[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Please reconsider closure of # 457151 -- it affects gfortran transition



Dear dpkg maintainers,

Let me present to you a situation caused by the change to dpkg-dev to
install Build-Depends in alphabetical order:

1) atlas3-base-dev package provides an alternative version of
liblapack.so and libblas.so to the packages refblas3-dev and lapack3-dev

2) Maintainer of those packages (Camm Maguire, in CC) has the runtime
library package lapack3 Depend on "atlas3-base | refblas3 | libblas.so.3"

3) Up till now, it's been possible for a package (for instance, my
package Cernlib) that build-depends on these -dev packages to make sure
the runtime library package "atlas3-base" doesn't get installed, by use
of "Build-Depends: refblas3-dev, lapack3-dev".  The rationale at the
time was just that atlas3-base{,-dev} are larger, so take longer to
download and unpack, causing more wear and tear on buildds.

4) Now that dpkg re-orders Build-Depends, atlas3-base always ends up
installed since lapack3-dev comes alphabetically before refblas3-dev.

5) Until recently this was just an annoyance.  But now that we are going
through the gfortran -> g77 transition, lapack3 and refblas3 (runtime
libs) have been renamed to lapack3gf and refblas3gf.  ATLAS has not yet
transitioned, so there is not yet an atlas3gf-base package existing.
But (in anticipation of it) lapack3gf already Depends on "atlas3gf-base
| refblas3gf | libblas.so.3gf".

6) As a result, the version of Cernlib I just uploaded to experimental
for the purpose of furthering the gfortran transition FTBFSes on all
arches due to unsatisfied dependencies:
http://experimental.debian.net/build.php?pkg=cernlib

7) If dpkg was reverted not to re-order Build-Depends, I could force
refblas3gf to be installed first, satisfying the dependency of lapack3gf
on "atlas3gf-base | refblas3gf | libblas.so.3gf" and preventing the
attempted installation of non-existent atlas3gf-base.

8) If dpkg is not reverted, I fear that the progress of the gfortran
transition will come to a halt until ATLAS can be made to transition
(which due to technical difficulties could possibly be a long time in
coming).

So please reconsider the closing of this bug.

best regards,

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

Attachment: signature.asc
Description: OpenPGP digital signature


Reply to: