Re: Merging / and /usr (was: jessie release goals)
Marco d'Itri <md <at> Linux.IT> writes:
> People use live CDs for rescue all the time, do you have some data which
> show that this is actually a problem in real life and not an imaginary
I’ve had Knoppix destroy the nvram of my laptop’s graphics chipset.
(I sent it in, and all they apparently did was booting WiXP once,
whose driver seems to have reset that; the problem caused by Knoppix
was OS-independent: happened under DOS and even in the BIOS Setup
screens. I do not own WiXP.)
That being said – is people like you and software like udev and
systemd really the reason for this?
Just use eudev and do not use systemd. Or fork it. Debian has a
history of patching packages to death away from what upstream,
or other distros, do.
And I absolutely do not buy the argument that Debian does not
have enough manpower to keep the / vs. /usr separation (for many
use cases) working.
Ben mentioned hotplug stuff. Can that not be deferred to until
after /usr is mounted? (If need be, patch eudev so that it looks
at whether a triggered’s dependencies are available and defer it
if not.)
Another thing to consider is: my m68k boxen run udev now too, but
they boot very finely without it (and devtmpfs.mount=1, with some
code in /etc/rc.local to create /dev/std{in,out,err} symlinks).
How about supporting / vs. /usr separation if either an initrd is
used or it’s a server/embedded setup, but require an initrd or /usr
to be on the same filesystem as / for a desktop?
Ah well, us Germans and convoluted sentences. Let me make a matrix.
\ desktop/laptop server/embedded
/ and /usr on same fs supported supported
separate, initrd used supported supported
separate, no initrd not supported supported, will patch
Right now, to me, from the discussion, it looks like the lower-right
corner is the only one doing actual work, and the only one in question,
and that the two upper lines are “a given” even by the upstreams in
question. Merging / and /usr, one way or the other, has never been
mentioned as a requirement as long as they’re on the same filesystem
or initrd mounts them early, it’s merely been proposed on the grounds
of “Fedora does it and they think it’s convenient”.
Fedora/RedHat being entangled with some upstreams also never was a
problem for Debian to do it differently in the past, either – just
look at eglibc when RHEL people were keeping development of glibc slow.
(Putting several responses and thoughts into *one* mail because some
“fellow” DDs seem to take offence at reading several eMails within
a few days from me. And I hope, dear unnamed co-DD, that the things
I proposed here are constructive enough. I also hope Debian will
continue to support freedom of choice and appreciate people looking
at things from a different PoV, as for the latter if only to have
a reflection.)
bye,
//mirabilos (still not subscribed, reading via GMane)
Reply to: