I wrote:
> Robert Haas <robertmhaas@gmail.com> writes:
>> My main concern was actually whether we ought to be detecting this
>> earlier in the process, before it gets as far as the executor.
> Yeah, that might be an appropriate response too. The executor is
> coded so cavalierly because it expects the planner to have replaced
> the CURRENT OF node with something executable. As things now stand,
> whether that happens or not depends in part on the behavior of FDWs,
> so maybe we'd better have the planner check whether it happened.
> I'm not sure though if there's any suitably-painless place to do it.
After looking at this a bit, my memory was faulty: the CurrentOfExpr
isn't really transformed to something else, we just make sure it ends up
in a TidScan node's tidqual list, where it won't be executed in the
normal way. The equivalent thing for a foreign table would be for the
FDW's execution code to have special smarts about what to do with a
CurrentOfExpr qual. So there's no way for the core planner to know
whether a ForeignScan incorporating a CurrentOfExpr qual is OK or not.
Short of a major redesign of the way this is handled, treating the
execution-time error as user-facing seems like the thing to do.
regards, tom lane