Architectural changes
  - The Great Refactoring (Factor and merge out common code):
    # uClibc (cross/native/static toolchains)
    # binutils/gcc (cross/native/static toolchains)
    - ccwrap (cross/native/static toolchains)
      - Wrap existing toolchain
    # busybox/toybox (hostutils/rootfs)
    - distcc (hostutils/rootfs)

  - Improve individual tarball package granularity
    - binutils/gcc essentially one package, useless without ccwrap.
      - Also useless without uClibc, but uClibc useful on its own?
        - *.a files in with compiler. *.so files separately?
      - native compiler needs make.  Cross compiler doesn't.
    - Make setupfor/cleanup tarballs, then second pass to make better ones
    - Reorganize downloads directory
      - and snapshots

  - Prebuilt binaries
    - static cross compiler: ship only that but with right directory name
    - test static native compiler
      - start with static busybox only
      - add uClibc tarball
      - add native compiler tarball
      - build dropbear

  - ccwrap existing cross compiler to make root-filesystem.sh orthogonal
    - Break out ccwrap.sh in great refactoring?

  - Add hdc -> /mnt functionality
    - have sbin/init.sh call /mnt/init if it exists instead of shell prompt

  - Simplify root-filesystem.sh logic
    - directories (yes or no)
    - uClibc: static or dynamic?
      - static system with no toolchain, don't need uClibc installed
      - Just build against existing toolchain libc, unwrapped
      - Otherwise add ccwrap to existing cross comp to force uClibc usage
    - native toolchain (plus make, distcc)
      - Currently separate build stage.  Packaging?
    - busybox (yes/no, static/dynamic)

  - Upgrade busybox ash to eliminate need for bash

Bugfixes and tweaks

  - fix distcc (accidentally broken in The Great Refactoring)
  - uClibc should install getent?
    - Merge/test my implementation?
  - Add armv4tl-eabi (arm920t, openmoko)
    - rename armv4l to armv4l-oabi (all other arm targets eabi)
  - add /etc/passwd with root line so dropbear works

Check QEMU targets for 1.0 release:
  - current: arm, mips, mipsel, sh4, x86_64, x86
    - broken: armv4eb, m68k, sparc, ppc
  - pending: cris, microblaze, ppcemb, sh4eb
    - 64bit: mips64, mips64el, ppc64, sparc64
  - theoretical: alpha, s390
  - How much is really needed for a 1.0 release?

  powerpc-440fp should use 64k pages.
    - actual hardware does, so emulator should too

  m68k
    - Try other qemu tree
    - aranym?

  s390
    - hercules

  - armv7 (cortex-A8/beagleboard)
    - gcc 4.4 or 4.2 patch?
  - xylinx microblaze patch(es)

- Gentoo From Scratch
  gentoo script cleanups (bash, tar, patch)
    work towards a full stage 1 (list from schlotzky's)
    Package up gentoo stage 1 image (tarball and boot script).
  portage removal script
    remove unnecessary packages
    remove portage

- Fedora from scratch
  - Download koji
  - Ask on #fedora-admin

# Teach ccwrap to use libgcc_eh.a when available.
  - Export ccwrap for use by uClibc upstream.

gigabit ethernet
  - test all targets with this, migrate the ones that can use it

Try uClibc-git
  - new pthreads (not old)
  - nptl where available

Improved nightly cron job
  - web page with status

Push patches upstream
  - linux
    - dnsdomainname patch
    - perl removal patches
      - remove remaining perl from kernel build

Documentation
  - Post slides from OLF talk
    - Convert keynote to html slideshow
    - MP3 recording
    - Post pricing details for 1-day and 2-day training sessions
  - Redo documentation.html (split up?)

Benchmark
  - How much faster is distcc?
  - ./configure with static binaries?
  - On each platform:
    - loop 10 million times
    - launch 1000 instances of "true"
      - static binaries

----------

# build/sources should be build/packages
# functions.sh uses $TOP/sources when it should be using $SOURCES.

#use squashfs as hda by default?
# Boot from squashfs, built as normal user, with stable version.

toybox:
  - add fuzz factor support to patch