Re: gfortran-4.3 the default gfortran compiler, blas and lapack uploaded to unstable
On 24/01/2008, Kevin B. McCarty wrote:
> As soon as your new NMUs of the lapack and atlas packages have passed
> through NEW into Sid, the following will hold:
Sorry to interrupt the thread, but if the Ubuntu build logs are
anything to go by, I guess we could face lapack FTBFSing on us on
several architectures (notably, even i386).
https://launchpad.net/ubuntu/+source/lapack/3.0.20000531a-1.1ubuntu1
Though I cannot say anything about the cause (I have to test it out
myself), one thing I am unhappy about is that the current Blas in
Debian is one I re-packaged based on the latest upstream sources, part
of which is based on Lapack sources (see
http://lists.debian.org/debian-toolchain/2007/12/msg00012.html for
details), and Camm's Debian packaging. Now, building the "old" Lapack
against the new source code based Blas doesn't sound right to me.
Matthias agreed with this, but said he was unable to switch to the new
Lapack since he, like me, did not know how to handle Camm's heavy
patches.
My cursory glance (last month) to this problem led me to believe that
several of Camm's patches have been merged upstream in Lapack, while
others (which do not apply per se) may or may not be required.
Therefore, my first attempt (unless Camm comes back to take over by
then) would be to rebuild the Lapack upstream sources without Camm's
patches. The rationale behind this is the fact that the last upload of
a gfortranized lapack3 to experimental (which Matthias did based on a
package I made some time back) built on all architectures where the
old refblas3 built, and on other arches, it didn't build due to Blas
absence:
http://experimental.ftbfs.de/new/package.php?p=lapack3&suite=experimental
I would, therefore, try out two things; one would be a totally
unpatched Lapack, and the second level would be to painstakingly merge
Camm's patches with the new Lapack. However, the latter, I assure you,
will be _really_ painful; I had to spend several hours for Blas, and
if that is any indication, re-patching Lapack is going to be worse
(though I am willing to do it, should the need to do it arise). But,
as always it'd be great if someone else could do this before me! :-)
(Please note that I am no expert in either the toolchain or Fortran,
but I just hope my remarks make sense. Apologies for any mistakes I
make).
Thanks.
Kumar
--
Kumar Appaiah,
458, Jamuna Hostel,
Indian Institute of Technology Madras,
Chennai - 600036
Reply to: