Jon Barnett <jbarnett@pobox.com> writes:
> From a practicalities point of view, if the result is not compliant
> with the SQL standard, the question is whether to hide it in the
> abstraction layer (JDBC driver) or fix it at the source
If you can demonstrate that this behavior [NULL result from aggregate
functions] is not compliant with the SQL92 standard, then the Postgres
backend will get fixed.
There's been discussion of this point before on the hackers list,
with some people feeling that the current behavior is OK and others
not happy with it, but so far no one has made a convincing case about
what the standard expects.
regards, tom lane