Re: State of graphical installer
On Tue, Sep 28, 2004 at 12:53:09PM +0200, Martin Schulze wrote:
> State of graphical installer
>
> Does anybody know the current state of the graphical installer? Using
> debconf for d-i was said to have the benefit of hooking different
> frontends to it and the gtk/gnome frontend is one of them.
>
> Michael Cardenas also provided the first version of a graphical
> installer a couple of years ago (1-2) but I don't remember any
> progress or work being done to it.
>
> Is anybody interested in reviving it and working on a graphical
> installation method?
I'm busy trying to revive it, and most of my time at Oldenburg was spent
doing this. I'm working on the following immediate issues:
* cdebconf-gtk has bitrotted
gtk.c hasn't been touched since March, and it shows; it spends a lot
of its time segfaulting or misbehaving in other ways. I have a
series of patches in my queue to clean it up.
* several udebs are only built on i386
I'm developing on powerpc, and there are a number of GNOMEish udebs
that are only built on i386 due to an accident in debian/rules. Bug
reports will be forthcoming tonight.
* directfb udebs are a nightmarish mess
The directfb and gtk+2.0-directfb udebs are laid out very strangely;
I think they reflect some very old ideas on how to do udeb library
packaging. As they stand, they thoroughly confuse mklibs (kov posted
a patch to work around this, but the libraries really need to be
fixed). They need to be repackaged in a more normal way, with
ordinary runtime and development debs and a runtime udeb.
* mklibs issues
Even after I experimentally repackaged directfb and
gtk+2.0-directfb, I still ran into issues with mklibs reporting lots
of unresolved symbols. I'm still investigating; help from somebody
who knows mklibs well would be appreciated.
(So, as it stands, d-i only builds with the gtk frontend if you beat
it up a lot one way or another.)
* cdebconf extensions
We almost certainly need some cdebconf extensions to allow us to
make the GTK frontend a good user interface; I think the ability to
register a callback that gets triggered when the answer to a
question is changed (but not necessarily OKed) would be a good
start.
--
Colin Watson [cjwatson@debian.org]
Reply to: