MPICH update

Anton Gladky gladk at debian.org
Sat Aug 10 21:40:56 UTC 2013


Hi Torquil,

I am CC-ing the debian-science list in case some other members
have better ideas how to proceed.

2013/7/10 Torquil Macdonald Sørensen <torquil at gmail.com>:
> Hi Anton!
>
> We are down to these lintian warnings:
>
> I: mpich-doc: conflicts-with-version mpich2-doc (<< 3.0.4-1)
> W: libmpich2-dev: virtual-package-depends-without-real-package-depends
> depends: libmpich-dev
> I: libmpich-dev: conflicts-with-version libmpich2-dev (<< 3.0.4-1)
> I: mpich: conflicts-with-version mpich2 (<< 3.0.4-1)
>
> where i have neglected three unimportant "binary-without-manpage" warnings,
> because upstream is working on that.
>
> I have two questions and a comment:
>
> 1) The "virtual-package-depends-without-real-package-depends" appears
> because of the ancient MPICH1-based packages:
>
> http://packages.debian.org/source/sid/mpich
>
> which have "libmpich-dev" in their Provides fields, and therefore
> libmpich-dev is listed as a virtual package in
>
> /usr/share/lintian/vendors/debian/ftp-master-auto-reject/data/fields/virtual-packages
>
> How would we go about introducing our new "mpich" package at the expense of
> the old one with the same name? I did contact the authors of that package a
> while ago, but never got a response.

What if we announce it in NEWS-file [1]?

[1] http://www.debian.org/doc/manuals/developers-reference/best-pkging-practices.html#bpp-news-debian

> 2) The lintian warnings "conflicts-with-version" are related to the
> transitional packages that I introduced. I don't know if the upgrade scheme
> I have provided will be satisfactory or not, or if it can be done in a way
> which lintian likes better. All I know is that the upgrade worked on my
> computer. What do you think?

It seems, it is ok.

> 3) I rebuilt (not using pbuilder, though) all of the reverse build depends
> packages for libmpich-dev. All but one compiled successfully, as you can see
> from my notes:
>
> * gromacs, meep-mpich2,  yorick, netpipe, python-scientific (builds OK)
>
> * hdf5 (builds OK. Some warnings about not finding /usr/include/mpich2 (I
> think...) during the build, but it finished successfully despite that.)

Ok, good.

> * mpi-defaults (Its debian/rules must be updated. It analyses files in
> /etc/alternatives in a way that returns the string "mpich" with the mpich
> packages. Problems arise because it doesn't match the string "mpich2", and
> it then complains about an unknown MPI implementation.)

I think we can do it after uploading of mpich-package. mpi-defaults
requires changes/upload anyway, as the last one was about 2 years ago.

> Best regards
> Torquil Sørensen

So, generally I am ready to upload the package. I think, it will not
be harmfull,
as mpich is even not in testing.

We just need probably a NEWS-file.

So when, you are ready with the package, let me know.

Thanks for your work.

Anton



More information about the debian-science-maintainers mailing list