On Sat, May 20, 2017, Robert Wohlhueter wrote:
>
> Thanks for your responses. They were helpful, and I got "analyse.sh"
> working.
>
> That my system python is "bad" is a relative statement: I have dozens
> of programs that use python, some requiring version x, another version y
> (and some, like amber which distributes their own version). And I admit
> to a baroque set of scripts and aliases to try to satisfy each
> pthon-dependent-application.
>
> The most expedient solution for me was simply to replace the
> synaptic-distribution-numpy version of _umath_linalg.so with the Amber16
> version. (Hopefully, that won't break other numpy-dependent programs.
> But is it does, I'm sensitized.)
I know that these can be quasi-religious items for people in the python
community. Here's my view, which may have serious flaws:
Various python incompatibilities (versions of the interpreter
itself, numpy versions....) have led many programs in our field
to distribut a version of python along with their codes, and use
that, avoiding dependence on the "system" python.
This approach is simplified by the anaconda/miniconda approach, which now
allows us to ask users to install a python that we know works with Amber.
Strong objections from key people have prevented us from making this a
requirement, however. And some examples and tutorials have made the
assumption that the "python" found first in the PATH is the one to use, even
if there is also an "amber.python" in the PATH. The latter is both more
likely to work, and is certainly easier to debug than an unknown system
python. Using amber.python should also discourage people from replacing
a system-wide numpy component with something else, with potentially bad
consequences for other programs.
I'm cc-ing this to the developers mailing list. My recommendation is
that we re-double our efforts to have users install a python inside the
$AMBERHOME tree, and access it via the "amber.python" alias. Programs,
examples and tutorials should consider using amber.python as the
interpreter, only falling back on "/usr/bin/env python" if the former
doesn't exist. This makes life a little less convenient for developers,
but decreases the chances that users will find python-related problems
in the first place, or that they may adversely affect non-Amber work in
trying to fix these problems.
(Apologies for being so long-winded.)
....dac
>
> Thanks again,
>
> Bob W.
>
> On 5/20/17 9:31 AM, David Case wrote:
> > On Fri, May 19, 2017, Hai Nguyen wrote:
> >
> >> sounds like your system python is bad. Can you try to replace "python"
> >> by "amber.python" (to use $AMBERHOME/miniconda python)?
> > Sounds like this may already have been done:
> >
> >>> Using `find`, I also see an instance of this shared-lib in
> >>> $AMBERHONE/miniconda/pkgs/numpy-1.12.0-py27_nomkl_0/lib/python2.7/site-packages/numpy/linalg/_umath_linalg.so.
> >>> NONE of these instance defines the symbol
> >>> "ATL_zttrsm" (as probed by `nm -Ca <lib-path-name> | grep ATL_zttrsm`)
> > But if you haven't tried using amber.python, it's certainly worth a try.
> >
> >>> Any hints appreciated,
> > I don't have any real hints, and I think this is the first time this problem
> > has been reported.
> >
> > [Developers: we don't have any systematic way at all of keeping the
> > tutorials in sync with the current code. Suggestions are welcome. Maybe
> > new students/postdocs could be required to work through a tutorial,
> > comment on problems or things that are difficult to understand, and post
> > results to a wiki page? Other ideas might be better.
> >
> > It would be great if someone now could check tutorial A9.]
> >
> > ....dac
> >
> >
> > _______________________________________________
> > AMBER mailing list
> > AMBER.ambermd.org
> > http://lists.ambermd.org/mailman/listinfo/amber
> >
>
> _______________________________________________
> AMBER mailing list
> AMBER.ambermd.org
> http://lists.ambermd.org/mailman/listinfo/amber
--
====================================================================
David A. Case | david.case.rutgers.edu
Dept. of Chemistry & Chemical Biology |
Rutgers University | office: +1-848-445-5885
174 Frelinghuysen Road, Rm. 208b | cell: +1-609-751-8668
Piscataway, NJ 08854 USA | http://casegroup.rutgers.edu
====================================================================
_______________________________________________
AMBER-Developers mailing list
AMBER-Developers.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber-developers
Received on Sat May 20 2017 - 22:00:02 PDT
