gb spam wrote:
On 3/10/06, Rahul Sundaram <sundaram@xxxxxxxxxxxxxxxxx> wrote:
and still there is no quantification of "high". if i said that not
having an everything install button has a high cost, would you accept
that if i just continued to wave my hands in the air? (please don't
take this as abuse or anger, i just have trouble accepting something
when a question is not answered several times)
Again, I did very specifically answer this question in various occasions
in the earlier discussions. I cannot quantify the exact amount of code
required to reimplement this beyond the assertion that was it is non
trivial as communicated to me by Anaconda developers. What I can tell
you is instances where I have been involved where users had problems
trying to recover from a everything installation because they installed
a whole lot of packages unnecessarily.
Combine that
with the effect of this "feature" and the time spend in fixing and hand
holding new users who choose to use this option innocently
If there are problems merely because multiple packages are installed
together, that is entirely another problem, which should _also_ be
dealt with. we shouldn't be trying to hide from those problems. on
the contrary, we should be looking for them. if one problem comes up
frequently (like the oft-cited gfs kernel) then it quickly becomes a
known problem, work arounds are made public, mailing lists are well
aware of the problem and can direct people experiencing them in the
right direction and hopefully the problems also get resolved too. the
net result being that the project wins in the long run. this is
actually the direction we should be aiming for, not shirking away
from.
In many cases, for example with multiple mail servers, you just cannot
run them together. So this is not a flaw at all and there is nothing to
fix. The problem with earlier definitions of user profiles - personal
desktop, workstation, server etc
(http://shots.osdir.com/slideshows/slideshow.php?release=335&slide=7) is
that users didnt fit this precisely and choose custom. There is a effort
in resolving that through a new package selection design
(http://people.redhat.com/~katzj/tasksel.png). While it makes sense to
install everything (using kickstart or yum ) and test them if we can
during the test/development phase rigorously and testers should be
focusing on testing applications that they use, it isnt something we
should be encouraging otherwise for regular end users. The problem with
the GFS and other kernel modules was already known and well documented
in the release notes
(http://fedora.redhat.com/docs/release-notes/fc4/errata/#sn-kernel) and
something we have been trying to resolve with great effort
(http://fedoraproject.org/wiki/Extras/KernelModuleProposal) but this
problem wouldnt be there for users who didnt choose the everything
installation despite having no need for a storage cluster. So in the
longer run, there are discussions happening on reorganizing the packages
in core to meet the needs better
(http://fedoraproject.org/wiki/CoreBrainStorm) and then installing every
package in the ISO images would be much better. Fedora Core 5 Anaconda
uses yum as a dependency resolver and the next step would be support yum
repositories like Fedora Extras during installation and consider moving
more of the packages there. In the longer run (which might be as soon as
Fedora Core 6( our arguments are actually converging and installing
everything in the Fedora Core would be good if the images contained just
the default packages. So dont scream at me just because I am trying to
communicate the understanding that we have better.
--
Rahul
