I wrote:
> The more I think about this, the more I like the other idea of just
> throwing an error rather than trying to fix up cases like bug #19055.
> I don't think we have much evidence that anyone is trying to do that
> in the real world (otherwise reports would have surfaced years ago).
> And this discussion is making it clear that fixing it up is harder
> than it sounds.
Hearing no further comments, I've pushed the v2 patch that does it
that way, restoring the previous behavior in cases that would not
have failed before #19055.
regards, tom lane