Test suite of LibXC is not portable

Bug #1248600 reported by Yann Pouillon
2
Affects Status Importance Assigned to Milestone
LibXC
Fix Released
Medium
LibXC Developers

Bug Description

I put here the comments of Miguel Marques on this topic, so that as many people as possible know that a solution is under way.

-----8<-----
   Hi,

   This subject has been showing up in the mailing list quite often, so
please let me explain you the current state of affairs.

   First note that to a considerable extent I am the only continued
developer of libxc. There were important contributions by other people,
of course, by the large majority of the commits are made by me.
Therefore, I developed the "testsuite" (or more precisely the regression
tests) to make sure that I did not break up anything, not to see if the
compilation succeeded in some lambda machine. As such, the "testsuite"
should only run in my computer, with my compiler and my options, and
this to machine precision! Note that libxc is basically just a bunch of
formulas, and there is very little logic. Therefore, it is a very simple
task for compilers ;)

   However, in the past few months several people started compiling
libxc and complaining about the testsuite. I understand the problem, and
I decided therefore that the testsuite should change to be also helpful
to these people. There are two important points:

1) at the moment the densities used in the testsuite are not physical,
and they were taken by convenience several years ago to compare to the
results of the density functional repository. This will change, and I
will be using atomic densities that were calculated with APE by Micael
(and that are already committed to the svn)

2) The comparison will not be performed up to machine precision, but to
a lower precision (still not sure which).

3) second and third derivatives should be included consistently for all
family of functionals (this is not true at the moment).

   Of course, my time is extremely limited, and my current priority is
to have the third derivatives working for the GGA functionals. I already
sorted out the exchange, and I am currently working on the correlation
(35 different derivatives!) After that I will rewrite the testsuite. If
someone wishes to help, I would be very happy indeed!!!!

   Hope you understand the situation now ;)

   cheers,
   Miguel
-----8<-----

Changed in libxc:
status: In Progress → Fix Released
To post a comment you must log in.
This report contains Public information  
Everyone can see this information.

Other bug subscribers

Remote bug watches

Bug watches keep track of this bug in other bug trackers.