On Tue, Jun 03, 2008 at 04:17:04PM +0200, Alexander Vlasov wrote: > Alan Coopersmith wrote: > > Alexander Vlasov wrote: > > > >> Hello, > >> > >> dependency on file is TERRIBLY wrong idea. Anyone who maintained big > >> enough set of RPM-based linuxes can confirm this. > >> > > > > And for those of us fortunate enough to not have done this, can you > > explain why? > > > > Some reasons are already described in my mail to Michal (see the > thread), but I can extend some thought here: > Maintainer selects name for package knowingly. If he gives some name and > "provides" metadata to the package, he does it because he is sure this > package really does what it should. > Contrary, third-party developers are not aware of situation in the > distribution and can give their tools/libraries any name they want; > presuming their product satisfies some distribution requirements a bit > optimistic. And even more: sometimes it's possible to build bits from > source code in a different ways, and some variants will be ok for > particular task, some will not.
Well, that would be a problem even with FMRI-based dependencies unless we use full metadata with standardized values for dependency handling. (GTK+ built with the GNU toolchain, etc., etc.) I agree that file-based dependencies cannot replace true package version dependencies, but in cases like '/usr/lib/sendmail' and '/usr/bin/pkg-config' where the filenames are stable interfaces, it could be useful. This would depend on the requirements of individual packages, of course. Mutt might depend on /usr/lib/sendmail while the SpamAssasin milter plugin would depend on the sendmail package explicitly. It would be up to the package maintainer to choose the right alternative, but then isn't that always the case? Venky. _______________________________________________ pkg-discuss mailing list [email protected] http://mail.opensolaris.org/mailman/listinfo/pkg-discuss
