Re: New module proposal: tracker

[Martyn Russell <martyn lanedo com>]

On 18/08/09 13:42, Emmanuele Bassi wrote:
> On Tue, 2009-08-18 at 13:05 +0100, Martyn Russell wrote:
>
> > The response was positive. So I would like to propose Tracker as a new
> > GNOME module.
>
> though I've seen the amount of work that has been done on it, I admit
> I'm still ambivalent on Tracker, and the fact that nobody has come clear
> with the scope of the project yet.

That's fair.

Scope: To take over the world :)

I am not entirely sure what you want to hear regarding the scope, the 
website and Bugzilla description are quite clear to me:

"All-in-one indexer, search tool and metadata database."

and

"Tracker is a tool designed to extract information and metadata about 
your personal data so that it can be searched easily and quickly."

Perhaps this needs updating slightly, but ultimately it is about storing 
and querying metadata about your *personal* data as efficiently as 
possible. The scope can include web applications like facebook (we have 
a GSoC student with a patch working on this right now) for example.

It is also a freedesktop project so it is not just for GNOME, but also 
for KDE.

So depending on what you mean by scope, I hope I answered your question 
there as far as the "goal", "data" and "platforms" are concerned.

> we keep hearing stuff like "Tracker
> will host all metadata about everything, ever" but since I have
> objections to that I'd rather have somebody to give a balanced overview
> of what Tracker is and why applications should use it.

Well, first and foremost it depends on the build. It can be used for 
full text searching or just metadata searching (for example). Also 0.7. 
works to the Nepomuk[1] standard for ontologies. This means we not only 
have methods for gathering data (called miners, like the one that crawls 
the file system) but we also store the data with relationships which 
describe the data in terms of entities and hierarchy.

> the fact that Tracker hasn't ever gone through the external dependencies
> list is also what worries me: are we proposing a module that it's just
> not being used by anything in the platform?

Well, our initial thought was that it should probably be in the "extras" 
with GNOME given not many modules use it yet.

> > Right now Tracker 0.7 is currently in development and we are hoping to
> > get the 0.7. unstable release out the door in the coming month or so.
> >
> > Right now we are considering making the miners (the file system crawling
> > at this point) optional so it acts purely as a store if needed by ISVs.
> > This is not yet done in master but can be if that's a GNOME requirement.
>
> if Tracker is proposed as a metadata extractor, is there a library or an
> API that allows applications to access the metadata, or am I expected to
> use raw D-Bus calls?

There is tracker-sparql (a binary) and an API using libtracker (which is 
quite limited at this point, but does the job). You can also use DBus calls.

> if Tracker is proposed as a metadata storage then are there applications
> that are part of the platform that are using it?

The storage is only a part of it, the querying is also very important 
and has been designed for speed, for example: "List all images created 
on a specific month and order by date".

Currently we are working with the Zeitgeist team on an implementation 
that allows them to use Tracker as a backend. If used this means 
gnome-shell will ultimately depend on Tracker as I understand it.

Also nautilus depends on Tracker, but I think with 0.7 it might need 
updating, I would have to check.

[1] http://nepomuk.semanticdesktop.org/

--
Regards,
Martyn