[Pkg-scicomp-devel] Debian package for Scotch

[Ondrej Certik]

>> the fact that you provide the scotchmetis interface is a very good
>> thing. Unfoirtunately some functions are missing to be able to use
>> scotch as a full replacement.
>> for example it would be very nice  to have the full  "metis" interface
>>  for suitesparse supported by scotch.
>
> The point is that I do not plan to replace all of the
> routines of the MeTiS library. As a researcher, I do things
> only if they provide an improvement over the state of the
> art. This is why I do Scotch : because I think the algorithms
> I implement are better than the ones of MeTiS.  ;-)

I don't use metis because it's not free.

>
> Therefore, I cannot invest time in just recoding pieces of
> code that bring no added value to the taxpayers who fund me.
> For instance, I do not plan to recode the "mesh2dual" or
> "mesh2nodal" routines of MeTiS, because there is no scientific
> added value for me to do that. I could do it as a hobby, but
> my involvment in the libre software community is already taking
> me much time, on other issues...   :-)

The added value is that your software can be freely used and thus be in Debian.
So if it's not a lot of work, it be nice to have it. :) But of course,
we are all doing it in our spare time, so I understand you.

>
> I got in touch with your group now because in the Scotch 5.0.6
> version that I am about to release, I have tried to comply more
> to the GNU packaging standards : now, the files go to separate
> "./bin", "./include" and "./lib" subdirectories, and I am also
> writing man pages for the binaries. The "make install" target
> sends files to distinct subdirectories relative to the $prefix
> variable. I can provide you with a tarball of it if you want,
> prior to its upcoming release.
>
> I'd like to be able to do something better in term of porting,
> such as basing on autoconf and automake, but I have not been
> able to invest too much in it, so its arcanes are much too
> obscure for me at the time being...  :-(

If it was me, I wouldn't go the autoconf/automake route, as it seems just
messy for me. I like your clean Makefiles. I just tried to compile it myself
and it just works.

>
>> You are most welcome to join the group (list and project). As Ondrej
>> mentionned,  let me know your login name and I will add you to the
>> project.
>
> I am already overworked and over-emailed, so I think I'd better
> say no...     :-)
>
>> PS: I work at the Lab. Jean Kuntzmann in Grenoble. my main interest is
>> scientific computing and more specifically the numerical solution of
>> PDEs
>
> OK. A colleague, then.    ;-)

I am interested in the same things.

Ondrej