[Debichem-devel] RFH/RFC: gfortran transition (status information)

[Michael Banck]

Hi,


On Tue, Feb 26, 2008 at 02:29:55AM +0100, Daniel Leidert wrote:
> This is a short information about the status of the gfortran transition
> of our packages. For a general overview of this transition see
> http://wiki.debian.org/GfortranTransition.

Thanks for a all the work so far!

> Packages affected: mpqc, mopac7, libghemical, ghemical
> (other packages like apbs, psicode have already been migrated)
> 
> Status of mopac7 (1.13-2/UNRELEASED):
> 
>   The package builds with gfortran and f2c, so the main goal to get rid
>   of g2c has been achieved.
> 
>   The library package(s) have NOT yet been renamed nor are there any
>   fixes for .shlibs files or dh_makeshlibs calls. This probably needs to
>   be done.
> 
>   The package (build-)depends on libf2c2(-dev) instead of libg2c0(-dev)
>   now. However, we could get rid of the dependency by building the
>   library from the FORTRAN sources. An explanation how to achieve this
>   can be found at http://bugs.debian.org/465723 (including a statement
>   by Tommi Hassinen). But I'm quite busy atm, so please feel free to
>   step forward and take over testing this solution if possible. The
>   FORTRAN patch to fix the programming bugs has already been added to
>   our SVN. The s_copy() call in fdate.c needs to be replaced.

I wondered about this as well a while ago, and I think it'd be an
elegant solution if ghemical et al. work fine afterwards with this.
 
We would have to check the C source for additional changes/bugfixes not
done on the Fortran source though (but it is my understanding that this
is discouraged or even forbidden)

> So the *main* problem left is:
> 
>   - check if we need to rename the library packages and/or fix .shlibs
>     files/dh_makeshlibs calls (see wiki.d.o)

Haven't looked into this yet either.

> Additionally:
> 
>   - get rid of libf2c2 in MOPAC7 (see #465723)
> 
> 
> *ghemical packages are involved in the openbabel transition. So they
> should not be uploaded until this transition is done. gchempaint needs
> to be built on mips to achieve this:
> http://buildd.debian.org/pkg.cgi?pkg=gchempaint
> http://people.debian.org/~igloo/status.php?packages=gchempaint
> 
> However, mopac7 and mpqc could be uploaded (especially because if we
> rename the library packages, they have to go through NEW and this will
> take days or weeks as you know).

Well, NEW processing is quite fast right now anyway.

The important point here is that if we change the library names, then
the old libraries will be removed from unstable and the ghemical package
will be uninstallable, i.e. will need a rebuild, which would reset the
openbabel1->2 transition.

So we should really wait for the libopenbabel1->2 transition before
starting the gfortran transition, as painful as this is.

One thing we can do though is upload those packages to experimental.
That would (i) get us limited autobuilding (ii) remove the NEW wait time
for unstable and (iii) possibly some limited testing by users.
 
> As a bonus I built all affected packages with GCC 4.3 and gfortran in
> pbuilder to check if we fixed all issues. Seems we did :) Packages built
> and tested: the above + liboglappth and openbabel.

Cool, thanks again.


Michael