Aurelien Jarno wrote:
Denis Barbier wrote:On Mon, May 22, 2006 at 04:08:56PM -0400, Daniel Jacobowitz wrote:On Mon, May 22, 2006 at 09:06:01PM +0200, Denis Barbier wrote:IMO we should do the same, but how do we handle this transition? We must be sure that /etc/ld.so.conf is updated before any package which runs ldconfig and ships an /etc/ld.so.conf.d/foo.conf file. A simple solution is to add this line in libc6.postinst for etch and modify other packages only after etch is released. Or is there a safe way to allow these changes in etch and make sure that /etc/ld.so.conf is updated first?Why not make any package which ships an ldconfig conf file Depend on a version of glibc which will add the line?Libc binaries have a different name per architecture, so writing the Depends field is quite painful. But this can be done easily in conjunction with a shlib bump, like with 2.4, yes.What about an shlib bump, even for 2.3.6 (even if I would prefer to see 2.4 in etch). We did one recently on i386, and in any case there will be one from sarge to etch.The only thing we should do here is not to block migrations from unstable to testing with that bump. The solution, if the RMs agree, is to implement the feature, but clearly states it should not be used (except maybe in the glibc itself, for multiarch). Then when the package is ready for testing, we could bump the shlib, upload with urgency=high so that we block testing only 2 days, which should be acceptable.I wonder though if /etc/ld.so.conf.d/foo.conf is a good choice. Do these files really need to be configuration files, or can we use /lib/ld.so.conf.d/foo.conf instead? Admins can always edit /etc/ld.so.conf if they want to adjust it manually.How is it done on other distributions? /me prefers /etc, but I am not sure...
On SuSE: include /etc/ld.so.conf.d/*.conf -- .''`. Aurelien Jarno | GPG: 1024D/F1BCDB73 : :' : Debian GNU/Linux developer | Electrical Engineer `. `' aurel32@debian.org | aurelien@aurel32.net `- people.debian.org/~aurel32 | www.aurel32.net