Feature proposal: Rebootless Installer

[Douglas McClendon]

Colin Walters wrote:
> On Tue, Jul 14, 2009 at 4:04 AM, Douglas
> McClendon<dmc.fedora at filteredperception.org> wrote:
> 
>> i.e. simply a checkbox before beginning installation stating whether you want rebootless instead of traditional.
> 
> I don't think the installation process needs more questions.  Also,
> off by default means it won't get testing and will eventually bitrot.

I think we would only ever consider adding a checkbox (not a question) 
to anaconda, if the initial reception of the feature in, e.g. F12, was 
so positive that it justified it without requiring any advocacy on my part.

> 
> It's unclear to me what your plan is for handling firstboot, which is
> a critical part of the OS setup. Now, we should probably move most of
> the current firstboot questions into anaconda.  At a minimum, it'd be
> useful to ask them when you'd otherwise be staring at the image
> copying progress bar.

I'll try to add this to the wiki.  My intention was to start with the 
minimum possible for the first experimental release (f12, alpha freeze 
in 3 weeks).  With the feature accepted, and me currently being 
unemployed, the minimum possible might actually be quite a bit though. 
My answer of course is the same as what you mentioned, they go into the 
installer.  Honestly a create user, change rootpw, and similar 
complexity pages are not that difficult to add.  It is also quite 
acceptable to move anything that can be done as 'regular system 
maintenance and reconfiguration' out of the installer.  The whole 
philosophy of a LiveOS is that you boot in the most generic agnostic way 
possible, and let the user configure at runtime.  I.e. it's actually a 
really good thing IMO to force the user to say use system-config-time or 
whatever to change the timezone.  I know too many examples from personal 
experience, dating back to the days when you couldn't easily tab out 
system-config- how much of a pain it was to learn how to administer 
things that were done in the installer.


> Another issue that occurs to me is that the livecd user, besides not
> matching the target username, also has no password (and I believe
> won't have an encrypted gnome keyring), but this is different from the
> expected target installation.

My long term ideal vision for this, would be to have gdm login accept an 
arbitrary user/pass instead of the autologin of liveuser, then that 
user/pass would be the initial user.  Obviously with an option at 
install time to disable the mode subsequently, probably default on. 
Though possibly left off for people who want a very guest permissive 
system.  For now, a simple addition to my installer already in my 
ROADMAP is a checkbox in the installer 'delete liveuser account upon 
logout'.


> So this looks technically cool, but there are a lot of problems to
> investigate and solve that it brings up.

Thanks for the positive feedback, and I do agree.  In response to some 
of the prior feedback, I already adjusted the wiki release notes to 
emphasize the experimental status.

peace...

-dmc