Bug#328732: [Pkg-octave-devel] Bug#328732: please update octave2.1-headers's dependencies

Dirk Eddelbuettel edd at debian.org
Sat Sep 17 03:07:34 UTC 2005


On 17 September 2005 at 04:37, Rafael Laboissiere wrote:
| * Aaron M. Ucko <ucko at debian.org> [2005-09-16 21:54]:
| 
| > Package: octave2.1-headers
| > Version: 2.1.71-3
| > Severity: important
| > 
| > In general, octave2.1-headers's dependencies should probably reflect
| > octave2.1's build-dependencies.  In particular, building octave2.1 against
| > gfortran causes mkoctfile to refer to gfortran and its development
| > libraries, whereas octave2.1-headers still depends on g77 | fort77, causing
| > the new octave-forge package to FTBFS.  It's possible that other
| > dependencies will also need to be updated, though nothing else jumps out at
| > me.
| 
| Thanks for spotting this problem.  Besides the gfortran vs g77
| incompatibility, I see other dependencies that can also cause problems:
| 
|     refblas3-dev | atlas3-base-dev
|     lapack3-dev | atlas3-base-dev
|     libhdf5-serial-dev
| 
| The first two are problematic because of the alternation.  If Octave is
| build with, say, refblas3-dev then mkoctfile will fail if the system
| contains only atlas3-base-dev, even though octave2.1-headears could be
| happily installed.

No, as they all provide the virtual package "libblas.so.3" that the Depends:
is really on, as per the shlibs files.  (Unless Camm changed that
recently...)

| The dependency on libhdf5-serial-dev is problematic because it is
| versioned in Build-Depends.  It should also be so for octave2.1-headers. 
| 
| I am inclined to change both the Build-Dependencies and the binary
| dependencies for octave2.1-headers to the following:
| 
|     refblas3-dev
|     lapack3-dev

Whenever you contemplate a change here, *please* run it by Camm first, and
have patience to wait a day or two for his reply. Well worth it, in my
experience :)

|     libhdf5-serial-dev (>= 1.6.4)
|     
| I hope the other members of the Debian Octave Group will comment on this.

Sure thing. :)

Dirk

-- 
Statistics: The (futile) attempt to offer certainty about uncertainty.
         -- Roger Koenker, 'Dictionary of Received Ideas of Statistics'




More information about the Pkg-octave-devel mailing list