[pymvpa] upgrading from Ubuntu 8.04LTS? How?

Yaroslav Halchenko debian at onerussian.com
Thu Sep 17 13:44:51 UTC 2009

On Thu, 17 Sep 2009, Emanuele Olivetti wrote:
> In my lab (~10 quad core, mostly desktop) we all have Ubuntu 8.04LTS
> and use the usual stack of libraries: NumPy, SciPy, matplotlib,
> PyMVPA, MDP, NetworkX, OpenOpt, PyNIFTI, graph-tool, mayavi2,
> hcluster, etc.
Just to reveal a bit of "LTS" notion here... even looking at
into "Package classification and support" would reveal the truth.

              free software 	non-free software
supported 	  Main 	            Restricted
unsupported   Universe 	        Multiverse

so what is LTSed -- only "Main" whatever is critical for Canonical to provide
basic distribution viable to be supported for the commercial clients (i.e. base
libraries, kernel, etc).  For the rest, there is universe, which is not
supported by Canonical, just in a limited amount by the MOTUs.

So, long story short, THERE IS NO *proper* LTS for PyMVPA (and 90% of
other available packages) in Ubuntu.

> The current status is to rely on .deb packages provided
> my main repositories (Ubuntu) plus few "less official" ;-) ones
> (i.e. Michael's uni-magdeburg, Gael's ppa.launchpad.net etc.) and
> very few installation from sources.

Quick news: Michael's uni-magdeburg repository generalized into more


You can use it to have recent PyMVPA installed/upgraded on the majority
of Debian and Ubuntu releases: http://neuro.debian.net/pkgs/python-mvpa.html

> 1) Upgrade to Ubuntu 9.04 which would result in a general improvement,
> even though not the bleeding edge of everything.

> 2) Install locally each library from sources using the latest versions
> and keep the underlying - and very stable - Ubuntu8.04LTS. I don't
> know whether higher-level libraries (e.g., PyMVPA) have issues with
> too recent lower-level libraries (e.g., NumPy) so some limits about
> too recent code could apply.

> 3) Add the lenny repository and have a mixed Ubuntu/Debian
> installation (even though python-numpy will not be updated...).

Alternative might be

4) Use OS with proper LTS, ie Debian  stable ;)  it has numpy
   1.1.0, scipy 0.6.0.

   If you would need more recent releases, and do not want to backport
   anything manually, stick to a mix of
   stable/testing (with proper apt pining) or just testing (with keeping
   some chroot with a system where you could test upgrade path before
   actually doing the upgrade).

   At least in this case there would be no veiled LTS, everything would
   be clean, want stable -- use stable, want to risk a bit - use
   testing ;)  On cluster I maintain(ed) I've always used stable + set
   of backports.

> from sources. But is this a smart solution? About option 3 I really
> never tried it extensively and anyway some basic libraries will not
> upgrade, which is one of the main reasons for transitioning.

mixed Ubuntu/Debian might fail quite miserably at many points... I guess
it would be better to mix Ubuntu releases or just Debian releases.

> Any idea on how to handle this issue? What is your experience? What
> would you suggest?

Set up chroot with Ubuntu8.04LTS in it, install the same list of
packages as you have on your system.  Then start experimenting (you
might like to targz full chroot so later on you could easily start from
the same point):  just go through your tentative solutions and see which
one works ;)  PyMVPA should tolerate quite a bit of misuse, some
scipy functionality is monkey patched (if needed) whenever you
import mvpa.suite... but never know for sure how well it would hold ;)

Just FYI: in 0.4.3 there is a handy mvpa.wtf()  which would provide
valuable information if you run into problems and decide to share them
with us ;)

Cheers Emanuele,
=------------------------------   /v\  ----------------------------=
Keep in touch                    // \\     (yoh@|www.)onerussian.com
Yaroslav Halchenko              /(   )\               ICQ#: 60653192
                   Linux User    ^^-^^    [175555]

More information about the Pkg-ExpPsy-PyMVPA mailing list