Re: server side sorting & paging?

[Jon Roberts <man@mentata.com>]

First, let me clarify: I am not taking the side that this should be a 
priority for the OpenLDAP project. I agree that it makes more sense to 
sort on client side, have implemented it in the past, and am 
implementing now in my software. It works fine and doesn't kill any of 
my clients for result sets in the tens of thousands. In the servlet 
world (I have separated my core library from servlets) I am even 
planning to cache result sets so that various pages of sorted results 
can be retrieved without searching again. In my post was merely 
providing a client developer perspective on what is empirically viewed 
by some as a deficiency in OpenLDAP vs. proprietary LDAP servers.

Howard Chu wrote:
> > > > 1) Server Side Sorting
> It's certainly not a gut/rewrite/overhaul task.

I couldn't say, so please forgive the bit of cheekiness.

> > > > 2) Paging.
> > >
> > > 2.2 supports paged results.
> >
> > In my mind, paged results are of limited value
>
> In my mind, you should have stopped right there. 

Why? The point about correlating various syntaxes to appropriate sorts 
is a key implementation consideration that this thread would otherwise 
ignore. I recognize there is some value of paged results without 
sorting, which is why I chose the qualifier "limited".

> Clients should grab results
> and use them or display them as they arrive, in realtime. I suppose that's
> not great for HTTP clients, but HTML isn't fully interactive in the first
> place.

This point is misplaced. I personally handle results as you say, and 
while HTML is what the browser eventually gets, a Java servlet container 
"client" like I build on can involve some serious interactivity over the 
course of a session (I manage sessions now, too).

> OpenLDAP's indexing mechanism doesn't preserve ordering information,
> so sorting of indexed attributes doesn't come for free. The server
> uses hashes for indexing, to save on storage requirements.

This clears a lot up, but what about when the schema specifies an 
ORDERING for an attribute?

The only other thing I wanted to respond to is that the (sn=A*), 
(sn=B*). As I recall, we would need to manually patch OpenLDAP before 
building it in order to allow one or two letter filters. If that's still 
the case (Ace's results seem to indicate such), I would think you'd want 
to mention that gotcha anytime you sell that solution.

Thanks for responding. I'm still perfectly happy with things as they 
are, so please don't think I was griping.

Jon Roberts
www.mentata.com