Re: Feedback on Ted T'so's initscripts proposal

[tytso@mit.edu]

Sorry for the delay in responding, but I'm writing this from
Australia.....

First of all, my intent was that the original implementation would not
support scripts being able to declare that they provided a certain
initscript dependency.  Instead, LSB scripts would only be allowed to
depend on certain LSB-defined dependencies: network, syslog, netdaemons,
etc.  This means that it would be very simple for the initscript
installer to map that to a specific SysV rc.d SXX and KXX number.  

The whole point of doing things this way is so that we don't have to
force distributions to harmonize start and stop levels.  Given that we
haven't been able to get distribution to agree on /etc/rc.d/init.d
versus /etc/init.d, I figured that getting distributions to agree on
numbers would be nothing short of impossible.   (Who would change?
"Incremental upgrades would be too hard" (tm).)

Being able to use this to do parallel rc scripts was just a bonus.  I
wasn't planning that we had to implement parallel scripts first off, or
perhaps at all.  (Like Jim, I'm concerned about depending on having
standards committee design a new mechanism.)  However, I'd like to make
sure that the minamalistic LSB specification wouldn't preclude someone
from making init.d scripts from being parallelized in the future.

As far as circular dependencies are concerned, that's easy.  They're not
allowed, and so if you try to install a script which would introduce a
circular depedency, the init.d installer will refuse, print an error and
exit.  But that's not something we need to implement for the the initial
LSB release.

							- Ted