Gracefully handling variable-length data types at the 8k limit?

I sent this to general:
> I have written an aggregate function which returns a "varlena" type whose
> length is proportional to the number of rows passed to the function.
>
> My question is what is the consensus on the best way to handle the case
> where the input to the aggregate causes the length of the variable to
> exceed the 8k limit? Abort the query? Generate an error message, and
> discard subsequent data? Silently discard data? Something else?

Apparently, not only is there no consensus, but nobody seems to have any
opinions either. So, I would like to solicit opinions.

What is the proper thing for aggregate functions and operators to do when
they discover they need more than 8k for a variable length data type?

-Michael Robinson