From gentoo-dev Thu Apr 25 23:15:32 2013 From: Ambroz Bizjak Date: Thu, 25 Apr 2013 23:15:32 +0000 To: gentoo-dev Subject: Re: [gentoo-dev] Re: rfc: oldnet scripts splitting out from OpenRC Message-Id: X-MARC-Message: https://marc.info/?l=gentoo-dev&m=136693174424722 I *really* hate those virtual dependencies that don't actually satisfy a real dependency, and require manual choice-specific intervention by the user anyway. For example, packages that build external kernel modules tend to depend on virtual/kernel-sources. However, this dependency doesn't make sure that a kernel is built and hence a module can be built against it. The only thing it does is get in the way, like when someone gets a kernel source outside of the package manager, and has to use package.provided to make the package manager happy. A virtual/network-provider package would have exactly the same problem. You'd have to find all packages in portage which can serve as network managers, and even after you do that, someone will be annoyed because he uses something else. Masking unwanted network managers may even be needed to work around intricacies in dependency resolution. I think that a virtual only makes sense when there's some degree of interchangeability, for example, a virtual for all network managers which understand the standard Gentoo network configuration files. But not a virtual for some generic service. We don't have a virtual for an HTTP server or desktop environment, either (AFAIK). On Fri, Apr 26, 2013 at 12:51 AM, Carlos Silva wrote: > On Thu, Apr 25, 2013 at 10:15 PM, Duncan <1i5t5.duncan@cox.net> wrote: >> >> > > > Maybe I was miss understood here. I know that there are tons of ways to have > gentoo *running* in a box without it having network connection. The thing is > that makes like 0.01% of the total installs. It's not a default install, it > isn't on any gentoo manual I know of (besides the chroot one, but I really > don't consider that an installation), and most importantly, AFAIK, it's not > something any John Doe would do. Offline installations and "runtimes" are > for geeks that use linux for a long time and know how the system work and > have the knowledge to build a stage4 or chroot and move it to another box. > It's not something technically difficult for us "geeks", but would take ages > for some non-geek to do it. > Hell, a friend of mine normally calls me when he needs to do something to > his box other that "pacman " (yeah, he's on arch) and he's using > linux for some time now. > > The bottom line here is, does @system have to have virtual/network-provider? > - Yes -> Make it RDEPEND; > - No -> don't care and just set some use flags. > > The question above is more a political one than technical. Everyone here > knows that a system doesn't have to have networking support for it to boot, > we can't even guarantee that networking support is in the kernel (at least I > don't see it using kernel-*.eclass), but is it a safe default meaning that > 99% or more of the people will use or *need* it? <--- political > > Sorry if I was too long on this :)