Re: Installing non-interactively/improving interactive postint-scripts
Martin Buck wrote:
> On Fri, Aug 01, 1997 at 01:10:06AM -0400, Steve Dunham wrote:
> > One nice thing about RPM is that the packages can be installed
> > non-interactively.
> Can you give a short summary how rpm achieves this?
Well it just doesn't ask any questions. Any questions that they ask are
centralized and seperate from the packages. (So this wouldn't work for
us both because we are decentralized and because we want more choice in
the behaviour of the packages we install.)
> > (This isn't an implicit fault of the .deb format,
> > rather the packages that are available in the .deb format.)
> I couldn't agree more. What I'd propose is a standard way for
> postinst-scripts to ask questions. If we wouldn't use simple shell-scripts
> with 'read' but something a bit more advanced (maybe like the
> kernel-configuration scripts), installation could be made much more
> user-friendly (for novice *and* advanced users).
I think this has been discussed before, but as far as I can tell it was
What would be nice is a system that entails:
1. A global configuration file of answers to _install-time_ questions.
2. Changing the postinst scripts to look in the configuration file
3. A utility for asking the questions and creating the file.
One option would be to standardize a file describing the variables that
program uses (and their datatypes, string, number, etc.)
> 'read' is completely unusable for an X-based installation (one of the goals
> deity wants to achieve, IIRC) and it's not very comfortable. It doesn't
> allow you to go back to the previous question if you made an error or would
> like to change your opinion (like in the Windows '95 installation, where
> you always have 'Back' and 'Next' buttons that even allow you to peek at
> the next question before answering the current one). Also, providing
> default answers for non-interactive installations is difficult with this
> If we would have a standard program to ask questions during the
> installation, this could do the right thing depending on then installation
> frontend used. This program could have a Tcl/Tk backend for X, a
> dialog-based one for the console or even just a simple 'read' for those who
> like the current state. It also could force the package-maintainers to give
> symbolic tags to every question, so that you can identify all questions and
> put default answers for them into a file. This would be very useful for
And, of course, the option to read from an install configuration file.
(You can do this with NT to acheive a network install - graphics drivers
and all - with no questions asked.)
> A simple query-program won't be able to go back to the previous question,
> however. To do this, we'd need something like the kernel-configuration
> scripts I mentioned above. This would require rather big changes to the
> postinst-scripts, but I think this definitely would be worth the effort.
> What do others think about it?
I don't think it would be too big of a deal to change all the postinst
scripts - the big problem is coming up with a system and standardizing
This is something that could be discussed to death.
TO UNSUBSCRIBE FROM THIS MAILING LIST: e-mail the word "unsubscribe" to
Trouble? e-mail to email@example.com .