On Wed, Apr 16, 2003 at 03:18:29PM +0200, Claudio Sacerdoti Coen wrote: > > And you really cannot put them in a separate C library that is common to > > both stub libs ? > > It would not make much sense. We are talking about a very small set > of #define and 1-line functions. Each binding should always provide > its own separate 50-lines C library. Moreover, I do not see > anything bad a priori in one binding using another one.
No the idea is to move the common part to a separate library, which is provided by the first package and used by the second. But it don't really solve anything and pose the same problems of the symbolic link in /usr/lib. > > 2) imagine you have more than one version of the library around, or > > two different ocaml runtime (ocaml-3.06 and ocaml-3.07 for example), > > then you would have to choose which of these you will have installed > > in the ld.so aware library or in the /etc/ld.so.conf file. > > That's convince me. Probably moving the symbolic links to > /usr/lib (and using versioned .so names) is the best option. Yes and no. What would you do when you build the same version of the library for both ocaml-3.06 and ocaml-3.07 ? Would you bump the so name for the second or something such ? Altough normally the stublibs should not suffer from incompatibility between different ocaml versions, should they ? > > Finally, maybe all this shows that there is a propper usage of rpath > > finally, and that the ocaml_packaging_policy should allow rpath for > > these cases, and these cases only. > > Maybe. But the scary scenario of Denis where you end up compiling > against an old version still applied, doesn't it? We don't care. It is just a matter of discipline, you build your packages in pbuilder or upload source only and let the autobuilders do their job. Friendly, Sven Luther
