Concrete implementation discussion based on the DebCamp BoF moved to MetaInit

Init Hackfest

I (ErichSchubert) propose that we should do a ?HackFest some time after Etch was released, to update our init stuff.

Debian ships with a couple of init systems, e.g.

Other closely related programs include:

However, out of these only sysvinit is really supported (others work only in a compability mode, taking away many of their benefits).

Therefore it's proposed to do a "Init hackfest"; either face-to-face or via IRC (or both) with the aim of adding full support to most of these init systems to Debian. (And to monitoring systems such as monit.)

Proposed schedule:

  1. get the maintainers/authors of the relevant packages together to write up requirements
  2. from these requirements, derive a migration document (think of the Python policy documents)
  3. supply appropriate tool changes, e.g. updating invoke-rc.d, update-rc.d
  4. update *init packages to be able to 'compile' an init script for their config scheme from the meta-init-script
  5. do a hackfest to fix as many packages as possible in a sprint (it's easier if only a few people have to dig into how to convert an init script to the meta-init-script format). Supply patches to the maintainers and do 7-day NMUs
  6. 7 days later, the NMUs will go in.

Random notes:

Shipping meta-init-scripts allows other init systems to be added (or bugs fixed with them) without having to NMU all other packages again. We just need to get the meta-init format right in the first go... maybe it's also easier to continue shipping packages with a sysvinit script and the meta-init script for upgrade purposes, and schedule the removal of the sysvinit scripts for lenny+1.

For those who like something a bit more concrete - how do we stop something like #400952 from happening?

Please check the discussions on the initscripts-ng mailing list: http://lists.alioth.debian.org/mailman/listinfo/initscripts-ng-devel

Participants: