On Mon, Jun 11, 2012 at 11:25:48AM +0800, Igor M Podlesny wrote:

On 11 June 2012 08:17, Tom Gundersen <tomegun@archlinux.org> wrote: […]

...

   rc.sysinit: remove warnings

   These were useful in a transition period, but no need to keep them around.

   If /usr is not mounted we probably won't even get this far. If devtmpfs is not supported    the previous mount of /dev would have given an errormessage.

   This was prompted by a suggeston from Igor Podlesny to improve on the error handling, which    made me realise we could drop it.

1) Does it mean "grep" can be hosed out of PKGBUILD as well?

Interesting choice of words...

2) Surely I welcome cleaning up the code, but transition period is something subjective. Another one good thing about Arch is it's simplicity and flexibility, allowing people to use their own kernels easily, so it's reasonable that If initscripts have some assumptions they would be better tested, probably at the very beginning of it, wouldn't they?

Removal of these assumptions has little to do with initscripts itself, but rather with the rest of userspace's expectations. There's little we can do in initscripts to combat the fact that recent versions of udev _require_ devtmpfs for proper operation. There's little we can do to work around the fact that not having /usr mounted from early userspace leaves us handicapped and presents unfixable bugs. So no, this isn't a problem of wanting to let users be "flexible" in their config or using their own kernels. We've offered all the warnings about this sort of thing ahead of time, and I've personally gone out of my way to ensure that mounting /usr from early userspace is supported. dave

On Mon, Jun 11, 2012 at 5:55 AM, Igor M Podlesny <for.poige+archlinux@gmail.com> wrote:

On 11 June 2012 11:31, Dave Reisner <d@falconindy.com> wrote: […]

...

but rather with the rest of userspace's expectations. There's little we can do in initscripts to combat the fact that recent versions of udev _require_ devtmpfs for proper operation. There's little we can do to

  I don't think that stating aloud of assumptions in initscripts would hurt someone. Even if the only thing it could do is emitting a few warning and then even pause booting — at least it gives (gave) hints.

The point is that I realised that the warnings are redundant. A missing devtmpfs will print an error when we trie to mount it (before this warning would have been printed). If /usr is not mounted we won't ever get this far in the bootprocess anyway, so the warning would never be printed. The "transition period" has admittedly been short, but not because we chose it. We started printing the warnings when we realised that tha hard requirements would change in the future. Now that the hard requirements have changed (udev was updated and /lib moved to /usr/lib) the warnings are of no use any more (as explained above). This reminds me: a warning that I'd like to ADD is to warn when HARDWARECLOCK in rc.conf and the third line in /var/lib/hwclock/adjtime are out of sync. This is causing some subtle bugs. Cheers, Tom