[Pkg-octave-devel] Packaging Octave 2.9.3

[David Bateman]

John W. Eaton wrote:

>On 28-Sep-2005, Rafael Laboissiere wrote:
>
>| * John W. Eaton <jwe at bevo.che.wisc.edu> [2005-09-28 13:06]:
>| 
>| > There will likely be a 2.9.4 snapshot soon.  It will depend on
>| > ufsparse instead of umfpack.  We are expecting ufsparse to provide the
>| > libraries -lamd -lumfpack and -lcolamd.  You can check the current
>| > configure.in in CVS for more details.  Will you please check on the
>| > status of packaging ufsparse for Debian?
>| 
>| It does not appear to exist a ufsparse package in Debian.  I cannot even
>| find the upstream location.
>
>http://www.cise.ufl.edu/research/sparse/beta/Aug31_2005/
>
>| If needed, The Debian Octave Group will package ufsparse for Debian.
>
>OK.  There is one problem.  Cholmod from ufsparse requires METIS,
>which is distributed under terms that are not compatible with the GPL,
>so we won't be able to use it unless the license changes.  David
>Bateman has asked the author about licensing, but I don't know the
>outcome.
>
>jwe
>  
>
The outcome was silence for a very long time... So I'm not hopeful for 
metis.. Note that debian has the parallel version of metis (pmetis) 
packaged which includes the basic metis functionality. However, I 
believe that this probably poses a licensing problem for debian...

At the moment cholmod (and metis) aren't used in octave since I have not 
had the time to write the interface to allow sparse cholesky 
decompositions. Also cholmod can be used without metis, with the 
consequence of a significantly slower decomposition..

Note that UFsparse is still considered beta by its author, and in 
particular he has advanced the version numbers of certain packages and 
expects to make another release without changing them again. In 
particular I proposed several patches. Most of these aren't important 
for debian (build under mingw, split/unsplit complex parts in umf_solve 
for particular use of octave). However, there is a single patch that is 
essential

*** ./CCOLAMD/ccolamd.h~        Fri Aug 19 20:28:07 2005
--- ./CCOLAMD/ccolamd.h Sun Sep 18 02:32:33 2005
***************
*** 349,355 ****
      long Stack [ ]
  ) ; 
 
! int (*ccolamd_printf) (const char *, ...) ; 
 
  #ifdef __cplusplus
  }
--- 349,355 ----
      long Stack [ ]
  ) ; 
 
! extern int (*ccolamd_printf) (const char *, ...) ; 
 
  #ifdef __cplusplus
  }


without this patch, octave will fail to build as ccolamd.h is included 
in multiple places...

Regards
David

-- 
David Bateman                                David.Bateman at motorola.com
Motorola Labs - Paris                        +33 1 69 35 48 04 (Ph) 
Parc Les Algorithmes, Commune de St Aubin    +33 1 69 35 77 01 (Fax) 
91193 Gif-Sur-Yvette FRANCE

The information contained in this communication has been classified as: 

[x] General Business Information 
[ ] Motorola Internal Use Only 
[ ] Motorola Confidential Proprietary