Re: Properly mark NULL returns in numeric aggregates

David Rowley <dgrowleyml@gmail.com> writes:
> For testing, can't we just have an Assert() in
> advance_transition_function that verifies isnull matches the
> nullability of the return value for INTERNAL returning transfns? i.e,
> the attached

FTR, I do not like this Assert one bit.  nodeAgg.c has NO business
inquiring into the contents of internal-type Datums.  It has even
less business enforcing a particular Datum value for a SQL null ---
we have always, throughout the system, considered that if isnull
is true then the contents of the Datum are unspecified.  I think
this is much more likely to cause problems than solve any.

            regards, tom lane