Kernel page, once again
Alexander E. Patrakov
patrakov at ums.usu.ru
Wed Jun 16 00:13:38 PDT 2004
(sorry for private mail, Jeremy told me to stop beating this dead horse)
Ken Moffat wrote:
> On Tue, 15 Jun 2004, Alexander E. Patrakov wrote:
>>1) Move all optional/replaceable packages to ch8. This currently
>>includes: e2fsprogs, reiserfsprogs (just for showing the reader that an
>>alternative exists), make_devices, udev, hotplug, module-init-tools, the
>>"modules" script, grub and lilo. Yes, I am strongly against the "one
>>true path" approach.
> As a believer in "build on as many platforms as possible" I have to
> support the general idea. However, I can see the following issues -
> (i.) Isn't make_devises | udev a choice to be made before building glibc
> in chapter 6 ? I don't understand how we can defer them to chapter 8.
Correct for the current book. But it still can be solved: instead of
using udevstart to populate the /dev directory before ch6, we can mount
--bind the host's /dev. In ch8 we can either install udev (and do
nothing more) or create the rest of devices with make_devices.sh
However, if ch8 is a bad place for choice, not everything is lost. My
idea was to concentrate all choice to be made to one chapter or one page.
> (ii.) since at least LFS-3 the various editors have been
> unable/unwilling to support alternatives. The more choice the book
> offers, the more it will allow people to select unworking options.
> Example - in my first play with BELFS I used lilo, got it building, but
> lilo fails to boot. Almost certainly an issue with the particular
> version of libc-headers that I used, but I haven't yet gone back and
> rebuilt everything. Options in the (released) book are only useful if
> people are going to test them before a new edition.
This is just an organizational question. Make two editors responsible
for testing each configuration. We have the potential to support three
or four configurations. This is sufficient because e.g. udev vs
make_devices choice is independent on lilo vs grub. So we can test the
and have a good reason to think that make_devices+lilo will work.
> (iii.) Gerard might regard this as bloat or too much hand-holding.
> e.g. reiserfs used to be in the book.
For reiserfs, a pointer is really sufficient because there are not many
traps. For udev+hotplug+modules, the situation is very different.
> (iv.) This also invites us to reopen the FSF/HJL debate, at least until
> the community makes the right decision :-)
I prefer not commenting on this one, except that I use HJL.
> FWIW, my view on architecture-specific things (and by extension, lilo
> and any other bootloaders where there is a choice) is that optional
> sections could be maintained alongside the book by people who use them,
> preferably with a note showing for which version of the book the
> instructions were last tested. Sounds suspiciously like a hint, in
> fact. I'd love to see the book reaching a point where a builder can
> check-out a custom version of the book for his/her hardware, but I don't
> think we're close to that at the moment.
The difference is that a hint cannot be converted to a PDF file.
> For filesystems, just continue to point people to other fs's, e.g. in
> BLFS. Maybe add reminders that clueless builders may wish to ensure
> their host's filesystem can be mounted, and can be understood by grub.
Alexander E. Patrakov
More information about the lfs-dev