On Sun, 23 Sep 2007 22:28:34 +0300 "Ville Skyttä" <ville skytta iki fi> wrote: > I tend to agree. Is the strategy documented somewhere so one can > educate oneself and verify that this indeed is the desired behaviour > that follows from it? I'm having hard time coming up with any > arguments that would support the way things currently work, but would > be interested in finding out if I've missed something. Jeremy was working on documenting it. It is a strategy that has been developed over a period of releases that offered multilib. Quite simply, we want both runtime library and development package availability of the secondary arch installed by default. To accomplish this we use the existence of a -devel subpackage to trigger selection as a multilib package. We then require the package that holds the symlink destination of any .so files in the -devel subpackage. We then depsolve that to make sure all the secondary arch deps are resolved. There are a few more things going on, a few whitelists, a bare minimum blacklist. http://git.fedoraproject.org/?p=hosted/mash;a=blob_plain;f=mash/multilib.py;hb=HEAD has the gory details. This is the method by which we select which packages are multilib (finally a programatic way instead of a hand generated list we had before). The second part of the strategy was that tools like yum will install the available packages by default if simply asked to 'install foo' and "foo" wasn't arch specified. Yes, some of us would like to change this around a bit in Fedora 9, by making it configurable client side which strategy to follow, and to determine at install time which compat arch packages to pull in based on configured strategy. This would eliminate the need to pre-populate the repos with secondary arch content at compose time. -- Jesse Keating Fedora -- All my bits are free, are yours?
Description: PGP signature