Re: [Tracker] The Utopian idea, Tracker as it should be

[Philip Van Hoof <philip codeminded be>]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

On 22/09/2014 12:30, Martyn Russell wrote:
> On 21/09/14 17:05, Vishesh Handa wrote:

> > We, in KDE, were quite fed up with the ontologies. With Baloo,
> > we're no longer using any ontologies. The project simple aims to
> > be a good search
>
> The tracker-miner-fs is already really good at this and the black 
> listing rules are quite comprehensive. I plan to make things a
> little more user-friendly here over the coming year or so too.
>
> > index for files. Baloo does not store any data and is just an
> > index. The

> In terms of indexing content, we do this mainly to provide
> "snippets" and to find content within files on the machine. This is
> really quite important when people are searching they expect this.
> In my experience, people are really not good at naming files well
> and that means you can't rely on just file names and directories
> for finding content by a keyword.


I do think it's important to do more than just indexing of keywords
and so called tags.

The Nepomuk ontology aims pretty hard to allow for rich graphs
connecting different domains of interest with each other.

There ought not to be a difference between an attachment of an E-mail
and a file in the Documents directory of the $HOME of the user.

The $HOME/Documents is for me an implementation detail. The user cares
only about data of the past now, and about data he or others will
create for the future.

Rapid acquisition of data about what is being created for the future
and formal and well known relationships on data about what was created
in the past is of utmost important.

Not just tagging of some keywords.

Who made it? What is my relationship to that person? What are his or
her E-mails? What other communications did we have? Which friends do
we share?

When was it made, in what context? Where was the author when he made
it? What is the content about? What can we learn about that content?

Who is nearby that location? Can I talk with them right now about
this? Are they busy or available?

All these questions are much more important than: what strings can we
find in the content. Besides, much content has no strings or whatever.
No lyrics. No easily discoverable data. Nothing searchable but the
relationships about the data.

These relationships are vastly more important than typical
content-search data.

It's a nut that we must crack, rather than give up and create yet
another persistently stored Map<string, string>.

If Baloo is about a persistently stored Map<string, string> on
key-value pairs related to content of files on your FS: then Baloo
doesn't get metadata at all.

updatedb and locate are good, and grep is good too. Combine the two
and you have a perfect solution. No need for a Baloo index. This is
not what Tracker and Nepomuk are about, at all.

> > small amount of information that we need to store - user tags
> > and ratings are stored in the xattr of the files.

But no relationships with other domains? Pity. With huge respect for
the technical implementation of Baloo, that nonetheless means you
developed not that much.

It also doesn't solve not that much that strings, cat, grep, updatedb
and locate also didn't already solve.

> For xattr and tags, this is actually one of Tracker's weaknesses
> IMO. I was approached by a student (IIRC) asking about this in
> Tracker not long ago and they said they could have some time to
> build support into Tracker. I would like this and it would really
> fill the problem we have right now of persistent data for user
> tagging.
>
> > The Nepomuk KDE project is pretty much dead.
>
> Maybe for the desktop it's not so useful OR apps are simply not 
> embracing the relationships between data like smart phones or set
> top boxes are. What's clear to me is, searching for simple file
> content is always the first requirement and it's never enough. Even
> for the desktop.

First requirement yes. Final goal, no. Not at all.

> I quite like the Nepomuk ontologies and standardising a DB schema.

Me too. Let's stick to Nepomuk and the more complete goal that we have
in mind for Tracker.

> Good luck with your project though Vishesh!

Yes. Good luck nonetheless. Maybe someday we can outsource simple
indexing of easily searchable content to Baloo?

It's boring to do that, so I would very much like to outsource that.

:-)

Kind regards

Philip



-----BEGIN PGP SIGNATURE-----
Version: GnuPG v2.0.20 (MingW32)

iQEcBAEBAgAGBQJUIAiwAAoJEEP2NSGEz4aDARAIAIxlhinSNHMPiUx/Z9VaGvrh
lbppLoemETARiFETz5W7mRL9ZK2fDCxJZriGjiH8/evJs624vB4Jsj3gvjLlJESd
yplZmnImr4ATbvb1xz1e3TPgEeM4EjpEQKLWmHYxgj+QbYv/B8hy/9Xr1waD0M5w
SEY/3dwyJ5wsrQyOfmZOn3oRllwzr1kQbvWxIz+8Gs0XrJfotK5rR3xdoK+J3B46
GacGK0PyUWJnSA2KiwFc88aD9Ui61jVoVxSOCa334TzJCQC5w5t6l4S9Djdh5ADb
HpVIVnb/0MEdu71x4IIiGjz4aXGHG/2HPaUJdyyiZKm/2gtrrJcac5FoYqpMSOY=
=TIRt
-----END PGP SIGNATURE-----