On Jan 6, 2014, at 12:04 AM, Adam Williamson <awilliam@xxxxxxxxxx> wrote: > On Sun, 2014-01-05 at 22:52 -0800, Adam Williamson wrote: >> On Thu, 2014-01-02 at 23:32 +0100, Jean François Martinez wrote: >>> I have a nice booter setup and a nice _main_ Linux installation. Last >>> thing I would want is a distribution I am _testing_, that is Fedora 20 >>> forces on me it will be my main installation and forces me to choose >>> between installing Grub on the MBR or not at all. >> >> Why? "Not at all" is precisely the correct action for a grub2-based >> distribution in this case. >> >> I think we should do grub2-mkconfig for such installs, though it's a bit >> tricky to refactor anaconda's bootloader install code to do so. I might >> have a shot at it if I get a spare minute, though. > > Well, hum, it doesn't look hard at all, in fact, at least a simple > version. See attached patch (untested, but what could possibly go > wrong?!). I'm not subscribed to anaconda-patches-list ATM, but CCing > anaconda-devel-list to see what the anaconda devs think. I think it's > reasonable to install a config file and device.map when 'skipping' > grub2-bios bootloader install, especially given our standing advice to > people who want to do chainload-style multi boot is "skip bootloader > installation then setup configfile booting after install". There's an RFE bug floating around for this. I asked a while ago on grub-devel@ whether grub-mkconfig depended on grub-install being done first, and I got a kinda yesish answer that wasn't all that convincing. But the suggestion was to use 'grub2-install --grub-setup=/bin/true /dev/sdX' before grub2-mkconfig. This causes all the parts of grub to be installed except the singular thing people object to, and is the sole reason why they say they don't want a bootloader installed - which is boot.img being written into the MBR. I've tested that combination and it's safe. So if the code were changed just that "no bootloader" simply means *adding* --grub-setup=/bin/true and doing everything else we already do, that'd be the easiest fix that helps the most people IMO. BTW I think device.map is deprecated (?). Chris Murphy -- devel mailing list devel@xxxxxxxxxxxxxxxxxxxxxxx https://admin.fedoraproject.org/mailman/listinfo/devel Fedora Code of Conduct: http://fedoraproject.org/code-of-conduct
