On Mon, Aug 30, 2004 at 21:21:26 +0200,
Pierre-Frédéric Caillaud <lists(at)boutiquenumerique(dot)com> wrote:
> >>Also, count(*) is likely to always generate a seq scan due to the way
> >>aggregates are implemented currently in pgsql. you might want to try:
>
>
> By the way, in an ideal world, count(*) should only read the index
> on the timetamp column, not the rows. I guess this is not the case. Would
> this be an useful optimization ?
It's in the archives. The short answer is that no, postgres has to check
the heap to check tuple visibility to the current transaction.