Control: severity -1 normal

On Sun, Jun 18, 2017 at 10:49:15PM +0200, Andreas Heinlein wrote:
> The meta-package grub-efi depends on either grub-efi-amd64 if
> installation architecture is amd64, or grub-efi-ia32 if installed on i386.

To persuade me that this is important, you'll need to first persuade me
that the grub-efi binary package actually matters.  As far as I'm
concerned, it exists only for upgrade purposes (note that it's in
Section: oldlibs).  Although d-i does use it as a fallback, it generally
tries to work out the firmware architecture instead.

> So either need to determine EFI architecture at install time and then
> pull in the correct package (hard way),

Not so hard considering that the installer has code to do this already!

I think we should probably change d-i to stop expecting grub-efi to
exist even as a fallback, and instead fall back to the userspace
architecture (of course not ideal, but good enough as a fallback given
that it will look at /sys/firmware/efi/fw_platform_size first).  We can
then simply drop grub-efi entirely rather than debating what its
dependencies should be.

> or correct the dependencies and depend on *both* grub-efi-amd64 and
> grub-efi-ia32, regardless of installation architecture (easy way). I
> did the latter for tests on our machines and found no problems so far
> with both of them co-existing.

I don't think you can have tested that very extensively, since
grub-efi-amd64 and grub-efi-ia32 declare Conflicts on each other.  In
general, multiple grub-<platform>-bin packages are coinstallable, but
multiple grub-<platform> packages are not; this is intentional because
the semantics of grub-<platform> are to assert ownership of the system's
boot process.


Colin Watson                                       []

Reply via email to