Representation for DISTINCT in aggregate functions

[JamesRTaylor]

What's the best way in the spec to represent DISTINCT usage in aggregate functions (i.e. COUNT(DISTINCT foo) )? There seem to be two options: specialized functions or an aggregate function invocation property. Any other ideas? Thoughts on which would be best?

[jacques-n]

Tagging @westonpace , @cpcloud, @andygrove, @hannes , @pitrou for feedback.

[andygrove]

In DataFusion we made it a property of invoking an aggregate function.

[cpcloud]

In principle any column whose values can be compared using == can be DISTINCT input to an aggregation, so I'm +1 on having it as a property of aggregate functions instead of having separate {agg}_distinct functions.

[westonpace]

I'm -1 on {agg}_distinct. I don't understand the alternative enough to say. When we say a property of invoking the aggregate function are we really talking about a property on the AggregateFunction message? Or is this part of the AggregateRel::Measure messages?

[JamesRTaylor]

I was thinking it'd be a property on AggregateFunction message.

[JamesRTaylor]

For example, like this PR: #191

[hannes]

We have this as a flag is_distinct in the BoundAggregateExpression https://github.com/duckdb/duckdb/blob/master/src/include/duckdb/planner/expression/bound_aggregate_expression.hpp#L28

[jacques-n]

Note, this is now addressed via #191 and will be in the next release.