ametel01 commented on PR #23038:
URL: https://github.com/apache/datafusion/pull/23038#issuecomment-4835975340

   hanks @alamb , that makes sense.
   
   I agree the physical execution semantics are weak: with no input arrays, a 
generic accumulator does not have row cardinality, and disabling 
`convert_to_state` makes the feature much less useful for normal multi-stage 
aggregate plans.
   
   Given that, I think this PR may be the wrong abstraction for #16453.
   
   Your struct-returning aggregate suggestion seems like a better direction: 
model the operation as an aggregate that receives the relevant input columns, 
for example `augmented_avg(ts, value)`, and returns a struct containing window 
metadata plus the aggregate result. That keeps the aggregate connected to real 
input data and avoids adding nullary special cases across the aggregate 
execution paths.
   
   Unless you think there is still independent value in nullary aggregate UDF 
support, I can close this PR and instead explore/prototype the struct-returning 
aggregate approach for #16453.


-- 
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