Re: [Tracker] Don't crawl directories after start

[Martyn Russell <martyn imendio com>]

Theo Wollenleben wrote:
> Am Samstag, 11. April 2009 17:53:46 schrieb Ivan Frade:
> >  Tracker needs that initial crawling to know if something have change while
> > tracker wasn't running and set the monitors on the folders. Take into
> > account that tracker is just "crawling" and not reindexing the content of
> > the files. It is just comparing the last-modification-date of the
> > files/folders with the "index-time", so if there is no change, the files
> > are not indexed/extracted/analyzed again.
> >
> >  There is no way to disable _only_ the crawling and keep the monitoring
> > using options. You could still modify the code to adapt it to your
> > requirements.
>
> Does this mean the code can be changed such that monitoring without initial 
> scan is possible? In the meantime I found 
> [http://bugzilla.gnome.org/show_bug.cgi?id=502875], where Jamie McCracken 
> claims that crawling is necessary and it is a flaw of inotify. Is that true? 
> I used beagle for a while, which uses inotify to watch files without initial 
> hammering of the disk.

He is right. Inotify was designed to watch specific directories. 
Consider the case where you want to watch ALL files under directory X, 
in that situation (right now) you have to recursively crawl the file 
system and add watches on all directories there. If inotify would allow 
us to just watch $HOME (i.e. X) we would be better off here of course.

However, that's one part of the problem, You should also know that we 
still need to get directory listings because we have to stat() the 
directory for its mtime to compare it to what we have in the database to 
know if we should check individual files in that directory. Also, the 
mtime on a directory doesn't update recursively up the tree, only the 
immediate parent directory mtime is changed, so crawling is absolutely 
necessary. We could of course add an option to only do this the first 
time round, but then you risk not tracking data if something crashes or 
anything is changed outside the run time operation of Tracker as Ivan says.

--
Regards,
Martyn