[postgis-devel] EMPTY is as EMPTY does?

[Sandro Santilli]

On Wed, Jun 24, 2015 at 12:59:19PM -0700, Paul Ramsey wrote:
> Yes, it should be safe at an app level. I ran into the case though, because we had a query that was bringing down one server, but couldn’t be replicated on others… why? because when we tried to move the data to a test server it had to go through dump/restore which effectively removed the case (MULTILINESTRING(EMPTY,EMPTY)) that was causing the crash. So it’s a problem of correctness in that we cannot replicate a database exactly using dump/restore. Of course, we already have that in one case (POINT EMPTY) but maybe we don’t want to deliberately have it for other cases too.

Agreed, it's best to always keep the same structure of emptiness.
And have the homogenizer function collapse to a single empty
(I still think we need an ST_Homogenize that doesn't necessarely want
a "COLLECTION" as input).

--strk;