On Feb 21, 2005, Chris Adams <cmadams@xxxxxxxxxx> wrote: > Once upon a time, Alexandre Oliva <aoliva@xxxxxxxxxx> said: >> Think of it this way: would you push OOo to extras? Why not? If so, >> why insist on pushing other very, perhaps even more useful pieces of >> software there, before they have be regarded as actually part of the >> distro, not as second-class citizens? > How many Fedora users will use OOo vs. how many will use Java > development tools? That's not my point. My point is that people claim pushing to extras makes little to no difference to end users. If that *was* the case, why not pushing OOo there? It's a very big package, and nothing else depends on it. I'm not seriously advocating for us to do that, it's more of a thought experiment to point out the flaw in the argument of people who say `you can always get it from extras'. The point that OOo is a very big download offsets the higher costs of bandwidth and slower network connections in most of the world, so you can get an idea of how inconvenient it can be for most of the world to get to extras if it's not available in CDs like the core of the distro. The other annoyance with having to install packages from extras is that the install can have a significant number of drawbacks: - it's slower: anaconda/rpm can take a number of shortcuts when doing an install from scratch that they can't when the database is in a fully consistent state. - it pretty much requires network connection: you have to download packages, and unless you have them all already downloaded onto your box, and you have an alternate repo config to use them, you're out of luck if your network goes down - it requires far more storage: yum, up2date and all other depsolvers/installers first download all packages you need, then start installing them. So you need local space for the rpms plus space for the installed packages. Compare that with having the rpms in installable media, be it CD, DVD or a local NFS server. - longer down time: it requires at least one box to remain down while all packages that are not in the install media are downloaded, have deps solved and then installed, adding one more interactive step to the install, unlike the ideal model of `make all decisions, go', which means the box is fully operational as soon as it comes up. It means you can't start installing 50 boxes and go out to have dinner, you have to stay around until they get to the first boot, then do whatever amount of interaction it takes to get them functional, and by then restaurants will all be closed so you go straight home, hungry :-) The advantages? Can't really think of any. Mirrors will want to carry extras anyway. Not making them available as ISOs means they will get more requests for smaller files, which thrash their disks, and pushes people away from bittorrent et al. Users will be sad to find that their favorite apps are no longer available in the CD set they got in a magazine or on a store, and might not even realize they're still available in extras, *if* they have the network connection and the bandwidth to get to it. The DVD will be half-full instead of half-empty. Maybe that's a good one :-) FWIW, I don't have a DVD burner myself, and most if not all of my installs are over NFS; I have a reasonably fast and reliable network connection, but I happen to know the reality of most that live around. Just getting updates is a lot of trouble already. -- Alexandre Oliva http://www.ic.unicamp.br/~oliva/ Red Hat Compiler Engineer aoliva@{redhat.com, gcc.gnu.org} Free Software Evangelist oliva@{lsd.ic.unicamp.br, gnu.org}
