Re: ROADMAP in git tree needs to be updated.

[Ludovic Courtès]

Omar Radwan <address@hidden> skribis:

> I think we need to rewrite the roadmap from scratch.

In place of a road map, here’s my take at a relatively detailed list of
things that I think we should have for 1.0 to satisfy a range of use
cases, and things that can happen later:

* Features scheduled for 1.0

  - larger & more robust build farm
    + we need a powerful, dedicated front-end
    + armhf-linux build machine
    + leave Hydra in favor of 'guix publish' + custom code?
  - more OS features
    + LVM support
    + encrypted root
    + configurable name service switch
    + whole-system unit tests, using VMs
  - more service definitions
    + mcron, postfix(?), wicd(?), etc.
  - better 'guix system'
    + 'reconfigure' should be able to restart non-essential services
    + support for '--list-generations' and '--delete-generations'
  - better 'guix pull'
    + using Git to fetch the source instead of re-downloading everything
    + build more quickly
    + install new .mo files and new manual
    + authentication of the Guix source: use signed commits?
  - simplified, purely declarative service list in 'operating-system'
  - GUIs
    + integrate guix-web?
    + guile-ncurses installer?
  - 'guix publish'?

* Features for later

  - complete GNU/Hurd port
  - use content-based addressing when downloading substitutes to reduce
    bandwidth requirements
    + design nar v2 format where file contents are replaced by their hashes
    + leverage /gnu/store/.links
  - binary origin tracking
    + keep signatures in sqlite.db
    + preserve signatures upon import/export
  - peer-to-peer distribution of updates (GNUnet?)
  - more deterministic builds
    + identify & fix sources of non-determinism in builds
    + strengthen guix-daemon containers to further increase reproducibility
    + trusting-trust: bootstrap with different tool chains
    + fixed-point: re-bootstrap until fixed point is reached
    + distributed validation: compare contents of store items with others
      * resist a hydra.gnu.org compromise
  - reproducible containers: mix of 'guix environment' and 'guix system vm'
  - execute code with least privilege
    + build containers like guix-daemon does
    + provide a Plash-like interface in Bash
  - daemon rewritten in Guile
  - more dmd integration
    + monitor network interfaces and start/stop events based on that
    + include a DHCP client written in Scheme

The second part is particularly unstructured, but that’s because I think
all these items are fairly orthogonal and can happen in any order.  It’s
also non-exhaustive, I’m sure.  :-)

The first part is a list of things IMO required before we can claim to
have reached 1.0.

It’s a bit of a brain dump, but I hope it makes sense.

What do people think?  Anything missing?  Opinions on relative
priorities?

I’m happy to put that into ROADMAP if that is understandable.

Thanks,
Ludo’.