Gilboa Davara wrote:
*AND* because the fedora distribution does nothing to assist their users in installing the driver that Nvidia makes freely available. The party-line argument that third party drivers cause support problems kind of falls on its face when the included driver doesn't work at all...
Oh come-on! Not that argument again! Ignoring for the second the GPL problem (so called party-line), which is the basis on which... err... Linux is built: Please solve the following problems:
It is more a question of _who_ should deal with the problems created by the inability or refusal of the kernel developers to define a driver interface. In the 'enterprise' release the distro packagers handle it and the users win. In fedora the users lose.
Case 1: (Happens once every 6 months or so) 1. nVidia tends to release their driver on a 4 month - 1 month schedule. (Major each 4 months, with a minor bug fix after a one month) 2. Fedora kernel people are pushing a new upstream kernel that is incompatible with the current nVidia release - while nVidia is during the their major-update-wait. ... Which in turn: A. Fedora should not push a new kernel update because nVidia has yet to support it. B. Fedora should push a new kernel update, getting hammered by users that have broken system on their hands.
Users lose either of these ways. The first question is why does anyone who has a working kernel and device drivers _ever_ need to install a behavior-changing replacement - and particularly within an FCx release where the next one is only months away? If there is an answer to that question, then the next one becomes how to make yum understand kernel module dependencies and schedule kernel updates only after all needed modules (i.e. currently loaded in your kernel) are available in the repositories so you at least have fewer unpleasant surprises.
Case 2: 1. Fedora is about to release new major release (Fx) which uses a kernel which is completely incompatible with the current nVidia driver. 2. nVidia, due to Vista driver development problems, decides to spend less resources on Linux driver development.
Or, perhaps being snubbed and excluded by popular Linux distros is what makes them decide to reallocate those resources...
... Which in turn: A. Fedora should use the old Fedora Fx-1 kernel just to maintain compatibility with nVidia - until nVidia releases a new driver. (If and when) B. Fedora should release the new kernel with Fx, breaking support for the nVidia driver. (Which is officially supported - following your advice) Both Case 0 and Case 1 happened before. (8K stacks, X.org 7.x, 2.5.x kernels, a couple of 2.6.xx releases, Xen, etc)
Again, proper dependencies in the packaging system should permit such changes to propagate at the speed that the slowest needed component becomes available.
While I use the nVidia driver on a daily basis on a number of i386/x86_64 machines (and I have the utmost respect for nVidia's driver development team for producing close-to-rock-solid-drivers *), Fedora should not officially support, nor for that matter should they even care about breaking my setup when it's time to push a new update. (Kernel, X.org, etc) It's my choice to use binary-only out-of-the-tree drivers and it's not Fedora's problem if things break because of it.
Perhaps - but things are different now that users have other choices. I think it's fedora's - and eventually RH's problem if users are forced to a distro that uses different packaging and admin techniques to get a usable desktop. Fedora tends to be usable in this respect at or near the end of its life cycle, but then quickly becomes unusable due to the ending of security updates.
-- Les Mikesell lesmikesell@xxxxxxxxx
