Re: [gentoo-catalyst] mounts changes was: [PATCH 5/6] modules/generic_stage_target.py, modules/stage1_target.py: Add a target_mounts dictionary

["W. Trevor King" <wking@tremily.us>]

[-- Attachment #1: Type: text/plain, Size: 2614 bytes --]

On Wed, Jan 01, 2014 at 06:05:35AM +0000, Jorge Manuel B. S. Vicetto wrote:
> On Mon, 30 Dec 2013, Brian Dolbec wrote:
> > On Sat, 2013-12-28 at 19:52 -0800, W. Trevor King wrote:
> >> On Sat, Dec 28, 2013 at 06:49:11PM -0800, W. Trevor King wrote:
> >>> Hmm.  I'm not how this is handled at the moment, but don't we want:
> >>>
> >>>   mount -t proc none /tmp/stage1root/proc
> >>>
> >>> instead of:
> >>>
> >>>   mount --bind /proc /tmp/stage1root/proc
> >>>
> >>> The handbook is currently using --rbind as well [1].
> >>>
> >>> Maybe food for a future series :p.
> >>
> >> After Douglas Freed talked me off the devtmpfs ledge on
> >> #gentoo-releng, my current suggested mount changes are:
> >>
> >>   -t:
> >>     proc
> >>   --rbind:
> >>     sys
> >>     dev
> >>     portdir
> >>     port_logdir
> >>     port_tmpdir
> >>     kerncache
> >>     ccache
> >>     icecream
> >>   drop:
> >>     devpts (now that we --rbind /dev)
> >>     distdir (now that we --rbind portdir)
> >>     packagedir (now that we --rbind portdir)
> >>
> >> For distdir/packagedir, I'd be find if the source defaulted to None
> >> (meaning “don't mount anything”) but could be configured to a path
> >> (meaning “--rbind this after you've --rbind-mounted portdir”).
> >
> > No, we can't drop distdir and packagedir.  The default locations for
> > those are waiting for the catalyst changes to be completed.  That is the
> > reason I got started rewriting catalyst a year ago.  Those 2 directories
> > are being relocated OUT of the tree.  Like they should have been all
> > along.
> 
> As I've already mentioned on #gentoo-releng, there are also 2 reasons this 
> is a no-go for me:
> 
> 1. it's unlikely we want to share pkgdir between the host and the targets 
> we build
> 
> 2. sharing pkgdir between incompatible targets would fail miserably. For 
> example, skimmer does amd64 builds for hardened and non-hardened, multilib 
> and no-multilib. It also does x86 builds. Sharing a package between any or 
> at least most of those targets would fail and cause broken stages.

So packagedir should just come from the pkgcache.  That makes sense.
I still think /dev should be mounted via --rbind, and likely other
user-configurable sources as well.  If we remove a user-configurable
packagedir source and keep the rest of this --rbind proposal, does
that address all of your concerns?

Cheers,
Trevor

-- 
This email may be signed or encrypted with GnuPG (http://www.gnupg.org).
For more information, see http://en.wikipedia.org/wiki/Pretty_Good_Privacy

[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 836 bytes --]