Re: A proposal for Summer of Code.

[Debarshi 'Rishi' Ray]

I have something to add regarding this new addition to the SoC ideas page:
"4. Near-real-time enhancements for locate and updatedb"

Once implemented with the "extended meta-data support", it can make
GNU Findutils a command line counterpart of Beagle and Tracker.
Instead of directly interfacing with the kernel's inotify sub-system,
we can also have a look at libinotifytools which is a part of
inotify-tools (http://inotify-tools.sourceforge.net/), and for the
Hurd we can make use of translators. However since my knowledge of
Hurd is limited, I will limit myself to GNU/Linux systems.

Here is a brief outline:

a. Depending on the availability of a meta-data extraction library,
configure Findutils during the build stage to enable/disable the
ability to read the attributes which are unique to a particular type
of file. This should affect all the components-- find, locate,
updatedb, etc..

b. Have a separate updatedb daemon, depending on whether inotify is
available in the running kernel, apart from the conventional updatedb
executable. The daemon should make use of the inotify feature of the
kernel to keep the database updated. An update operation should be
invoked by the events generated by the inotify subsystem and not
involve processing the names or meta-data of files in the database
that have not been changed. Whether this daemon will store the
meta-data is also decided as in (a).

c. Make sure that both updatedb and updatedb daemon are not running at
the same time to prevent database corruption.

d. Try to keep all these features as transparent as possible to users
of find and locate to prevent breaking existing shell scripts. This
means that the default behaviour should not change, and ability to
look at the newly added extra meta-data should only be enabled via
flags.

What do you think?

Regards,
Debarshi
--
GPG key ID: 63D4A5A7
Key server: pgp.mit.edu