Dependency on non-existent atlas3gf-base breaks experimental buildds

[Kevin B. McCarty-3]

Hi list, Camm,

The cernlib package I just uploaded to experimental, to try to build it
against gfortran, is failing to build on all arches:

http://experimental.debian.net/build.php?pkg=cernlib

As far as I can tell, the reason is that the atlas3gf-base package does
not yet exist -- see the bottom of the build log for 2006.dfsg.2-6 for
ia64 (for instance).

Any workaround, or do I just have to wait until you upload
atlas3gf-base?  Unfortunately dpkg-dev no longer installs build-depends
in the order referenced in the control file.  It now installs them in
alphabetical order (see http://bugs.debian.org/457151 ), which means
that lapack3-dev is installed before refblas3-dev.  Hence lapack3-dev's
dependent package (lapack3gf) tries to pull in the nonexistent
atlas3gf-base, and now that dpkg re-orders Build-Depends, there is no
way to set the Build-Depends to get refblas3gf to be installed first in
order to prevent this from happening.

I must say this is a bit annoying...

best regards,

--
Kevin B. McCarty <kmccarty@...>
WWW: http://www.starplot.org/
WWW: http://people.debian.org/~kmccarty/
GPG: public key ID 4F83C751

[Kumar Appaiah]

Dear Kevin,

On Sun, Jan 13, 2008 at 12:11:41PM -0800, Kevin B. McCarty wrote:
> Any workaround, or do I just have to wait until you upload
> atlas3gf-base?  Unfortunately dpkg-dev no longer installs build-depends
> in the order referenced in the control file.  It now installs them in
> alphabetical order (see http://bugs.debian.org/457151 ), which means
> that lapack3-dev is installed before refblas3-dev.  Hence lapack3-dev's
> dependent package (lapack3gf) tries to pull in the nonexistent
> atlas3gf-base, and now that dpkg re-orders Build-Depends, there is no
> way to set the Build-Depends to get refblas3gf to be installed first in
> order to prevent this from happening.

The package in question (for me) is libitpp. I worked around this
easily because I just had to pass the --with-blas=-lblas --with-lapack=-llapack
options to configure to override it's best guesses. So, even while the
buildds installed Atlas, they were safely ignored by configure.

A similar thing for numpy, and I had to work around it in a rather
ugly way:
http://svn.debian.org/wsvn/python-modules/packages/numpy/trunk/debian/patches/02_dontuse_lapack.diff?op=file&rev=4017&sc=1

From my cursory glance at cernlib, I can't say anything, though I do
understand that the solution is less likely to appear like the one for
libitpp, and you may have to to an ugly hack like the one I had to do
for numpy.

> I must say this is a bit annoying...

++++++++++++++++++++++++++1. While the dpkg people are right in saying
that the package build must not the ordering of the Build-Depends, it
is tough to make the change for some pacakges. :-(

Kumar
--
Kumar Appaiah,
458, Jamuna Hostel,
Indian Institute of Technology Madras,
Chennai - 600 036