Re: Followup: opinions on Search services

[Jamie McCracken <jamiemcc blueyonder co uk>]

Manuel Amador wrote:

> > 3) It really needs to be a freedesktop solution written in C cause it 
> > looks likely that KDE will produce their own solution so adding to a lot 
> > of needless wheel reinvention.
>
>
> The KDE solution has a different focus.  I've been talking to the klink
> people regularly, and they have placed their faith in the value of
> relationships instead of the value of data.  I hope that turns out great
> as well =).

The optimal solution IMO is to have a 100% C daemon that uses libgda for 
DB access (IE provides abstraction to sqlite, mysql, firebird, postgres 
etc). KDE is more interested in conceptual linking of data and thats 
really easy to do with an sql DB  cause the data is highly relational in 
nature (and its one area that flat file databases like lucerne will 
*probably* struggle - I've not used lucerne so thats a guess but I 
haven't read anywhere that lucerne is a fully relational database).

The big problem with daemons written in garbage collected languages is 
resource usage and the inability of ordinary users to turn off an 
invisible background service (its not a big deal if a gui app is eating 
too much memory cause the user can close it down). Im not saying dont 
use them for this but we need to be cautious about their impact - they 
are a *potential* threat to system stability particularly on systems 
with lower amounts of free RAM.

Supporting embeddable DBs (sqlite, firebird) is very important as it 
means they will just work out of the box with no configuration. KDE's 
Tenor plans on using Postgres which looks foolish as setting up postgres 
is a technical task beyond most users and so looks certain to fail.

Its desirable to make use of things like Dbus rather than plain sockets 
as it makes interfacing and integrating stuff a lot easier.

THe text searching can be done with sql "like" or optionally with 
"glob". We dont need google style ranking or show me similiar pages etc 
so the sql DB route looks very feasible.

We also want to avoid significant dependencies (like a big RDBMS, 
additional runtimes, libraries and frameworks etc) so that adoption is 
much easier.

I believe a freedesktop implementation is possible and desirable as I 
cant believe anyone will be happy running two separate indexers if they 
use a mixed kde/gnome desktop. If anyone is interested in doing this 
please let me know as Im considering implementing it myself at some 
point (when I have the time!).

jamie.