[yt-users] installation issues with yt 2.5.5

[Matthew Turk]

Hi Luigi,

Thanks for the report!

On Tue, Sep 17, 2013 at 6:20 AM, Luigi Iapichino
<LUIGI at uni-heidelberg.de> wrote:
> Dear all,
>
> while installing yt 2.5.5 I had to face some new issues. As you maybe
> remember, I have yt on a machine (SuperMUC in Garching, Germany) whose
> firewall permits only incoming SSH-connections. Running the install script
> there is rather annoying but, by making use of remote mounting through sshfs
> on my local PC, I have been able to always install yt. In this process, it
> is crucial that, whenever possible, the install script is organised in two
> distinct parts: a download part, run on my local PC, where I get all
> dependencies, and an installation part, run on the remote machine.
>
> This has worked pretty well so far but, in the newest script of yt 2.5.5,
> during the installation matplotlib-1.3.0 looks for additional dependencies
> of its own, and tries to download them from the Web, crashing the script.
> These dependencies are, according to
> matplotlib-1.3.0/lib/matplotlib.egg-info/requires.txt :
>
> python-dateutil
> tornado
> pyparsing>=1.5.6,!=2.0.0
> nose
>
> I downloaded from the Web the first and third one, and installed them by
> hand before installing matplotlib. The second and fourth one are already
> among the yt dependencies, but I had to move in the right place, before
> installing matlotlib. In practice, I had to add the following lines before
> installing matplotlib:
>
> do_setup_py $NOSE
> do_setup_py pyparsing-1.5.6
> do_setup_py $TORNADO
> do_setup_py python-dateutil-1.5
>
> and commenting do_setup_py $NOSE and do_setup_py $TORNADO in other parts of
> the script. In this way, the installation worked again.

Ah, thanks -- this seems like a straightforward change.

>
> In the yt website it is stressed to mention any issue with installation,
> therefore my email on this subject.

Thank you!

> Apparently my installation works well
> now, so you may consider to test and eventually incorporate these changes in
> install_script.sh , and whether to include pyparsing and python-dateutil to
> the yt dependencies. Alternatively, one could consider to go back to the
> previous version of matplotlib, which did not attempt such downloads, but I
> am not an expert in matplotlib, so possibly there were better reasons fo you
> to move to 1.3.0.

There were some crucial bug fixes in 1.3.0, although I must confess it
*has* brought its share of problems -- the change in the backend
selection, the dependencies you mention, and a few other little
things.

...but on the topic of the install script, I think I speak for some of
the other contributors when I say I'm growing increasingly uneasy
about the process we're having to go through ( http://goo.gl/VFKa8G )
to keep it working.  Basically, we're trying really hard to keep up
with changes, keep things modern, but the install script is a game we
can't win for much longer.  The changes to the matplotlib dependencies
(which were made for perfectly reasonable reasons) are another
example.  While in all likelihood they will work just fine for
connected machines, for non-connected they will pose the problems you
describe above.

So, what we have been exploring has been a three-pronged approach.
This is still in the discussion phase, but we've been trying to
determine if it's viable.

 * Strip the install_script.sh down considerably, removing
non-essentials.  This should function
 * Add a "get_yt.sh" script based on Conda (
http://docs.continuum.io/conda/index.html ), which will do binary
installs of a very small set of packages.
 * Make yt exist much better as a python package, with
pip-installation being first class, and with better instructions for
how to install it from both source and released packages.

Anyway, this is all kind of hard, and we're running into issues with
Conda that we don't yet know how to get around.  (This is all in the
thread entitled "Packaging." on yt-dev.)  But ... this is something we
really need to figure out.

Thanks for your comments, and I'm glad it's now working!

-Matt

>
> Cheers,
>
>   Luigi
>
> --
> Luigi Iapichino
> Universität Heidelberg, Zentrum für Astronomie
> Institut für Theoretische Astrophysik
> Philosophenweg 12, D-69120 Heidelberg, Germany
> Tel: +49 6221 548983, Fax: +49 6221 544221
> e-mail: luigi at uni-heidelberg.de
> URL: http://www.ita.uni-heidelberg.de/~luigi/
>
> _______________________________________________
> yt-users mailing list
> yt-users at lists.spacepope.org
> http://lists.spacepope.org/listinfo.cgi/yt-users-spacepope.org