On 17-02-13 21:54, Graham Inggs wrote: > I finally see what you were hinting at. The missing piece was above. But > indeed, because you can not guarantee that all three will be installed > if somebody installs a package that requires libmotif3/libmotif4, I > don't think this is the right solution. Providing a virtual package > (lets not call it transitional then) is I think better. > > > I am open suggestions. But (on Ubuntu at least) if you install a > package, all recommended packages are pulled in as well. In Debian as well, by default. But people are allowed to override this behavior, also on Ubuntu. The policy [1] is clear however: Recommends This declares a strong, but not absolute, dependency. The Recommends field should list packages that would be found together with this one in all but unusual installations. So, if you want to be sure that a package that DEPENDS on libmotif3/4 has its dependency fulfilled, we need a package that provides ALL dependencies, i.e. an (empty) libmotif4/3 package. > I want compatibility with commercial packages that need libXm.so.3. > Seeing that we provide these symlinks, we may as well advertise that we > provide libmotif3 compatibility. I thought so. Sure, no problem. > No. If he/she absolutely needs libmrm4 or libuil4 to run his/her foo-bin > package, he should get it no matter what. So libmotif3/libmotif4 should > absolutely pull that in or fail. Therefore, a virtual libmotif4 package > that provides libmotif3. I propose we remove the sentence "...can safely > be removed..." This problem will go away soon, as packages will depend > on the proper library directly in the future. > > As above, when installing a package, all recommended packages are pull > in as well. In most cases, but not guaranteed, also not on Ubuntu. Paul [1] http://www.debian.org/doc/debian-policy/ch-relationships.html#s-binarydeps
Attachment:
signature.asc
Description: OpenPGP digital signature