Bug#739581: update-grub: fails at 30_init-select, no Windows entries in grub.cfg

[Colin Watson]

Control: reassign -1 init-select
Control: severity -1 grave

On Wed, Feb 19, 2014 at 09:40:44PM -0500, Ron Murray wrote:
> Since the last grub update in testing, update-grub appears to fail
> during 30_init-select and doesn't include Windows partitions in
> grub.cfg, although it sees them:
> 
> > khufu:~# update-grub
> > Generating grub.cfg ...
> > Found background image: /usr/share/images/desktop-base/desktop-grub.png
> > Found linux image: /boot/vmlinuz-3.13.3-khufu-2
> > Found initrd image: /boot/initrd.img-3.13.3-khufu-2
> > Found linux image: /boot/vmlinuz-3.13.2-khufu-0
> > Found initrd image: /boot/initrd.img-3.13.2-khufu-0
> > Found memtest86+ image: /memtest86+.bin
> > Found memtest86+ multiboot image: /memtest86+_multiboot.bin
> > Found GRUB Invaders image: /boot/invaders.exec
> > Found Windows 8 (loader) on /dev/sda1
> > Found Windows 8 (loader) on /dev/sdc1
> > done
> 
> Last lines of grub.cfg, however, come out as:
> 
> > ### BEGIN /etc/grub.d/22_invaders ###
> > menuentry "GRUB Invaders" {
> > 	insmod part_msdos
> > 	insmod xfs
> > 	set root='hd1,msdos1'
> > 	if [ x$feature_platform_search_hint = xy ]; then
> > 	  search --no-floppy --fs-uuid --set=root --hint-bios=hd1,msdos1 --hint-efi=hd1,msdos1 --hint-baremetal=ahci1,msdos1 --hint='hd1,msdos1'  47193b17-a99a-4a83-8818-fc54982e3996
> > 	else
> > 	  search --no-floppy --fs-uuid --set=root 47193b17-a99a-4a83-8818-fc54982e3996
> > 	fi
> > 	multiboot	/invaders.exec
> > }
> > ### END /etc/grub.d/22_invaders ###
> > 
> > ### BEGIN /etc/grub.d/30_init-select ###

Thanks for your report.  /etc/grub.d/30_init-select is provided by the
init-select package, not by grub2 itself.  Reassigning.

Even under the best circumstances where all prior output has been
flushed (not, I think, guaranteed), the effect of the "sed -i" in
init-select is going to be to break grub-mkconfig's open redirection to
/boot/grub/grub.cfg.new, which is probably what's going wrong here; that
will then break all output from later scripts, hence the grave severity.
init-select should really be extending the kernel arguments in an
/etc/default/grub.d/ file instead, as I think I've advised before; I
don't think its current strategy is ever going to work reliably.

Cheers,

-- 
Colin Watson                                       [cjwatson at debian.org]