Re: [gentoo-soc] GSoC - Package statistics

[Brian Dolbec <brian.dolbec@gmail.com>]

[-- Attachment #1: Type: text/plain, Size: 2949 bytes --]

On Tue, 2011-03-22 at 00:39 +0100, Michael Seifert wrote:
> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA1
> 
> Hello Gentoo team,
> 
> I am interested in taking part in this year's Google Summer of Code and
> would like to discuss some general things. More specific, I am talking
> about Christopher Harvey's idea to design an application that raises
> statistics on the installed packages on a system (see [1]).
> 
> Let me explain my issue:
> In the first place, I am a Java developer and I consider my Java skills
> to be quite well. The point is that java is not part of a minimal gentoo
> system, whereas Python is. I would not have problems with Python, but I
> would have to learn the API of a GUI toolkit (preferably PyQt).
> What is your opinion: Would you be okay with a Java application or would
> you rather like to have a Python based one or even some other language?
> And finally, how much time do you think will it cost to work through the
> Qt API?
> 
> 

Well, a good deal of the info gathering methods are already coded in
python (by me) and installed on many users systems already, soon to be
even more as gentoolkit-0.3.0 final is about to be released.

Have a look at the analyse module.  It is designed for accessing the
installed package database and reporting/repairing things about it.
There is room for many more types of reports depending what is needed.
Adding an anonymous data upload should not be difficult.  So far I have
only been adding report types for problems users have run into.  It has
USE flags and keywords reports so far.  It has an api to use so as to
not need terminal output parsing to get the relevant data out of it.
Also so does most of equery and eclean have usable api's now (if there
is anything needed from there (the reason I got involved in gentoolkit
coding).

I don't think a gui would be needed, but rather webapp interfaces to the
data gathered.  Most of the devs are cli die hards, so a simple command
line interface to query the central database should be primary. I
believe it would be more widely used for the ebuilds they maintain. A
browser could be used to connect to it and get graphs, etc. for more
elaborate info displays.  

I would think the main portion of this project would be the database and
webapp query tools.  Potential controversy aside, I would also think you
should add a mechanism for a dev to trigger a simple query for people
having a cat/pkg-ver installed to optionally fill in a few questions
regarding an pkg's stability,... all anonymously.  Those queries could
be generic in nature and come with the data gathering tool, that way
only a simple small string need be downloaded at the time of data upload
(for ex: pkgname, query #, possibly a bug #) no executable code.

So I would concentrate your research and proposal efforts on the
database and webapps.
-- 
Brian Dolbec <brian.dolbec@gmail.com>

[-- Attachment #2: This is a digitally signed message part --]
[-- Type: application/pgp-signature, Size: 490 bytes --]