[Neurodebian-upstream] PyNN + NeuroTools packages

[Yury V. Zaytsev]

On Sun, 2010-11-14 at 00:04 -0500, Yaroslav Halchenko wrote:

> heh SfN... it charged us more than 2k$ to setup our basic Debian
> booth at SfN2010, despite our description of what Debian is about and
> what (enthusiasm with absent financial insensitive) makes it.
> Great to have James Haxby's support for all our "silly" endeavors ;)

Could you please send one copy of James Haxby my way? ;-)

> meanwhile, we are thinking about creating our "NeuroDebian-branded" live
> USB/DVD... since we make everything public, customizing it for a
> particular need might become a trivial job ;)

Well, the LiveCD I was talking about is basically imagined as
uncustomized Lucid LiveCD with some packages removed to shrink it and
some installed. This requires minimum amount of work, but, on the other
hand, it's something that empowers users to try simulators without any
pre-knowledge of Linux as well as software installation and packaging.

This is extremely important because many candidate scientists for using
state of the art simulation technology instead of unvalidated, slow and
buggy homebrew Matlab scripts can be attracted this way. I have seen
reports that some already use similar LiveCDs as a production simulation
environment, because their demands in what concerns the speed of the
simulation are rather modest (after you have used Matlab everything
seems to be so fast all around!).

I am not sure how you envision your ND branded distribution, but if I
ever had more time, I would go for writing a UCK [1] script and if time
permits, making a separate package for branding. 

This way you can rebuild your LiveCD regularly to include latest updates
in a fully automated fashion and in the same way I could also make use
of it by throwing in some more package names that I need with minimal
effort.

Also unless it's all packaged, I would go for keeping customizations to
the minimum. After all, one purpose of these LiveCDs is that an
attracted user can just install it on his system permanently, and now if
he is getting something completely different from what he saw while
working on the LiveCD, this would be extremely annoying and cause a lot
of frustration.

On the other hand, it's easily possible to throw in extra NeuroDebian
sources in the sources.d upon installation if you have it packaged as
neurodebian-desktop for instance and trigger the installation of the
same set of packages you have on the LiveCD.

Are you aware of the existence of the INCF LiveCD [2] by the way? They
have a somehow similar goal, but the way it is implemented in completely
non-transparent and community-unfriendly (obviously not because of the
bad will though, but most probably lack of resources). Maybe if your
project already has some traction and money you can somehow negotiate
with them to transfer you all they have so far and name it the official
successor of the INCF thing (which will bring you more users)...

[1]: http://uck.sourceforge.net/
[2]: http://software.incf.org/software/cnslive

> well... there is not much actually in that policy, which is also not per
> se a strict official policy but rather guidelines, thus not always
> strictly enforced.  For distutils/setuptools-build packages simple
> packaging (no unittesting during build etc) is trivial especially with
> debhelper7

Unfortunately, most of the packages you would want to get done are not
exactly properly packaged trivial pure-Python distutils/setuptools
scripts :-/ 

I think Debian wiki definitively needs updating with recipes and a set
of exemplar packages utilizing each of the currently used packaging
helpers (python-support, python-central, dh_python2 etc.)

Also, everything that is discussed on IRC / mailing list is reflected in
the policy with a huge lag. Take for instance the recent discussions on
XS-P-V / XB-P-V / X-P-V. It took me a lot of reading and bugging people
on IRC to find out what is the common wisdom on this topic at the
moment.

>  some terms for "Nest license" seems to be utter overkill imho which
>  simply forbid sensible re-distribution, thus making it not even a
>  candidate for Debian non-free.

This is a well-known issue and I am not quite sure if it will get solved
in the near future (I certainly do hope so, but...). The major problem
is that corporate contributors want to control the distribution and
limit the use of the technology by the competitors.

On the other hand, this licence doesn't prevent me from packaging it for
my own use and releasing my package under MIT, so if for instance
someone wants a cluster deployment they don't have to go through the
same pain. And of course, I do have an agreement with myself :-)

However, there is another beacon of hope floating around, that is that
NEST initiative can officially set up an apt repository in the same
controlled manner, but at least, users will get updates automatically
and won't need to compile themselves, which is not trivial in all cases.
Obviously, the rules of NEST initiative do not apply to the initiative
itself ;-)

> well, we have IRC OFTC/#exppsy but we are not very active there I must
> say
> 
> may be it would be more productive to use the mailing list, e.g.
> neurodebian-devel if you like to take it off -upstream list

OK :-)

P.S. Could you please set the lists in a way so that they don't send out
duplicates when they see a subscriber in the list of direct recipients?
I remember it was somewhere in the mailman options...
 
-- 
Sincerely yours,
Yury V. Zaytsev