/dev/sda1 is a windows partition to be mounted as /windows /dev/sda2 is an ext3 partition to be formatted and mounted as / /dev/sda3 is a swap /dev/sda4 is a fat32 partition to be mounted as /dos Graphics installer failed with an error about start being after the end. I then tried rerunning the command line installer, which fails at the roughly the same point. Attaching the install log and profile.
Created attachment 82843 [details] Install profile used
Created attachment 82844 [details] Install log
I should add that all partitions already existed and I made no changes to their sizes. I had another Linux distribution installed previously on sda2, which is why I wanted to format that partition first. So losing it is not the problem, installer not being able to proceed past that point is.
That error is from the second run. The first run didn't recreate all the old partitions (because it failed), so the second run couldn't find one of the old ones. We need the error from the first run to help you.
I can still mount all the partitions in LiveCD. Windows and DOS partitions are untouched. Linux partition has been wiped out, but it *is* mountable. So my immediate concern is why can't the installer proceed? The error message is not very descriptive, but I was hoping it might mean something to you. I will try manually running fsck and mkfs on the Linux partition, and tell the installer not to format it. Would you have any additional advice?
One correction: after restarting, I found that GRUB is hosed, and the DOS partition is gone. So the system is currently unbootable. I'm not sure what happened to that DOS partition, as it was mounted and accessible after the failed install.
Did you have it mounted when you ran the installer?
There is absolutely nothing I can do without more information. If you actually can provide me with more information, reopen this bug.
I was able to recreate this problem. The setup is slightly different than before, but the result is the same: windows partition is OK, DOS partition is removed from the table. I have been able to use fdisk to recreate the exact same layout and rescue the partition, but I'm not sure if this can always be done. Anyways, I'm attaching the installer profile and the log.
Created attachment 82978 [details] Install profile used (updated)
Created attachment 82979 [details] Install log (updated)
Reopening.
Ah, I see the problem. The partitioning code in the current version of the installer cannot handle a primary partition after an extended. The partitioning code in CVS should be able to handle it without a problem. I'd recommend using CVS, but I'm in the middle of rewriting the partitioning code, so it's not guaranteed to be working at any given time. Anyway, the problem is (or will be very soon) fixed.
That's one part of the condition that triggers the bug. The second is that is some (all Linux?) partitions are set to be formatted during the install. I was able to install using the LiveCD with this same install profile when I ran mkfs myself from the command line and told the installer not to format those partitions. Is there a way to mention this prominently somewhere in the online documentation for installing 2006.0? Two more observations, I'm not sure if they're important, but may be relevant to your fix. One, whenever this bug occured, the extended partition's type would be changed from 85 (Linux extended) to 5 (plain extended), I'm not sure why. Two, writing the partition table may have been unneccessary in the first place, because all I asked the installer to do was format the Linux partition(s) - they were already created before it was run by using fdisk. It seems like it would have been sufficient for the installer to have run mkfs, no?
Choosing not to format them should not have made a difference. While it would make sense to not touch the partitions and to just mkfs the ones that need it in a case where no existing partitions are being removed/resized, it would be messy to short circuit all that partitioning and jump straight to the mkfs'ing code. Although, it may be possible when I break up the partitioning code into separate functions as part of my current rewrite.
(In reply to comment #15) > Choosing not to format them should not have made a difference. Well, it did, thankfully. Or otherwise I would have a useless computer right now and be downloading Ubuntu to get it back to life. > While it would > make sense to not touch the partitions and to just mkfs the ones that need it > in a case where no existing partitions are being removed/resized, it would be > messy to short circuit all that partitioning and jump straight to the mkfs'ing > code. You'd know the issues involved far better than me, but what's messy about a single check? When the user has all these existing partitions that he wants to keep and just plop Gentoo into the one(s) allocated for Linux, as I did, what partitioning needs to be done at all? Especially when you have a user who's switching from another Linux distribution or installing a dual-boot system -- losing your existing data is not particularly fun, even when you do have a backup. I'm new to Gentoo, could you tell me how I could contribute to the documentation (either the handbook or the wiki) to mention this issue with an extended partition not in the last slot in the partition list?
Moving to Release Media/Installer.
