[gmx-developers] unbundling the lmfit library bundled with Gromacs

[David van der Spoel]

On 24/11/15 03:00, Roland Schulz wrote:
> Hi,
>
> Dominik, any contribution should be posted to gerrit.gromacs.org
> <http://gerrit.gromacs.org>. The current situation would only cause a
> problem if an application tried to link both to libgromacs and liblmfit
> correct? Because we don't have the lmfit API as public (but we currently
> still export all symbols so this currently only affects that the headers
> aren't installed).
>
> David, why did you change the source style for those files (e.g.
> brackets, doxygen, ...)? I think for any external code, we want to keep
> files as close to the original as possible to be able to look at the
> difference and also be able to update to a newer version if needed. I
> think if there is no special reason why this is needed we should undo
> those style changes.
I changed the interface to the routine as well. Since this is a fitting 
function I introduced a weighting array (errors in y). This is therefore 
a derived work now. Best way around clashes would be to rename the 
functions inside gromacs I guess.

>
> Roland
>
> On Mon, Nov 23, 2015 at 7:43 PM, Dominik 'Rathann' Mierzejewski
> <dominik at greysector.net <mailto:dominik at greysector.net>> wrote:
>
>     Dear Developers,
>     I am the maintainer of the GROMACS package in Fedora. Recently, I found
>     that GROMACS bundles the lmfit library
>     (http://apps.jcns.fz-juelich.de/doku/sc/lmfit), which is already
>     packaged
>     in Fedora (https://apps.fedoraproject.org/packages/lmfit). However, it
>     seems that the GROMACS-included version is modified from latest 5.1
>     version (cleaned-up patch showing the main differences attached) and in
>     a bad way: it breaks API and ABI (lmcurve_evaluate() and lmcurve()).
>
>     Now, this can be done properly by introducing new API that GROMACS can
>     use while keeping the existing API intact. To that end, I'm attaching
>     a patch that tries to do just that. I'm not claiming my solution is the
>     best or even entirely correct, but it should be a start.
>
>     I'm also attaching an initial patch for GROMACS to use a system-wide
>     shared library with this new API. It's most probably not suitable
>     to be applied as-is, but at least the internal test suite (make check)
>     passes.
>
>     I'm Cc'ing other interested parties, including the Fedora lmfit package
>     maintainer (Miro), the lmfit author (Joachim Wuttke) and Tobias Klauser,
>     who seems to have created a fork of lmfit on github
>     (https://github.com/tklauser/lmfit) which contains at least one fix
>     present in GROMACS fork of lmfit, too (the memleak fix:
>     https://github.com/tklauser/lmfit/commit/1dfe71ae26986b52e6037fb1933904d541ab100f).
>
>     Regards,
>     Dominik
>     --
>     Fedora http://fedoraproject.org/wiki/User:Rathann
>     RPMFusion http://rpmfusion.org
>     "Faith manages."
>              -- Delenn to Lennier in Babylon 5:"Confessions and
>     Lamentations"
>
>
>
>
> --
> ORNL/UT Center for Molecular Biophysics cmb.ornl.gov <http://cmb.ornl.gov>
> 865-241-1537, ORNL PO BOX 2008 MS6309
>
>


-- 
David van der Spoel, Ph.D., Professor of Biology
Dept. of Cell & Molec. Biol., Uppsala University.
Box 596, 75124 Uppsala, Sweden. Phone:	+46184714205.
spoel at xray.bmc.uu.se    http://folding.bmc.uu.se