =
array.slice(3, 3);` ideally be represented as `run_ends:[3],values:[2]` instead
of
`{run_ends: [3, 6, 9], values: PrimitiveArray<Int32>
[
1,
2,
3,
]}`
which is currently what `array.slice()` produces. due to `REE.slice()`
copying the values arrays.
Technically yes, but its just a result of how slicing is currently
implemented on run arrays (can be a little confusing)
> I think the best approach for now would be to remove the `run_ends()`
function and add it in later if there is a need.
Im not sure how we would avoid the extra work for `with_values()` without
updating the `.slice()` method. But this would require looking at all the
CallSites for slice first to make sure this doesn't cause breaking changes. In
theory this shouldn't causes regressions though it should be an implementation
detail.
This sounds good, thanks
--
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]
