Re: [Tracker] Some observations about tracker 0.5.2

[Jamie McCracken <jamiemcc blueyonder co uk>]

tpgww onepost net wrote:
> Hello all,
>
> I very much welcome tracker, and look forward to seeing it grow (not so much literally, I hope!). After using 
> it for a while, and exploring the code a bit, I have some thoughts to pass back to the dev's.
>
> 1. In keeping with the FD.O spirit, trackerd's configuration and "index" data might be moved from ~/.Tracker 
> to locations in accord with standards.freedesktop.org/basedir-spec/latest. Those places can readily be determined by 
> g_get_user_config_dir() and its siblings. I assume it's practical to amend Makefile to move existing data to new places 
> when upgrade happens.
possibly

> 2. If trackerd's initial data-trawl is interrupted, it seems that a full scan will not be completed unless 
> the user manually deletes the database. Restarting trackerd after interruption just produces quite a few:
> File /home/whome/tmp/Tracker-whome.10995/sqlite3-base/cache-journal has finished changing
> and
> Warning - file /home/whome/tmp/Tracker-whome.10995/sqlite3-base/cache-journal no longer exists - abandoning 
> index on this file

why are you indexing the tmp direcotry? Of course files in there will 
dissapear during indexing and hence you will get a ton of these 
messages. (its obviously pointless indexing tmp too and tracker creates 
a whole load of files in tmp which could cause endless indexinging too)

I guess I should make sure tracker never indexes the designated tmp 
directory especially in the unusual case of having it in your home 
directory (as in your case)

> It would be nicer to have a completion flag, and where appropriate, an offer to restart indexing, or better, 
> to resume from where interrupted.

At the moment we give priority to indexing directories first so that we 
watch them asap.

I guess we should really only index the directory once all the file sin 
it are indexed

will need to think about doing that


> 3. If trackerd is not interrupted, as the indexing gets further advanced I increasingly get blocks of 
> back-to-back messages of the types listed in 2 above. Eventually they greatly-outnumber messages about data 
> actually being processed. Is there some inotify-related optimisation needed there ?

cause is specified above


> 4. The torrent of information produced when trackerd first indexes your data is useful for debugging, but it should be made 
> optional now that tracker is going out into the wider community. The various levels of debug messages should all (or maybe, all 
> except critical warnings) be tied to the "--enable-debug" configure option. If you must, debug mode could default to 
> on, with a related "--disable-debug".

we already do this - we have --enable-debug option to provide more 
verbose output

trackerd by default should be invisible to end users (IE most will not 
manually run it in a terminal)

> 5. CVS includes several python files that don't appear in the release. Is there a problem with them ?

they are experimental additions

probably should add them to Make Dist

> 6. I haven't been able to find a systematic approach to running rdf queries, not even a default-search-path 
> for rdf scripts. Does anyone know of a tool to facilitate script creation ?

on the agenda

> 7. For the future - the command-line utilities which interrogate index data must continue to support 
> specification of search parameters in as much detail as for any GUI, e.g. and, or, not, phrase, anycase ... 
> Probably, GOption will have to be sacrificed.

we would like to keep goption

> 8. I understand that inclusion in gnome will require improved documentation. Some hints:
>  - some of the command-line utilities don't support a --help parameter
>  - Baptiste's initial manpage contributions need to be updated
>  - tracker-search-tool and tracker-tag have no manpage

yes patches welcome for this

> 9. A while back there was some chatter about a replacement icon. Has anything been done ? If not, maybe the 
> attached will start that ball rolling.

hehe I like the use of railway tracks!
But I prefer the tango magnifying glass


> 10. Closer alliance with gnome has some important benefits, but please try to keep tracker's essential 
> DE-independence in front of packagers at least, so that users of xfce and the like (not to forget KDE users 
> now that kat seems to be dead) are not dissuaded.

this should not be a problem. The recent patch for configure.in should 
help alleviate these


> HTH
> Tom

thanks for the feedback!
--
Mr Jamie McCracken
http://jamiemcc.livejournal.com/