David Rowley <dgrowleyml@gmail.com> writes:
> I don't think that's really going to help. The crash already tells us
> there's a problem down the line, but if the commit you mention is to
> blame for this, then the problem is elsewhere, either in our
> assumption that we can get away without the datumCopy() or in the
> aggregate function producing the state that we're no longer copying.
It does smell like the aggregate output has been corrupted by the time
it got to the plpgsql function. I don't particularly want to try to
synthesize a test case from the essentially-zero SQL-level information
we've been provided, though. And I doubt we can track this down without
a test case. So please try to sanitize the case you have enough that
you can share it.
regards, tom lane