Public SourceMage Discussion List

[Jeremy Blosser <jblosser-smgl AT firinn.org>]

On Mar 13, David Brown [dmlb2000 AT gmail.com] wrote:
> > The second harmful effect here is that the whole shadow cast fails,
> > combined with castfs, I guess. Because without castfs, no installed
> > files would be missing (am I wrong here?). the installation of
> > important stuff already happened.
> > It would be much saner if the dispelled shadow would be automatically
> > restored if the updating cast fails at some stage ... or not be
> > dispelled before the install is totally done inside the castfs.
> 
> Yeah, kinda... castfs does provide a completely transparent
> environment to run in. This means that (ideally) the entire system
> should act as it would without castfs. Another way of saying this is
> when you mount castfs it provides an exact copy of your system
> (without any copying) at the mount location so you can chroot to it
> and mess around. That is exactly how the install phase of cast works
> with stage root enabled.
> 
> chroot $STAGE_DIRECTORY/MOUNT delve INSTALL $SPELL
> 
> The only partial root is after the install is done and during the
> POST_INSTALL stage where the spell's cwd is $STAGE_DIRECTORY and
> people can mess with the installed files before they are moved to the
> system.

I think this is beside the point.  The problem is that sorcery dispels the
current version ahead of running INSTALL.  If INSTALL fails, the dispel
still happened, and nothing is left on the live system.

The pre-stage-root approach masks this problem because half of a make
install may still have a fair amount of the target on the live system.  In
either case, if the install fails we should probably resurrect the previous
one, or just wait to dispel until we know the install succeeded and we're
getting ready to move to the live system now that stage-root enables that
option.

Attachment: pgpDWQuP82DDZ.pgp
Description: PGP signature