From maintainers-request at octave dot org Mon Nov 14 10:14:03 2005
Subject: Re: Octave 2.9.4 available for ftp
From: Quentin Spencer <qspencer at ieee dot org>
To: "John W. Eaton" <jwe at bevo dot che dot wisc dot edu>
CC: octave maintainers mailing list <maintainers at octave dot org>
Date: Mon, 14 Nov 2005 10:06:36 -0600

John W. Eaton wrote:

>On 12-Nov-2005, Quentin Spencer wrote:
>
>| 2. It appears that the PKG_ADD file in
>| /usr/libexec/octave/2.9.4/oct/<canonical_host_type> now contains 
>| commands with absolute path names (it didn't as of the last release), 
>| which are created at the make install stage. This is no problem for 
>| someone building and installing locally, but in packaging sytems that 
>| install in a temporary path (creating a deb or rpm, for example), the 
>| path name could be wrong (I learned this the hard way) and require some 
>| post-processing using perl or sed.
>
>The commands are of the form
>
>  autoload ("fcn", "$octlibdir/file.oct");
>
>(with the actual value of $octlibdir substituted when the PKG_ADD file
>is generated) so how is it that octlibdir changes when you run make
>install?  Are you using a value of $prefix at install time that is
>different from the one used at configure time?  If so, I'd suggest
>using $DESTDIR for this purpose instead of changing $prefix.  Is there
>some reason you can't do that?  Do the package tools assume it is OK
>to change prefix at install time?
>  
>

The RPM build system on Fedora (I don't know if this is true of other 
RPM-based distributions or not) provides a makeinstall macro for use in 
spec files that redefines prefix at install time to install the files in 
a temporary location. I didn't realize that this distinction between 
$prefix and $DESTDIR existed; thanks for pointing that out. I have 
modified the spec file to use DESTDIR instead of the makeinstall macro 
and it appears to avoid this problem.

-Quentin