Dieter Plaetinck wrote:
On Sun, 01 Feb 2009 01:52:52 +1000 Allan McRae <allan@archlinux.org> wrote:
Gerardo Exequiel Pozzi wrote:
Allan McRae wrote:
Gerhard Brauer wrote:
Am Sat, 31 Jan 2009 20:23:34 +1000 schrieb Allan McRae <allan@archlinux.org>:
Would it be a good idea to move grub-gfx to [extra] if it is being used on the installer?
Hmm, abstain... First i wonder that we use a community package on/for installation sources, but it's only the grub-gfx package. So i have no problem belong this to community at all. (I also thought about maintaining a separate grub-gfx only for archiso...). I see the problem mainly on our testing procedure: this behavior must have be detected earlier - and not from a "beta tester"(Thanks Gerardo!), WE have had to detect this.... (I detect it on i686 and initiate that this got fixed. Myself (only i686) don't look if x86_64 was rebuild also....
Can someone clarify what package the grub being installed onto the users system is? From the problems here, I am getting the impression that it is the grub-gfx package.
Allan
Hi
In resume: * The package installed at: the root-fs of ISO is grub-gfx. the root of new user system $DESTDIR is grub, but... the MBR/BOOT of the new system is grub-gfx because $DESTDIR/sbin/grub-install (script) uses /sbin/grub (bin) and not $DESTDIR/sbin/grub (bin)
* Under x86_64 grub-gfx fails in ext2/3/4 FS since don't have latest patches that the i686 version have.
May be, a chroot to $DESTDIR and install grub from here with grub-install, or use the old method from 2008.06 that uses grub binary directly. Or another solution.
Thanks for the clarification. I was concerned about grub-gfx being installed to peoples systems by default but had not realized that this was because of a bug.
Allan
So do I understand correctly...
*problem 1*: grub-gfx vs grub: We can fix the problem that grub-gfx instead of grub is installed if we just chroot I think. Eg: chroot $DESTDIR /sbin/grub-install --recheck $ROOTDEV >/tmp/grub.log 2>&1
But, this only affects 64bit right? or not? (because in my tests the grub didn't look very fancy..) If so, can someone with a 64bit (virtual) machine change the above line in /arch/setup and test it? If okay, someone with access to archlinux-installer can fix the line and repackage.
If I understand correctly, this problem is unrelated to the "slightly outdated grub package" problem ( see http://bugs.archlinux.org/task/13068 ), eg if it were not for problem 2, we wouldn't need to update the package.
*problem 2*: Gerardo says "grub-gfx fails", this is the same as Alexanders problem, right? ("I'm afraid I confirm the grub bug. In x86_64 images grub isn't installed to the MBR of the chosen disk (when the boot partition is an extX, if it is reiserfs it works perfectly)").
IIRC Gerardo already tried the new package , tested it and confirmed that it works. So we just need to get it on the iso. See http://bugs.archlinux.org/task/13068
Dieter
Hi, OK Now the problem is solved, i see a grub-install parameter useful :) --grub-shell=FILE, with this override the use of the problematic /sbin/grub. 1) No chroot is needed. 2) Now the correct grub is installed at MBR/BOOT not the grub-gfx ;) Patch is attached -- Gerardo Exequiel Pozzi ( djgera ) http://www.djgera.com.ar KeyID: 0x1B8C330D Key fingerprint = 0CAA D5D4 CD85 4434 A219 76ED 39AB 221B 1B8C 330D --- setup.orig 2009-01-31 18:12:15.000000000 -0200 +++ setup 2009-01-31 18:12:51.000000000 -0200 @@ -1081,7 +1081,7 @@ DIALOG --msgbox "GRUB root and setup devices could not be auto-located. You will need to manually run the GRUB shell to install a bootloader." 0 0 return 1 fi - $DESTDIR/sbin/grub-install --recheck --root-directory=$DESTDIR $ROOTDEV >/tmp/grub.log 2>&1 + $DESTDIR/sbin/grub-install --recheck --grub-shell=$DESTDIR/sbin/grub --root-directory=$DESTDIR $ROOTDEV >/tmp/grub.log 2>&1 cat /tmp/grub.log >$LOG # unfreeze xfs filesystems if [ -x /usr/sbin/xfs_freeze ]; then