Peter Eisentraut <peter@eisentraut.org> writes:
> I think this is actually a bug in the implementation, not in the
> documentation. That is, the implementation should behave like the
> documentation suggests.
The trouble with that is that it presumes that the standard's
definition of IS NOT NULL is not broken. I think it *is* broken
for rowtypes; it certainly cannot be claimed to be intuitive.
We already have disclaimers about that in our documentation
about IS [NOT] NULL. I don't really want to propagate similar
confusion into COALESCE, much less everyplace else that this'd
matter.
Having said that, I'm not sure that substituting "is distinct from
null" in the COALESCE documentation is much better, because it's not
clear to me that we're entirely standards-compliant about what that
means for rowtypes either.
regards, tom lane