On Jul 1, 2015, at 6:45 AM, Per Øyvind Karlsen wrote: > 2015-06-28 22:24 GMT+02:00 Jeffrey Johnson <[email protected]>: > > > %ghost has many problems and is poorly designed. >
This is a fair statement of my current opinion(s) regarding %ghost. Meanwhile ... > If %ghost for the same path is present in more than one package, > that is already a packaging problem no different than any other > file conflict. Choose one package to use %ghost, remove the other. > > If you don't have criteria for choosing, flip a coin: it really doesn't > matter. > You don't think current existing behaviour of rpm, allowing %ghost files > packaged in multiplie packages to be installed without conflicting as the > right behaviour? > I described the existing behavior of rpm, which detected a file conflict between a path contained in 2 packages for Tomasz, and responded to a bug report about different behavior on 2 architectures that were implicitly claimed to be identical. > As I see the most correct behaviour would be, is for several packages to be > installed with same %ghost files without conflicting, while during erase of > packages, whenever last package with ownership of the %ghost file gets > uninstalled, the %ghost file should be removed with it.. > As you know, "last" is de facto and indeterminate both because of "partial ordering", as well as whatever heuristics might be present to remove dependency LOOP's. Computing "last" is quite expensive too: the %ghost bit must be compared to every other path in every other package, that is a very expensive calculation for a rather minor semantic. The %ghost "ignore all but the last" semantic you describe ends up adding an additional constraint to package ordering as well. > I'm a bit perplexed about potentially different opinions and implementation > behaviours around there otherwise as well, so I'm CC'ing this to > rpm-ecosystem@ for broader input... > Perhaps you should respond to rpm "bugs" in OMA sooner instead of inventing features to discuss and implement. 73 de Jeff
_______________________________________________ OM-Cooker mailing list [email protected] http://ml.openmandriva.org/listinfo.cgi/om-cooker-openmandriva.org
