Alexander Korotkov <aekorotkov@gmail.com> writes:
> Do you think we don't need int64 GUCs just now, when 64-bit
> transaction ids are far from committable shape? Or do you think we
> don't need int64 GUCs even if we have 64-bit transaction ids? If yes,
> what do you think we should use for *_age variables with 64-bit
> transaction ids?
I seriously doubt that _age values exceeding INT32_MAX would be
useful, even in the still-extremely-doubtful situation that we
get to true 64-bit XIDs. But if you think we must have that,
we could still use float8 GUCs for them. float8 is exact up
to 2^53 (given IEEE math), and you certainly aren't going to
convince me that anyone needs _age values exceeding that.
For that matter, an imprecise representation of such an age
limit would still be all right wouldn't it?
regards, tom lane
