Re: prefix btree implementation

On Thu, 6 Oct 2005, Jim C. Nasby wrote:

> On Wed, Oct 05, 2005 at 03:40:43PM -0700, Qingqing Zhou wrote:
> > We do the prefix sharing when we build up index only, never on the fly.
>
> So are you saying that inserts of new data wouldn't make any use of
> this? ISTM that greatly reduces the usefulness, though I'm not objecting
> because compression during build is probably better than none at all. Is
> there a technical reason compression can't be used during normal
> operations?
>

Yes, there are. Think if we do it we when build up index, we can choose
the shared prefix optimally w.r.t. maximizing the number of index items on
the page. But on the fly, if we do so, I am afraid this will (1) kill the
performance; (2) introduce more complexities. I don't exclude the
possibility of doing prefix sharing on the fly, but for current stage, I
would like to first come up with a proof-of-concept patch not including
this part.

Regards,
Qingqing