On Tue, 30 Nov 2004 12:00:27 -0800, Mike Ramirez <mike@xxxxxxxxxxxxxx> wrote: > On Tue, 2004-11-30 at 11:10, Jeff Spaleta wrote: > > In > > fact, i believe such testing is a waste of limited resources that > > could be used for other purposes. > > > Hi Jeff > > I think you answered my why question and can agree with you an almost > everything said. This here I think is wrong. This testing is necessary > or we will always have this same problem. I think a compromise must be > made in this area to have cross compatibility between all repos for a > better user experience. All repos? ALL REPOS? thats a very very big number. You realize that individual red hat developers have their own repos at people.redhat.com used for a variety of reasons...hell even projects at sf.net have package repositories that some users eat packages from. You must count those micro repos in that ALL REPOS matrix. To test against ALL REPOS its an absurd waste of time. My answer is to stop using the outside repos for as many packages as possible. "Established" is a misnomer, and suggests the only repositories worth doing regression testing against are a small handful of repos. If the number of overlapping "established" repos grows over time the testing matrix grows exponentially harder to manage... this does not scale. This system either throws up barriers to entry to new packagers with new repositories...or it breaks down continually into smaller groups. 100 repositories will not be able to cross-communicate as affectively as 10. 10 repositories will not be able to cross-communicate as effectively as 5. and so on. What scales... is moving more and more packages into the SAME build system into a non-overlapping tree, and encouraging users to use packages from that build system. My sincere hope is that this project and this community will move towards a situations where you have 2 centralized build systems. One build system managed by Red Hat and containing everthing Fedora policy can legally distributed. A second non-US build system maintained by community for packages that can not be re-distributed by red hat. In my perfect future, Red Hat will be able to provided a blueprint and tools for the non-US community buildsystem to replicate using community donations for hardware and community volunteers to staff the system with manpower. In my perfect future the non-US buildsystem builds on top of the Red Hat build system but does not overlap. In my perfect future both build systems follow very closely to the current Core development model, where much of the energy is focused on a development "truck" and individual packagers in the system have the freedom to create experiment sidebar repositories non unlike what we see in people.redhat.com today. > > Otherwise it will stay as it is. Is this a good thing? or a bad > thing? I think this thread has pointed out that its bad and needs to be > fixed. It does need to be fixed... i would fix it by no longer having multiple 3rd party repositories. I would fix it... by establishing a common buildsystem and common distribution tree. If you don't have multiple repositories aimed at general consumption... then you don't have repository conflicts. > asking again. How can we get interoperability between ALL repos or is > that just impossible now? ALL repos... is impossible and will continue to be impossible. There are hundreds of small repositories out there. Do you really think you can get hundreds of individual repositories all working together? -jef
