PG Bug reporting form <noreply@postgresql.org> writes:
> While working with Chris Hajas on merging Postgres 12 with Greenplum
> Database we stumbled upon the following strange behavior in the geometry
> type polygon:
Yeah, I'd imagine that none of the geometric operators/functions take
any thought for NaNs; if they behave sanely it's purely accidental.
Not really sure what we'd do about that --- operators returning
boolean, for instance, don't have a great way to handle it. (I doubt
returning NULL would be nice.) Going forward it might be best
to disallow NaNs in geometric values, perhaps?
regards, tom lane