dwsmith1983 commented on PR #5874:
URL: 
https://github.com/apache/datafusion-comet/pull/5874#issuecomment-5666324196

   > Could you add a short comment at the registration recording what it 
depends on? ... Would it be better to assert through 
`create_scalar_function_expr` so the test follows the path the planner actually 
takes?
   
   Both done. The registration comment records that `SparkLengthFunc` unwraps 
no dictionaries and gets no planner coercion, so it relies on `ScanExec`, the 
shuffle scan and the Parquet adapter delivering plain `Utf8` or `Binary`. The 
test now builds the `ScalarFunc` proto over a bound column and plans it through 
`PhysicalPlanner::create_expr`, asserting an `Int32` scalar function over the 
column with no cast for both string and binary; the wider and view encodings 
keep a return-type check only, since no Comet input path produces them. The 
branch is merged with `main` and the `jni_api.rs` conflict resolved.
   
   > Could you and @adibmbrk agree on which PR owns `length` and have the other 
drop that part?
   
   I agree this PR should own `length`, for the reason you give: a native 
kernel over routing through the dispatcher. @adibmbrk, would you drop the 
`length` part and the `length.sql` edits from #5607 and keep `bit_length` and 
`octet_length` there? I am happy to borrow `CometBinaryLengthBenchmark` for the 
numbers @sunchao asked for rather than add a third harness.
   


-- 
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]

Reply via email to