Tom Lane <tgl@sss.pgh.pa.us> writes:
> I do think hash aggregation is a plausible alternative implementation of
> plain DISTINCT, but I don't see the case for using it for DISTINCT ON.
It could be done without presorting the input though not with a simple
first()-like function. It would have be a sort of two-argument min() function
that kept a state variable for the smallest value found so far of the sort
key.
My main motivation here is that it's odd to have two code paths for
implementing the two language constructs when one is really just a special
case of the other. It's a source of cases like this where the code to
implement a query path exists but isn't accessible due to the way the query is
written.
--
greg