[Pkg-sysvinit-devel] Bug#674517: initscripts: RAMTMP is turned on during upgrades

[Roger Leigh]

On Fri, May 25, 2012 at 02:57:27PM +0100, Roger Leigh wrote:
> On Fri, May 25, 2012 at 02:12:32PM +0200, Cyril Brulebois wrote:
> > Roger Leigh <rleigh at codelibre.net> (25/05/2012):
> > > We previously didn't enable this for upgrades primarily because it
> > > wasn't possible to.  It wasn't possible to upgrade /etc/default/rcS.
> > > Also, quite a few people complained about the semantics of RAM* in
> > > rcS changing since it used to refer to /var/* rather than the new
> > > locations (mainly from RAMRUN and RAMLOCK).  To address these concerns,
> > > all the RAM* settings were moved to /etc/default/tmpfs, putting all
> > > the tmpfs-related settings in a single place.  However, this does
> > > result in tmpfs being enabled on upgrade.
> > > 
> > > Whether the default is set to enabled or disabled, we now have a single
> > > place to configure it, which will take effect for upgrades or new
> > > installs.  Having a consistent default for both upgrades and new
> > > installs is, I think, generally desirable, irrespective of what that
> > > default ends up being.
> > 
> > Changing the default on upgrades in *not* acceptable.
> 
> We should probably only have done that for RAMLOCK.  It's definitely
> possible to continue to look in rcS for RAMSHM and RAMTMP.  I'll
> look at it on Monday.

To keep you up to date, this is what I've done so far.  It's in
sysvinit.git.

  - We no longer comment out RAMSHM or RAMTMP in /etc/default/rcS,
    and these settings are used if present (the settings in tmpfs
    take priority, but these are unset and disabled by default).

This takes care of upgrades, so the existing RAMTMP setting will
be preserved and will continue to be used.  We could also reverse
the commenting of RAMTMP entries in rcS from recent upgrades with
a little more work.  If you feel that this should be undone, I'll
do this as well.

New installs: defaults to default set in initscripts; this is
unchanged.

Upgrades from squeeze: these have never had a RAMTMP setting, so
it will still get enabled by default.  The above only helps
testing/unstable users.  We do need to prevent it being enabled
in this scenario as well.  Thoughts?
- We could, if we are upgrading from the squeeze version, add
  RAMTMP=no to rcS in the preinst.  Ugly, but would match the
  testing/unstable upgrade behaviour.  Alternatively, we could
  edit /etc/default/tmpfs (but it's a conffile).


Regards,
Roger

-- 
  .''`.  Roger Leigh
 : :' :  Debian GNU/Linux    http://people.debian.org/~rleigh/
 `. `'   schroot and sbuild  http://alioth.debian.org/projects/buildd-tools
   `-    GPG Public Key      F33D 281D 470A B443 6756 147C 07B3 C8BC 4083 E800