First, I'm sorry that I was absent for about one week (but, to say
honestly, I'm wondering why the development of GNU GRUB depends on me
so much).
Second, much more importantly, I noticed that there was a better
approach for grub-install than writing an additional C
program. That is, instead of implementing the install procedure
separately, run the grub shell in read-only mode and let it output
information necessary for the installation, then a shell script
interprets the information and writes down data to a disk actually. If
we can define the format of the information in a general way,
grub-install will not depend on the internal very much.
The operations required for the installation are essentially only
two types: copy and modification. So I think of two ways:
1) The grub shell prints out a list of "copy" and "modify" commands,
like this:
copy stage1 (hd0)0+1
copy e2fs_stage1_5 (hd0)1+15
modify (hd0)1+1 123456789ABCD
2) The grub shell puts modified images (i.e. stage1 and stage2 (or
stage1.5)) temporarily somewhere, and just tells grub-install to copy
them to appropriate locations.
At the moment, I prefer the latter, since it is easy to
implement. But the former may be superior, because changing the
installation scheme wouldn't affect grub-install at all. Any comments?
Okuji
