On Wed, May 22, 2013 at 06:13:55PM -0700, Adam Williamson wrote: > > Can you elaborate on those? (Or point me to something; I tried to Google > > but only found you saying a similar thing in a different post.) > dgilmore could do a better job of it, but basically, the way lmc was > initially designed was pretty much fundamentally incompatible with the > way releng builds live images. Ah. Primarily, building in a VM as opposed to a chroot. Secondly, the image preparation would run on the _host_ system's OS, rather than the to-be-installed one, so fixes to the image generation would have to happen at that level rather than in the tree. (The installer itself matches the installation, but the infrastructure firing up the VM and processing the output might not.) There _are_ issues to solve but I think it pretty much has to be the way forward, because building in a chroot, _especially_ with a tool other than anaconda itself, has its own set of issues. Using something like livemedia-creator or oz has enough advantages that it's worth spending the effort to fix the difficulties. Jay Greguske is working on integrating this into Koji, but, yeah, too much work to make F19. (Initially for cloud images, but eventually doing live images the same way.) Some kind of combined approach might be possible. (Especially when we have nested KVM on the builders....) I was just wondering if there were other potential pitfalls to be aware of. -- Matthew Miller ☁☁☁ Fedora Cloud Architect ☁☁☁ <mattdm@xxxxxxxxxxxxxxxxx> -- test mailing list test@xxxxxxxxxxxxxxxxxxxxxxx To unsubscribe: https://admin.fedoraproject.org/mailman/listinfo/test
