Comment 16 for bug 185878

Regardless of GRUB 2 (which cannot be assumed to fix this bug without checking; installation into a partition doesn't seem to be a particularly high-priority feature for GRUB upstream, from what I can tell), we should analyse and fix this one.

I've been looking into this today. It seems to be a race; running grub-install twice in succession, as previously observed, works fine, and one can reproduce the bug again just by zeroing out the region on disk occupied by stage 1.5. I think that pretty clearly makes this a grub bug rather than a grub-installer bug. I have some ideas on how to fix that and will try them out shortly.

specialk: Race conditions can be a bit like this; sometimes you just get lucky. I don't think anything specific was changed to break this, but some of the timings may have changed a bit. Of course, installing to the MBR or to some different partition will work fine even if /boot is on XFS.