Re: Confused about the release

[Martyn Russell <martyn lanedo com>]

On 31/08/11 22:26, Federico Mena Quintero wrote:
> On Tue, 2011-08-30 at 15:24 -0400, Shaun McCance wrote:

Hey Federico,

> > I noticed that we have Contacts, Documents, and Sushi in the
> > apps moduleset.
>
> gnome-documents worries me because of its dependency on Tracker.
>
> As far as I can tell, g-d is linked from
> https://live.gnome.org/ThreePointOne/Features/FindingAndReminding - to
> https://live.gnome.org/Design/Apps/Documents - which are just factual or
> planning pages, not discussion pages.  (FindingAndReminding is even
> marked "obsolete"! - what about the sublinks to Document-Centric Gnome?)
>
> With the inclusion of gnome-documents, we are validating Tracker as
> *the* metadata store for Gnome.
>
> Tracker has many implications, and I would like to see a good discussion
> of whether it is the right thing for Gnome.
>
> Some starting points:
>
> - How is an existing $HOME handled?

In terms of what? If you mean what data is indexed I can say that we 
index all files in $HOME (no sub-directories) and then we independently 
index all XDG directories thereafter recursively (e.g. $HOME/Documents 
or $HOME/Music, etc as it commonly is in distros).

> - What kind of maintenance needs to be done on the database?

Usually the maintenance done on the Tracker databases is done in the way 
of ontology updates. The ontology describes the schema which we spit out 
and use in SQLite. So any maintenance done on the database is usually in 
the form of: user asks for features x, y an z and we update our ontology 
(which updates the database).

We have gone to great lengths to ensure redundancy for user data and 
also to support various schema updates (e.g. changing a column from one 
type to another or adding new tables/properties). Most of this is 
handled automatically. There is more information on the wiki:

  https://live.gnome.org/Tracker/Documentation/SupportedOntologyChanges

> - Are we storing full-text indexes in the same database?

Yes. Stored as binary blobs. It used to be separate but SQLite is slower 
and doesn't guarantee ACID¹ with multiple DBs and WAL².

It also depends on how tracker is compiled. There is an option for it to 
be disabled. It's enabled by default.

¹ http://en.wikipedia.org/wiki/ACID
² http://www.sqlite.org/draft/wal.html

> - Are we storing non-metadata (e.g. contacts) in the database?

By "we", if you mean the GNOME community then no, however Nokia have 
extensively used it for the last year+. The ontology there is well 
tested at least but I don't think anyone is using that actively in the 
desktop yet. I could be wrong.

> - How do we handle metadata movement when files move around?

In the database one field changes, the filename (IIRC), the rest is 
relational and no new extraction or heavy database changes are required.

> - Is Tracker well-maintained now that Nokia is switching to something
> else?  (I'm not sure if this is the right question, but something along
> those lines.)

Good question. I think upstream will always get some attention, how much 
is hard to say.

Nokia is really providing a lot of bug fixing and testing which is why 
Tracker is quite a solid project right now. I and others on the team 
like Juergbi, Adrien, Aleksander, etc have been committing patches in 
GNOME bugzilla in their spare time also.

There are also quite some people submitting patches upstream - which we 
don't get from Nokia.

Intel are also using Tracker in MeeGo so it's in use in more than one 
place and I believe they're patching it too (internally and upstream).

--
Regards,
Martyn

Founder and CEO of Lanedo GmbH.