Re: [AMBER-Developers] [ Re: Amber 'configure' now forcing miniconda?]

From: Daniel Roe <>
Date: Thu, 22 Dec 2016 09:44:32 -0500

On Wed, Dec 21, 2016 at 10:04 PM, David Case
<> wrote:
> On Wed, Dec 21, 2016, Dan Roe wrote:
>> But I think it is important that these tests happen at 'configure'
>> time, and it doesn't seem to me to be that difficult....
> It may indeed be do-able, but it is not just a matter of seeing if "import
> numpy" throws an error. The configure script would have to compile python
> extensions in the same way the real code will (which is actually several
> ways), and make sure they all work. And since only developers will decline
> the miniconda download, and since they can find out by trial and error if
> their system python works, writing and maintaining this functionality seems
> like a lot of work for only a small payback.

I think that some users (maybe installs at HPC sites etc) may also
want to decline the miniconda download, but it does seem like python
detection is problematic. As long as the '--with-python' flag works I
think it's fine anyway. It would be nice to have configure be able to
test the python ecosystem but if it's too much work then I'll just put
it on my wishlist.

> For those who want to play with this, configure2 alreay *has* a
> check_compatible_python() function. We commented out the call to this
> function (about line 935) since it was not reliable enough: having
> configure say a python is OK when it is not is quite annoying and confusing.

Agreed - that's worse from a user standpoint for certain.

> Knowing I'm repeating myself: we *really* don't want users installing
> packages into their existing python just to install Amber. There is too
> much danger of creating incompatibilities with other things that require
> python: that is, we increase the danger that installing Amber breaks
> something else on the user's computer. [numpy, in particular, exists in
> several incompatible releases; it is quite easy for program A to require
> version 1.9 and program B to require version 1.10. We can't solve that
> problem, but should not make it worse, either.]

This is one of the gotchas of living in the python universe. Advanced
users can just use the '--with-python' flag. "Normal" users can have a
built-in miniconda. We just have to make it clear during configure
time exactly why miniconda is a good idea instead of the vague "well,
maybe stuff won't work" message of the past.

>> I'll have to test '--with-python', but if it works there is not much
>> of a downside that I can see other than that Amber installation is now
>> the "python keystone", so I'd better not ever move it or I will break
>> every Amber install based on it (and it's another configure flag that
>> has to be specified each time).

I was being a bit facetious about having to specify an extra flag :-)
- that stuff doesn't come through in email sometimes. I think the
'--with-python' flag is the correct solution.

>> > [Next thing we know, people are going to want to be able to specify a build
>> > directory, separate from the sources.....]
>> I still want this :-). For separate installs the size can really add
>> up, more because the Amber 16 source tree weighs in at 2.3 GB - the
>> miniconda install adds 635 MB to that (sizable but not egregious).
> Has anyone tried to just make a shadow directory (using "cp -as" or lndir),
> with links to a clean amber tree, then build in the shadow directory?

Unfortunately, this can't work for multiple parallel library/CUDA
version builds (maybe I'm not understanding what you mean).

> Realistically, I'd recommend bigger disks. You can get 500 Gb external
> SSD drives with USB-3 or USB-C interconnects that are quite fast and will
> allow you create all the Amber images you are ever likely to need....

Just because disk space is cheap doesn't mean we should be wasteful. I
would like to do with Amber what I can do with every other linux
source code package: have one giant source directory that I can point
to an install directory with '--prefix', and then that install
directory contains only what I need to run that version of Amber.
AMBERHOME can point to the common source, but PATH and LD_LIBRARY_PATH
point to the install directory. If I happen to get free time before
now and the code freeze maybe I'll take another crack at making this


Daniel R. Roe
Laboratory of Computational Biology
National Institutes of Health, NHLBI
5635 Fishers Ln, Rm T900
Rockville MD, 20852
AMBER-Developers mailing list
Received on Thu Dec 22 2016 - 07:00:02 PST
Custom Search