Draft bootloader selection test cases

[Chris Murphy]

On Aug 2, 2013, at 7:20 PM, Adam Williamson <awilliam at redhat.com> wrote:
> he feedback. I didn't want to get into too much detail in
> the test case, because things can be fairly different on different
> arches, BIOS vs. UEFI etc. But it might be a good idea to include some
> examples for the most common cases, I guess.

Isn't "no bootloader" meaningless on UEFI? It would mean that anaconda would have to not install the grub-efi or shim packages; grub-install isn't called in UEFI, the grubx64.efi bootloader file (and other files) are installed in the course of installing shim and grub. I could be mistaken but my vague recollection is that for F18 the no bootloader option in the installer for UEFI hardware was considered a bug, which was fixed in F19. I don't think the option exists on UEFI.

As for how to verify the lack of a bootloader, that's not what we want. We'd have to capture the first 440 bytes of LBA 0 before and after installation, and compare - if the "don't install bootloader" option works, they should be the same. That bootstrap region is almost certainly not empty, even parted drops some small bit of code in there when an MBR is created; and of course it's filled if the syslinux mbr.bin has been installed there, or the Windows boot loader.

dd if=/dev/sdX bs=440 count=1 | hexdump -C

> 
> I'm not sure that actually _is_ suboptimal, because if you're not
> installing a bootloader, there's really nothing at all the installer
> does to render the installation 'bootable'; there's nothing to test. It
> just installs the packages. It's up to the user to configure boot. So
> I'm not sure there's something it makes any sense to test, at that
> point.

It could be argued the test is to make certain the first 440 bytes of LBA 0 haven't been modified. That's really the ultimate metric as far as I can tell. It doesn't matter if anything else is installed.


Chris Murphy