On Tue, 18 Sep 2012 12:58:30 -0700 Zac Medico wrote: > On 09/18/2012 12:44 PM, Ciaran McCreesh wrote: > > On Tue, 18 Sep 2012 12:40:51 -0700 > > Zac Medico wrote: > >> On 09/18/2012 12:29 PM, Ciaran McCreesh wrote: > >>> On Tue, 18 Sep 2012 12:25:57 -0700 > >>> Zac Medico wrote: > >>>> Also, if we change the meaning of RDEPEND in the next EAPI, so > >>>> that it's a hard build-time dep like DEPEND, then > >>>> DEPEND="${RDEPEND} virtual/pkgconfig" can be reduced to > >>>> DEPEND="virtual/pkgconfig". This is what I would like to do for > >>>> the experimental EAPI 5-hdepend which is planned [1]. > >>> > >>> What're we going to do about the zillions of unsolvable cycles > >>> that that would create? (Does Portage detect those and error out > >>> yet?) > >> > >> Yeah, it would be treated just like a DEPEND cycle, which is > >> already detected and treated as a fatal error. As a result, when > >> bumping the EAPI of an ebuild, you may have to migrate some deps > >> from RDEPEND to PDEPEND in order to solve the cycles. > > > > What about the large number of RDEPENDs that are required for a > > package to be usable, but not for it to be installed? > > You will have to migrate those deps from RDEPEND to PDEPEND. ...but PDEPENDs aren't guaranteed to be installed before a package is used to satisfy a dependency. (And we can't change PDEPEND to do what RDEPEND currently does, because then some cycles can't be solved at all.) -- Ciaran McCreesh