Re: [Tracker] Tracker and Zeitgeist

[Richard Hughes <hughsient gmail com>]

2009/7/15 Martyn Russell <martyn lanedo com>:
> 1. We need to (because of inotify) crawl the file system to set up monitors
> right now. With the new work alexl and others are doing in the kernel with
> FANotify, we should be able to avoid this in the future.

Yes, this should be good. I've also noticed bugs where tracker (or
beagle) use up all the inotify watches, which breaks all PolicyKit
applications nicely.

> 2. We get all files and directories anyway because we don't know what
> changed since tracker was last run. We can circumvent this by having a "last
> checked" timestamp in the database perhaps and by not doing this on every
> boot.

Yes, this would be good.

> We only do this in the unlikely case tracker crashes or someone
> creates content while it is not running. This is quite unlikely of course
> and we could just force a scan like fsck does every n days and rely on
> tracker being constantly running for the rest of the time. We may even be
> able to just provide some applet menu item to recheck the system if files
> are missing.

I don't think it's needed. Perhaps for tracker developers, but not
real life users.

> With 0.7 the time taken to check all files is pretty small. Only on an
> initial index is it intensive, but I don't think you can change that much
> anyway.

Depends if you're sitting on a dual core 1.7Ghz or a OLPC, but I do
agree that tracker is going in the right direction from a performance
and power point of view.

> Precisely what I was suggesting at the summit actually. It is a steep
> learning curve and that's why tracker-sparql has a man page with some
> examples now. But I would like to also provide some simply common use cases
> (in another binary?) or with tracker-sparql for users to do some quick
> checks which are common.

I would be surprised if the average user even knew they were using
tracker, much less sparql. The average user would just notice that the
"search bar finds more stuff". If a user has to read about SPARQL
before they can find a document, then we've already lost.

Richard.