[FFmpeg-devel] [FFmpeg-cvslog] random thoughts about SoC (was: Re: random thoughts about refactoring)

[Michael Niedermayer]

On Tue, Jan 12, 2010 at 12:06:55AM +0100, Stefano Sabatini wrote:
> On date Monday 2010-01-11 15:09:38 -0500, Ronald S. Bultje encoded:
> > Hi Diego,
> > 
> > [move to ffmpeg-devel]
> > 
> > On Mon, Jan 11, 2010 at 3:29 AM, Diego Biurrun <diego at biurrun.de> wrote:
> > [cut stuff that wasn't allowed during SoC]
> > > rtp network encapsulations
> > [..]
> > 
> > Luca and I wrote a skeleton for this so students could fill it in. No
> > student applied for this particular project. Next year, same skeleton
> > may be used and one of us will mentor a student if a student is
> > available.
> 
> Random ideas about soc failures / ideas for improvements.
> 
> * Qualification tasks are an important pre-requisites for seeing what
>   a student is able to do and if he's willing to finalize the work. We
>   missed some of them the last year, for example I didn't see any
>   qualification task performed by Kevin, he disappeared after some
>   time, also this is not fair towards students which actually
>   completed qualification tasks. Already valuable contributors should
>   be exempted though.

we tend to be a little short on students who finished a qual task
This tends to lead to disscussion about
giving someone a chance even without qual task or giving a slot back
to google.
Both of these have been done in the past, factors considered are for
example if the respective mentor is beliving that his student would
succeed.


> 
> * IRC presence is important for monitoring the progress of students,
>   both mentors and them should try to be present there the best as
>   possible at least during the qualification / task completion period.

that one has been mentioned by jason and i agree as communication seems
to have been a major problem


> 
> * Nationality of the student can't hardly be considered a good way to
>   judge its competence/motivation,

i assume you meant "can" and his/hers

i think most of us agree here though the IRClog hints towards that some
might have different oppinions.


> in general students from countries
>   where the relative value of the prize is greater are more
>   incentivated to work on those projects, so we expect to have more of
>   them from those countries, maybe it should change the way how Google
>   assigns the amount of money, I mean most of the prize should only be
>   delivered if the task is completed so to discourage the students to
>   take just the first part of it without to do any significant work
>   and disappear.

we have no influence on when, how or how much google gives students we
can just disqualify them as far as i understand, so nothing we can do
here


> 
> * Students which complete the work even after the task completion
>   deadline (rather than just disappear) should be somehow awarded (see
>   the following point).

getting svn write accounts, becoming maintainers, a little fame,
experience in flame wars, ...


[...]
> * The level of difficulty of the task should be carefully weighted, we
>   had had in the past a bunch of very hard tasks which required much
>   time and work to be completed even from experienced FFmpeg
>   programmers (just to do an example: lavfi completion). This is not
>   encouraging students to go on with the task, even the more motivated
>   and skillfull.

lavfi was hard but bobby did very good work on this, i definitly count lavfi
as a success.


> 
> * It is important to define some metrics for evaluating the progress
>   of the student, setting many milestones and evaluating each of them
>   is better than having just one final goal, and usually leads to
>   better performance.

mike has pushed this area recently like in the sense of disqualifying students
if they dont reply for weeks or get nothing done at all ...
beyond that, i think that students who do work & communicate will also
provide some usefull code to us in the end, maybe not in perfect and
finished shape but usefull ...


[...]
> * Completion of already started and uncompleted tasks, with an already
>   existing codebase, should be preferred when possible, as the written
>   not-integrated code tends to rotten and loses its value as the main
>   codebase evolves...

yes but this can be difficult, getting to know a unknown codebase that is
unfinished and probably suboptimally documented while reading a spec and
all that is not something every student
will be able to just do while finishing a soc project in a summer. Some
might be better at writing a smaller project from scratch.

[...]
-- 
Michael     GnuPG fingerprint: 9FF2128B147EF6730BADF133611EC787040B0FAB

it is not once nor twice but times without number that the same ideas make
their appearance in the world. -- Aristotle
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 189 bytes
Desc: Digital signature
URL: <http://lists.mplayerhq.hu/pipermail/ffmpeg-devel/attachments/20100112/bc79b6c5/attachment.pgp>