TODO

1.  More OS support:
     - Ubuntu 5.04 and 5.10
     - Debian Etch, Lenny, Squeeze
     - SUSE Linux 10.0, 10.1
     - openSUSE 10.2, 10.3 (autoyast)
     - OpenBSD (http://people.cs.uchicago.edu/~brendan/howtos/openbsd_install/)
     - NetBSD (anita)
     - FreeBSD
     - Dragonfly BSD
     - Windows Vista, 98, 95, ME
     - RHL 5.2, 6.0, 6.1, 6.2 (only nfs installs work)
2.  Add partition/disk size support to the TDL.
3.  Add improved error logging to guest installation.  That is, if the guest
installation fails, we want to get detailed error information out of the
installer to find out why.  Fedora/RHEL support doing remote syslog via
network or virtio-serial, and I also believe anaconda will dump data to syslog
during installation, so that is one avenue to pursue.  NOTE: anaconda
virtio-serial support does not exist until F-14.  See http://fedoraproject.org/wiki/Anaconda/Logging
4.  Instead of calling out to an external 'gvnccapture' command to collect
a screenshot, we should instead write the thing in python using the gtk-vnc
library.  Unfortunately I have not figured out how to do it in python yet,
so this needs further research.
5.  Before attempting an install, check the output_dir to make sure it will
approximately have enough space for the output disk image.
6.  With windows from MSDN (at least 2000, maybe other versions too), the
top-level doesn't really contain the bits you care about.  What happens is
that there are 3 boot options on the CD: installer for Professional, Server, and
Advanced Server.  The directory structure on the CD is something like:

    english
        win2000
            adv_server
            pro
            server

and underneath each of (adv_server, pro, server) are the bits you care about.
To automate this you really need to figure out which one the user cares about,
then copy just that subdirectory.  You'll also need to extract the appropriate
parts of eltorito so it boots automatically, and then you'll also need to set
i386/txtsetup.sif [SetupData]: SetupSourcePath = "\".  See
http://old.bink.nu/bootcd/
7. Write a boxgrinder plugin.  This requires implementing a ruby class
for it, and implementing .after_init, .deliverables_exists?, .run, .deliverables.
8. Take a screen shot if some customization or generate-icicle steps time out.
9. Try to do automatic detection of distro/version/architecture from the ISO
that are provided to us.  That will make it so that a minimal TDL will only
include the path to the ISO, and we can figure out the rest of the information.
10. Add support for additional drivers during install (needed for Windows virtio
support).
11. Make "python setup.py bdist_rpm" work.  The problem is that bdist_rpm
generates its own SPEC file based on the information in setup.py.  When it
does this, it somehow messes up the %description (minor, probably a change to
the setup macro), fails to create the /var/lib/oz subdirectories (which could
be moved into setup.py as well), and also fails to find the gzipped man page
files (which is what eventually causes the build to fail).
12. RHL 6.2 does not work via HTTP because of a bug in the installer; when
parsing a URL passed in via "method", it fails to put a / at the beginning of
the URL.  What this means is that when the installer goes to fetch the install
images via "GET path/to/netstg2.img HTTP/0.9", the web server then returns an
error.  To do a fully automated install, we need to use an ISO, NFS or FTP
install method; I could not get FTP to work, but I did not try that hard.
13. RHL 6.1 fails because there is no netstg2.img available in the distribution
I have.  Unfortunately, I have not been able to find the netstg2.img, nor an
ISO of 6.1 to do an alternate install.  NFS may still work here.
14. RHL 6.0 fails because the kernel panics on boot:
     VFS: Cannot open root device 08:21
     Kernel panic: VFS: Unable to mount root fs on 08:21
15.  Enable SELinux in enforcing mode inside of Fedora/RHEL guests.  The current
problem with this is that prior to launching the customize step, we upload files
like ssh keys via libguestfs.  In the default mode libguestfs uploads them with
no SELinux context, which means reads are denied.

Jim Meyering suggests that we add a script to the end of /etc/rc.d/rc.local that
does a "restorecon" on the files we care about.  Then we don't need to worry
about knowing the context in Oz; we just make it do the right thing.  There
are two problems with this: 1) if we upload /etc/rc.d/rc.local, that will have
the wrong context that we have to deal with, and 2) there is a race condition
between icicle-nc checking in and this restorecon script finishing.  Neither
problem is insurmountable, but we should make the effort to get this working.

16.  Do preflight checks on detected bridges.  That is, if we detected the
bridge (i.e. it was not specified by the user), then we should do some basic
checks to make sure that we think it will work.  We should check the state
of at least:

    /proc/sys/vm/net/ipv4/ip_forward
    /proc/sys/net/bridge/bridge-nf-call-arptables
    /proc/sys/net/bridge/bridge-nf-call-ip6tables
    /proc/sys/net/bridge/bridge-nf-call-iptables
    iptables -t filter -L -v
    iptables -t nat -L -v

17.  We should add a mode where we assume the responsibility for uploading RPMs
into the guest (via ssh) and then doing a yum localinstall on those RPMs.
18.  We should add additional commands section.  In particular, we should have
"pre" package installation and "post" package installation command sections.
19.  On distributions that use NetworkManager, we can speed up the customization
and ICICLE generation step by using dbus-monitor to monitor when NetworkManager
has brought up the network.  This should save us an average of 30 seconds per
customization.  The script to do this would look something like:

#!/bin/sh
#
# This script will be executed *after* all the other init scripts.
# You can put your own initialization stuff in here if you don't
# want to do the full Sys V style init stuff.

while true; do
gdbus introspect --system --dest org.freedesktop.NetworkManager --object-path
/org/freedesktop/NetworkManager/Devices/0 | grep "State = 100" > /dev/null
if [ $? -eq 0 ]
then
  break
fi
sleep 1
done

20.  Support hashes for the root password support.
21.  Add a mode where we use a reverse ssh tunnel with a SOCKS proxy to
download the packages.  This means that the packages need to be visible to the
imagefactory machine, but not necessarily to the instance.
22.  During customization/icicle generation, we should check to make sure that
the bridge listed in the libvirt XML matches up with what we detected (if
anything).  This isn't really an issue when doing oz-install -g -u, but if you
use the short-circuit oz-customize or oz-generate-icicle, the user could have
fed us bad data.
23.  In TDL.py, do a real schema validation using tdl.rng
24.  Allow a URL argument for both <files> and <commands>, where the user gives
us a URL with file contents or commands, and Oz assumes the responsibility for
downloading the file and then uploading it into the guest.
26.  Currently we use libguestfs to extract the installation ISO, make changes
to it, and then rebundle the ISO using mkisofs.  This is time consuming,
partially because we are doing two copies of the ISO (once from the ISO to the
filesystem, and once from the filesystem to the new ISO), and partially because
we are only modifying a very small portion of the ISO.  We can improve this
situation and get rid of one whole copy by mounting the ISO with
fuseiso/guestmount, symlinking most of the contents to a temporary directory,
and only modifying the small bits we need to.  Then we can pack the whole thing
back up using mkisofs.  This gets rid of the first copy, so it should cut our
initial time in half (plus or minus the page cache).
27.  Change setup.py to make the directories /var/lib/oz/*.  Since that is
repeated in both of the RPM and deb building stuff, it would be better to just
do it on python setup.py install instead.