Re: [Tracker] [PATCH] Make tracker skip mountpoints

[jamie <jamiemcc blueyonder co uk>]

On Mon, 2007-08-06 at 21:30 +0200, Marcus Fritzsch wrote:
> On 8/6/07, Marcus Fritzsch <fritschy googlemail com> wrote:
> > On 8/6/07, jamie <jamiemcc blueyonder co uk> wrote:
> > > On Mon, 2007-08-06 at 13:36 +0200, Marcus Fritzsch wrote:
> > > > On 8/6/07, jamie <jamiemcc blueyonder co uk> wrote:
> > > > > On Mon, 2007-08-06 at 04:03 +0200, Marcus Fritzsch wrote:
> > > > > > This patch prevents tracker from descending into mountpoints which is
> > > > > > a directory other than a watch- or crawl-root. I often have some fs
> > > > > > (sshfs, nfs) mounted under my $HOME and found tracker going down all
> > > > > > these trees. Also, there was a bug report on launchpad about this[1].
> > > > > >
> > > > > > I am however not sure if this method works, also I haven't found much
> > > > > > references when searching for it by google or clusty. If there are
> > > > > > objections to this method of mount-point 'discovery' it would be
> > > > > > possible to read mtab and store it in struct Tracker for processing.
> > > > > >
> > > > > > However as tracker is not that simple I am not aware of any side
> > > > > > effects this patch may have on tracker, I tested it - with mount
> > > > > > points in watched dirs and crawled ones, with fuse and nfs mounts.
> > > > > >
> > > > > > Any comments?
> > > > >
> > > > > yeah its a good idea
> > > > >
> > > > > but take a look at tracker-utils.c:
> > > > >
> > > > > tracker_file_is_no_watched
> > > > > tracker_file_is_crawled
> > > >
> > > > Thought of this too. The function signature needs a 'current' dir and
> > > > a 'testdir' to check wether testdir is a different device then current
> > > > - there is no problem with that? That would be faster - however I
> > > > could also implement it by checking against the crawl and watch
> > > > directory roots list. Which one do you prefer?
> > >
> > > I would prefer the latter but can we store the device signature of the
> > > watch and crawl dirs somewhere so we are not constantly stating them?
> > >
> > > (Could use a GSList to store them and simply iterate over them)
> >
> > Here is an overview on how skipping mount points work:
> >
> > 0a. Fields added: Tracker.root_directory_devices and Tracker.skip_mount_points
> >
> > 0b. Functions added: tracker_add_root_dir (),
> > tracker_add_root_directories () and tracker_file_is_in_root_dir ().
> > All in declared tracker-utils.h and implemented in tracker-utils.c.
> >
> > 1. Each service directory is added to the GSList
> > Tracker.root_directory_devices - there will be no doubles - that is
> > checked with a loop, if it gets too slow I could switch it e.g. to a
> > tree.
> >
> > 2. The new functions check for Tracker.skip_mount_points - so no one
> > has to do the check but those added functions.
> >
> > 3. get_files ignores files returned as not being on the same device as
> > the currently read directory (this, only if skip_mount_points is
> > enabled, see above).
> >
> > 4. A config option SkipMountPoints has been added to the Watches
> > section of tracker.cfg.
> >
> > 5. The default is to descend into mounted directories, as it was before.
> >
> > I tested this patch quite a lot, including a --reindex run.
> >
> > Cheers, Marcus
>
> Hi Jamie,
>
> here is the fix for the config option being in the semantically wrong
> section. I moved it to the Indexing section. Both patches are against
> svn r745.
>
> Cheers, and sry for those spam mails, Marcus

thx applied