[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: Solution strategy: Re: use and abuse of debconf



On Wed, Jan 17, 2001 at 12:51:59PM -0800, Joey Hess wrote:

> Matt Zimmerman wrote:
> > What I would like is a shared-global template with a package-local answer,
> > such as shared/overwrite_config_file, into which can be substituted, say, a
> > package name and a config file.  This template would presumably be included
> > with debconf.  The question could then be asked, and the answer stored in
> > package/overwrite_config_file.  That way, the wording, formatting, etc.
> > would be correct and consistent for all packages which need to ask this
> > question (which would become very common).
> 
> We could make a package that contains such templates and make debconf depend
> on it.

Yes, but how would we store a per-package (or possibly n-per-package, for
packages with multiple config files) copy of the user's response?  The config
script would need to be able to say "ask shared/overwrite_config_file, but
store it as mypackage/overwrite_config_file_etc_foo_conf".

Anthony Towns suggested a method using md5sums that emulates the current
conffile mechanism in a very nice way.  The advantage this would give us would
be to avoid asking the user any questions unless they actually edit the config
file.  The drawback is that we would need to read the config file in .config,
or else ask the question in postinst.  Is it kosher for .config to touch the
filesystem?  What if it is expected to retrieve answers from a central
database, and behave the same on a cluster of systems regardless of filesystem
state?  Both of these options seem distasteful.  Is there any way to have a
conffile-like mechanism that allows for preconfiguration?  Or do we need to
always ask the user about every config file?

-- 
 - mdz



Reply to: