[Debichem-devel] RFS: libchemps2/1.4-1 -- spin-adapted DMRG for ab initio quantum chemistry

[Sebastian Wouters]

Hi Michael,

Quite a few commits back, setup.py was configured by CMake to fill in the
CheMPS2 version etc. from the CMake option list. Toon Verstraelen however
changed it to make the python build independent from cmake as he found it
unsatisfactory:
https://github.com/SebWouters/CheMPS2/commit/af1870983ce1127664079c9b5f0d0f068bcce665
.

As pkg-config is not by default installed on all systems (I don't know
about mac either), I'm a bit reluctant to change it upstream.

There is however a proper way to handle both problems:

(1) HDF5_INCLUDE_DIRS for cmake can be externally specified. For example
psi4 does this in
https://github.com/psi4/psi4public/blob/master/cmake/ConfigChemps2.cmake .

If I give the option -DHDF5_INCLUDE_DIRS=/path/to/nowhere to cmake, and
print the HDF5_INCLUDE_DIRS variable after every package is set,

    message( STATUS "HDF5_INCLUDE_DIRS = ${HDF5_INCLUDE_DIRS}" )

prints:

    HDF5_INCLUDE_DIRS = /path/to/nowhere;/usr/include

So how about we pass

    -DHDF5_INCLUDE_DIRS=/usr/include/hdf5/serial/

in the SID rules file and deprecate the hard-coded path
/usr/include/hdf5/serial/ in CMakeLists.txt?

(2) Likewise include paths for python can by specified by the CPATH
variable.

So how about we use

    CPATH=../CheMPS2/include:/usr/include/hdf5/serial/

instead of the current CPATH=../CheMPS2/include in debian/rules and
deprecate the hard-coded path /usr/include/hdf5/serial/ in setup.py?

I will update the installation instructions in the README.md and sphinx
webpage accordingly in due time for people who wish to compile and install
CheMPS2 themselves and who have non-default installation directories.

Let me know if you find this option satisfactory, and then I will adjust
upstream & the git repo on alioth.

Best wishes,
Sebastian



2015-06-18 6:34 GMT-04:00 Michael Banck <mbanck at debian.org>:

> Hi,
>
> On Thu, Jun 18, 2015 at 12:20:36PM +0200, Michael Banck wrote:
> > The attached patch uses pkg-config (an additional dependency, but quite
> > lightweight and likely already installed on a development machine) to
> > find HDF5, it looks a bit cleaner to me.
>
> Well, it wasn't against latest git obviously. Alternatively, you can
> just pick the os.environ.get change and set HDF5_INCLUDE_DIRS in
> debian/rules accordingly.
>
> An even more comprehensive change would be to set HDF5_INCLUDE_DIRS via
> cmake, by using configure_file or similar.  Then the person building
> PyCheMPS2 would not need to set an environment variable at all.
>
>
> Michael
>



-- 
****************************************************

  dr. ir. Sebastian Wouters
  Fellow of the Belgian American Educational Foundation
  Princeton University
  (address) Department of Chemistry
            Frick Laboratory 351
            Princeton, NJ 08544, USA
  (e-mail)  sebastianwouters at gmail.com

****************************************************
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.alioth.debian.org/pipermail/debichem-devel/attachments/20150618/b6e0f136/attachment.html>