Re: [gentoo-soc] Package statistics reporting tool

[Vikraman Choudhury]

On Tue, Mar 29, 2011 at 7:01 PM, Yanbo Wang <wybb86@×××××.com> wrote:

>  Hi all, I'm a student from China. I'm interested in this idea and have
> different ideas to share.
>
> 于 2011-3-29 22:20, Brian Dolbec 写道:
>
> On Tue, 2011-03-29 at 08:53 -0500, Donnie Berkholz wrote:
>
>  On 18:54 Tue 29 Mar     , Vikraman wrote:
>
>  I've been doing research on this project for sometime, and would
> really like my ideas to be reviewed by the community before
> submitting my application.
>
> * User runs a script as a cronjob (or a portage hook ?), which
> collects stats on installed packages, and uploads to the online
> database. Surprisingly easy to code using the gentoolkit API.
>
>  :) Thanks, anything else needed in there, just ask.
>
>
>  Sure, one of those two options sounds good. The choice can be easily
> changed later, but I prefer the idea of a hook that runs after the
> package manager completes installing a package set.
>
> People on laptops may not run cron jobs regularly, or at all.
>
>   good point.
>
>
>   * Webapp generates statistics on the packages (top packages,
> versions, popularity, packages from overlays). Thanks to ideas from
> an old thread on the forums. Devs, please suggest some more useful
> ideas/stats too.
>
>  For Gentoo users:
> - Search by package name
> - Search by packages often installed with this one (recommendations)
> - Add ratings. Maybe just via the GUI frontend for now.
> - Add user reputations and profiles. How many ratings, etc. Provide the
>   option of logging in as user or anonymously for package uploads.
>
>
>
>  I think the statistics about relationship between packages is more
> important, because if the packages installed together run well and appear
> together frequently, we can recommend one package to the user when the
> other package is installed. and if the packages installed together usually
> crash, there may be some problems between them.
> So there can be more types of deep statistics that provide more useful
> information.
>
>  Tools for Gentoo developers:
> - Least popular packages
> - Search by maintainer/herd
> - Integrate with Bugzilla by showing # of bugs & linking to the list
> - Integrate w/ BZ by showing # of tinderbox-related bugs
>
> For example, a useful search would be ≤20 users, package assigned to
> maintainer-needed or maintainer wanted, and ≥1 open bugs.
>
>
>  * Dev/user runs a cli script to read stats or a gui program to
> display graphs.
>
> I'm looking forward to coding the project in Python, gui using
> PyGTK, database using MySQL. For the web service implementation,
> should I use SOAP (soappy, which I'll need to learn) or something
> like web.py (I've used it earlier) ?
>
>  What is the role of the GUI? Please describe the use cases and justify
> why it should be a desktop app rather than a webapp.
>
>  As much of a gui person as I am, I too think a web browser would be
> preferred connecting for this. It could also be expanded as the need or
> new query type comes up and then becomes available to all without the
> need to upgrade.
>
>
>  Consider REST for the web service rather than SOAP. There's a nice book
> called "RESTful Web Services" you might like to take a look at.
>
>
>  It was also suggested earlier to allow a dev to trigger a popup
> feedback from the user. I think that would require the user ro run a
> daemon, which will listen for queries.
>
>  It doesn't necessarily have to be instant. Upon the next installation,
> when uploading new stats, the client could check for messages.
>
>  Yes, this was what I was trying to suggest at the time, perhaps I could
> have chosen better words :/
>
>
>   Please suggest what details to include and what not to include in
> the application.
>
>  Many resources exist on this topic:
>
> - http://www.gentoo.org/proj/en/userrel/soc/applying.xml
> - http://www.booki.cc/gsocstudentguide/_v/1.0/writing-a-proposal/
> - http://google-opensource.blogspot.com/2011/03/dos-and-donts-of-google-summer-of-code.html
>
>   The informations about installed packages are regularly uploaded to the
> online database, but they are usually reduplicate because most of them may
> be same. I think each host can be identified by a unique random id that is
> used to track submissions issued by the same host. Security is very
> important for this project, the host id should be kept secret.
>
> --
> Thanks,
> Yanbo Wang
>
>  Yes, smolt handles that by generating a uuid from
/proc/sys/kernel/random/uuid for the first time and saving it in
/etc/smolt/hw-uuid

-- 
vh4x0r