-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA256 On 08/08/14 03:34 PM, Peter Stuge wrote: > Kent Fredric wrote: >> dependencies are forward specifications from upstream telling us >> what their software needs to function properly. > > Unfortunately that's not the full story. :\ > > ebuilds often (for me) have artificial dependencies, when the > actual version required is too old to be in the tree, but maybe not > too old to be installed on an existing system. > > I think it's bad policy to lie about dependencies in ebuilds for > the sole reason of only ever depending on versions which actually > exist in the same snapshot. It's a too simplistic model of > reality. >
For the most part I don't think that happens very often; usually if all stable versions of a dependency can satisfy a package's needs then there isn't any minimum version specification (or the minimum version specification hasn't actually been updated in an ebuild's *DEPEND, despite the older versions having been removed). The main exception to this is the work done related to gx86-multilib (as for obvious reasons the multilib ebuilds are needed to supply multilib dependencies), and the refactoring that mgorny did a few weeks back to fix the EAPI<5 USE_EXPAND+IUSE_IMPLICIT undefined behaviour issue. That said, even if dependency atoms allow the older, no-longer-in-tree versions to satisfy a package's needs, as I said earlier I don't think any dev has the time and resources to test against anything older than latest-stable, and definitely not anything that's no longer in the tree. -----BEGIN PGP SIGNATURE----- Version: GnuPG v2 iF4EAREIAAYFAlPlKW0ACgkQ2ugaI38ACPDAewD/ebk6WQa4pA4VJQkpiXf2Ch/R uGz0HRy6/Y17eSxDL3wA/2gD8ciNsVWkIV6/kLGwwVXGItLL07A3OXITGLE1U8+N =EBWi -----END PGP SIGNATURE-----
