Re: Status of gfortran transition
Kumar Appaiah <firstname.lastname@example.org> writes:
> On Thu, Oct 25, 2007 at 11:05:19AM -0400, Camm Maguire wrote:
> > Greetings!
> First of all, it's great to have you back, Camm!
> > Just uploaded refblas3_1.2-9 to experimental. Hope to get lapack and
> > atlas in in the next few days.
> I just had a look at it. While it seems OK, I definitely have to crib
> about the diff.gz. If you observe, it has many lines of the following
> --- refblas3-1.2.orig/src/cgbmv.f
> +++ refblas3-1.2/src/cgbmv.f
> @@ -142,7 +142,7 @@
> LOGICAL LSAME
> EXTERNAL LSAME
> * .. External Subroutines ..
> - EXTERNAL XERBLA
> + EXTERNAL XERBLI
> * .. Intrinsic Functions ..
> INTRINSIC CONJG, MAX, MIN
> * ..
> @@ -171,7 +171,7 @@
> INFO = 13
> END IF
> IF( INFO.NE.0 )THEN
> - CALL XERBLA( 'CGBMV ', INFO )
> + CALL XERBLI( 'CGBMV ', INFO )
> END IF
> (And many more).
> I would really request you to move to a patching system like quilt or
> dpatch in order to isolate the differences to diff files, as that
> makes things very easy to handle, isolating the upstream from the
> Debian packaging.
Thanks for the suggestion! I heartily agree -- I am just so old that
I've wound up implementing an analogous facility in terms of makefile
rules (aka lapack debian/rules, not here of course), and always resist
learning anything new :-), especially that might be replaced in short
order. What is th 10 year solution?
> > I distinctly remember being required to come up with the naming scheme
> > (refblas3 provides libblas.so.3 (virtual)) to get around difficulties
> > with ldconfig and earlier incarnations of the packaging system (in
> > conflict/replace with the earlier libblas) though I cannot recall the
> > precise details at this point. It would make sense to call the
> > package libblas3/libblas3-dev and have these also name virtual
> > packages which atlas could provide. Will this work? Anyone have a
> > better idea? Now would appear a good time to put in what we'll want
> > to have for the next few years.
> This is best addressed by Riku or doko. I'll stay out of this! :-)
> > Finally, it appears as though we might no longer need to mess with
> > /etc/ld.so.conf to get the preferential loading -- that automatically
> > certain subdirs of /usr/lib are searched by ld.so first. Any
> > succinct update on the latest here would be most appreciated.
> Same here.
> > What is the plan from experimental to unstable?
> Well, I volunteer to compile rdepends one by one, if I can, and for
> the rest, ping the maintainer to do so. I think that would check all
> the issues with the refblas and lapack packages, and should bring us
> to good shape with respect to moving it to unstable. However, I am a
> trifle worried about the architectures, since that's where you may
> face most of the problems.
> Anyway, I think I'll now take the back seat, and wait for you to take
> over. But do consider my request to use quilt or dpatch.
> Kumar Appaiah,
> 458, Jamuna Hostel,
> Indian Institute of Technology Madras,
> Chennai - 600 036
Camm Maguire email@example.com
"The earth is but one country, and mankind its citizens." -- Baha'u'llah