Robert Haas <robertmhaas@gmail.com> writes:
> Well, again, I think the proposal here is not get rid of 32-bit
> support, but to have less code that only gets regularly tested on
> 32-bit machines.
That seems like generally a good plan. But as to the specific idea...
> If we made datums 8 bytes everywhere, we would have
> less such code, and very likely fewer bugs.
... it's not entirely clear to me that it'd be possible to do this
without causing a storm of "cast from pointer to integer of different
size" (and vice versa) warnings on 32-bit machines. That would be a
deal-breaker independently of any performance considerations, IMO.
So if anyone wants to pursue this, finding a way around that might be
the first thing to look at.
regards, tom lane