On Sat, Jan 25, 2014 at 04:30:30PM +0100, Ludovic Court??s wrote:

     > +    (native-inputs
     > +     `(("gfortran" ,gfortran-4.8)
     > +       ("perl" ,perl)
     > +       ("less" ,less)
     > +       ("pkg-config" ,pkg-config)
     > +       ("texinfo" ,texinfo)
     > +       ("ghostscript" ,ghostscript)
     > +       ("gnuplot" ,gnuplot)))
     > +    (propagated-inputs
     > +      `(("texinfo" ,texinfo)
     > +        ("less" ,less)
     > +        ("ghostscript" ,ghostscript)
     > +        ("gnuplot" ,gnuplot)))
     
     Oh, seems like a case where native inputs need to be propagated.
     
     Why do less, Texinfo, etc. need to be propagated?

The octave build system is rather naive.  These propagated inputs don't actually
*need* to be present at configure/build time.  But the ./configure (rather stupidly IMO)
checks for their presence, and turns off the relevant features if they are not found.
Therefore, one must declare them as native-inputs just to keep ./configure happy AND
as propagated inputs because they are called in a pipe from the octave program itself.
     
     If that???s because Octave expects them to be in $PATH, then maybe an
     option would be to use ???wrap-program??? to wrap the ???octave??? program with
     a shell script that initialize $PATH to contain these.
     
     That way, we make sure Octave will run with the ???right??? versions of
     these (those used at compile time and known to work), and the user???s
     environment doesn???t end up pulling them.
     
     An example of ???wrap-program??? is Git (in version-control.scm.)

I don't think this will work in this case, for reasons explained above.

J'
     
     

-- 
PGP Public key ID: 1024D/2DE827B3 
fingerprint = 8797 A26D 0854 2EAB 0285  A290 8A67 719C 2DE8 27B3
See http://sks-keyservers.net or any PGP keyserver for public key.