Steve Wampler <swampler@noao.edu> writes:
> Tom Lane wrote:
>> Even so, though, I think it would be quite foolish to design an
>> application around the assumption that the timestamps of successive
>> insertions will be distinguishable. Put in a serial column.
> I'll do that. I was a bit surprised to see that the sort wasn't
> stable, however. Was that intentional for performance, or just
> not considered worth the effort?
I think our sort code isn't stable --- it's a quicksort which usually
isn't. But in any case the physical tuple locations aren't guaranteed
to be strictly increasing, so changing the sort code wouldn't make it
safe.
regards, tom lane