Making a dynamically-linked root

[Julian Elischer]

On Wed, 4 Jun 2003, Mike Barcroft wrote:

> 
> I think if we identify the dependency graph we can restrict the
> parallelization to only the leaf nodes.  Granted, we're probably
> missing some non-obvious dependencies in rc.d like the cron/sendmail
> one you mentioned.
> 

Given a lot of daemons allow their shell to continue almost immediatly,
how do you tell the difference between the daemon starting,
but destined to eventually fail, and starting and destined to succeed?

The fact that the shell continued with no error condition is not
any reflection as to whether the service (e.g an IP address in the case
of dhclient) is available yet or not.)

In the end every possible 'service' has to have its own 'test of
completion'.. e.g. "does fxp0 have an IP address yet?", or
"can we resolve a hostname yet?", or, "is /usr mounted yet".
We solved this on teh Interjet because we had a known set of services
and we literally hard-coded a lot of stuff.