clintropolis commented on a change in pull request #10613:
URL: https://github.com/apache/druid/pull/10613#discussion_r584199144
##########
File path:
processing/src/main/java/org/apache/druid/query/dimension/DimensionSpec.java
##########
@@ -70,6 +71,11 @@ default MultiValueDimensionVectorSelector
decorate(MultiValueDimensionVectorSele
throw new UOE("DimensionSpec[%s] cannot vectorize", getClass().getName());
}
+ default VectorObjectSelector decorate(VectorObjectSelector selector)
Review comment:
I haven't yet implemented support for multi-value string object
selectors, but your question also seems applicable if we were to add array
types too. I guess the contents of the array would be dependent on whatever the
underlying selector is providing, and it would be the responsibility of the
`DimensionSpec` implementation to handle decoration appropriately? I guess we
should also probably consider adding methods for the primitive numeric columns
as well, since they are not masqueraded as "dimension selectors" in the
vectorized engine, (if we ever want to fully support all of the non-vectorized
behavior here I think).
I don't know, the more I think about this, do we really need
`DimensionSpec`, like at all? The stuff it does is all over the place, and it
feels like maybe this type of stuff should just be pushed down into either
expression virtual columns or something more specialized as appropriate (more
generic predicate filtered selectors virtual column?). Then we maybe we can
just ditch this stuff entirely in some sunny future instead of implementing
vectorized versions of decorate for the various `DimensionSpec`
implementations. Or is there an advantage to `DimensionSpec` that I am missing?
----------------------------------------------------------------
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.
For queries about this service, please contact Infrastructure at:
[email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
