I wrote:
> "David G. Johnston" <david.g.johnston@gmail.com> writes:
>> How about just erroring out?
> Hm, yeah, inserting a FEATURE_NOT_SUPPORTED error might be an
> appropriate amount of effort.
When I looked into this more closely, it turns out that in v10/HEAD
it takes less code to fix it than to throw an error ;-). So I just
fixed it and added some regression tests. But 9.6 blows up somewhere
in the executor, and it didn't seem worth trying to deal with that.
So in 9.4-9.6 I just made the case throw an error.
regards, tom lane