1 |
I like this idea. |
2 |
|
3 |
Geert Bevin wrote: |
4 |
> I've been thinking about the core engine, and what would be nice is an |
5 |
> abstracted installer storyboard language. This might for example use xml |
6 |
> to define actions, sequences, paths, default responses, etc etc |
7 |
> The storyboard in itself can be maintained independently and several |
8 |
> frontends can interprete it. I can freak out on the Qt version, Hallski |
9 |
> can make his Gtk+ version if he feels like it, and someone else (or |
10 |
> maybe me if I feel courageous) can do an ncurses version. |
11 |
|
12 |
In general for install/config/admin tools I think the key is to separate |
13 |
the functionality from the interface. The kernel config tools are a |
14 |
good example of what I mean: you get the same functionality no matter |
15 |
whether you make config, menuconfig, or xconfig. The kernel config |
16 |
tools are a good model in one other respect as well: help which is |
17 |
actually helpful most of the time. It is important that help is |
18 |
part of the functionality, so you get the same help no matter which |
19 |
interface you choose. |
20 |
|
21 |
I have one suggestion regarding hardware detection/configuration. |
22 |
Doing things automatically is not easy--this is something even |
23 |
big distributions get wrong frequently. There are several options |
24 |
intermediate between Windows-style automatic configuration and the |
25 |
current extreme do-it-yourself approach. One which involves minimal |
26 |
work is to use ebuild scripts to do all the work. For example I |
27 |
could write an ebuild for Epson 777 printers (USB version) which |
28 |
would download the upp files, check dependencies--in this case |
29 |
ghostscript and a USB controller package--maybe check /usr/src/linux/.config |
30 |
and warn if kernel support is missing. It could also use lpadmin |
31 |
to configure the printer, but maybe this shouldn't be automated. |
32 |
If everyone who had trouble geetting a device working under gentoo |
33 |
wrote an ebuild for that device then new users could save a lot of |
34 |
time getting their systems working. One advantage is that you could |
35 |
use whatever UI front end you write for choosing and installing |
36 |
applications to configure hardware as well. I don't see much point |
37 |
in hardware detection at the moment; I think most new users know |
38 |
what hardware they have. |
39 |
-- |
40 |
John Stalker |
41 |
Department of Mathematics |
42 |
Princeton University |
43 |
(609)258-6469 |