init: API

[Gilboa Davara]

On Sun, 2005-11-20 at 09:48 -0800, Kenneth Porter wrote:
> --On Sunday, November 20, 2005 6:52 PM +0200 Gilboa Davara 
> <gilboada at netvision.net.il> wrote:
> 
> > A. Who said that both / and /usr are on the same partition? (Or on the
> > same machine for that matter?)
> 
> Yep, this is a separate issue. Mounting /usr read-only is desirable, so 
> putting it on a separate partition is nice.
> 
> Ignoring XML specifically as the configuration language, what's important 
> is that the language have a common parser, so that every service doesn't 
> have to provide its own. If not XML, do we invent a new language, or codify 
> an existing one? If an existing one, someone needs to extract the parser 
> from some single implementation and declare it the one that all others will 
> be rewritten to use, so that there's no ambiguity in the syntax.

Even if you don't use a bash based solution (that does the parsing for
you), you can create a service manager library that can used by the
executable (to execute the services) and the CLI and GUI tools (to
manage the service manager).

I'm no against shared code....

> 
> So far all init scripts are written in Bash, and the ones I'm familiar with 
> use simple variable=value pairs, using Bash syntax. Is that sufficient? 
> What can't be expressed in that way?
> 
> The hierarchical configuration I've seen has been in the networking 
> scripts, and is encoded in the filesystem, using filenames or directory 
> trees to represent configuration nodes. What's the drawback of that, if any?
> 
> 

I fully second the above.

The main question should be: Can we voodoo-magic the current scripting
engine to improve performance considerably.
In my experience, if the said service manager is designed correctly (and
I'm willing to pitch in, design, coding and all), yes.

Gilboa