On Sun, Oct 06, 2002 at 01:56:04PM +1000, Anthony Towns wrote: > On Sun, Oct 06, 2002 at 02:56:31AM +0100, Andrew Suffield wrote: > > On Tue, Oct 01, 2002 at 01:25:17PM +0200, Erich Schubert wrote: > > > > I don't really see any reason why the same approach that is used for > > > > shared library ABIs will not also work for any other ABI. > > > Compatibility with other distributions. > > Is nonexistant when it comes to package names. > > It's not the package names -- it's the binary names, and the names > of the config files. Allowing, say, mozilla M16 and mozilla 1.0 to be > installed concurrently and both to work means having one of them not be > called mozilla and having one of them not use ~/.mozilla. I would expect the sysadmin (perhaps assisted by something like update-alternatives, but different. haven't thought about that part much) to maintain a /usr/bin/mozilla symlink to the current default version; they can then perform a controlled migration at their leisure. As for dotfiles, if the application is not going to upgrade their format on the fly, then yes, I'd expect them to be named differently. [Of course, it would be good if upstream did this in the first place]. > > Whoever said that you shouldn't version the name of the first package? > > ABI versioning breaks utterly if you have *any* unversioned entities > > in the available pool. > > Actually, it doesn't. "" is quite usable as an ABI version, as long as > you only use it once, just like any other ABI version. The problem being that we conventionally expect "" to be the virtual form or symlink, for when you don't want to know about the version (we don't use that in package names because you always care about that version for shared libraries, and we do use it in .so filenames because gcc does not care). -- .''`. ** Debian GNU/Linux ** | Andrew Suffield : :' : http://www.debian.org/ | Dept. of Computing, `. `' | Imperial College, `- -><- | London, UK
Attachment:
pgpPrEeBtVLDU.pgp
Description: PGP signature