Anaconda - Kadischi install to disk from LiveCD

[Jasper O'neal Hartline]

Hello.

I have a prototype Kadischi here that can be invoked from LiveCD made 
with Kadischi and also acts as a installer to disk from LiveCD.
The problems I am having relate to packages.py, yuminstall.py and 
grabber.py in the Python package under yum and urlgrabber site-packages.

I keep getting exceptions stating "Cannot open header" or "Cannot open 
package header" for instance on a Fedora Core 5 machine
lines 159 and 161 when in fact /tmp/cache/headers/*.hdr exists for the 
transaction set chosen.

I notice in the real installer boot.iso for FC6T2 we actually get 
headers populating a directory: /tmp/anaconda-base-200608030842.i386
I'm curious if anyone has any idea on how to resolve the issue with not 
opening package headers, and or where this directory is created.

Even symlinking anaconda-base-2006080308432.i386 to /tmp/cache where the 
headers really are, doesn't work.
Like Anaconda deletes the symlink in /tmp or something.

Any ideas where I can look to rectify this issue?
A Live CD installer using Anaconda would be excellent in my opinion, I 
almost have it working.

J. Hartline

[Jasper O'neal Hartline]

Jasper O'neal Hartline wrote:
> Hello.
>
> I have a prototype Kadischi here that can be invoked from LiveCD made 
> with Kadischi and also acts as a installer to disk from LiveCD.
> The problems I am having relate to packages.py, yuminstall.py and 
> grabber.py in the Python package under yum and urlgrabber site-packages.

Well, I finally got it working, but I had to hack the heck out of 
Anaconda to do so.
Currently this is what had to be changed: <attached>

Could we fix some of this in RAWHIDE perhaps, so this hackery won't be 
neccessary?
It seems alot of exceptions occur when a device already exists!

J. Hartline

[Jeremy Katz]

On Fri, 2006-09-01 at 04:30 -0500, Jasper O'neal Hartline wrote:
> Jasper O'neal Hartline wrote:
> > I have a prototype Kadischi here that can be invoked from LiveCD made 
> > with Kadischi and also acts as a installer to disk from LiveCD.
> > The problems I am having relate to packages.py, yuminstall.py and 
> > grabber.py in the Python package under yum and urlgrabber site-packages.
>
> Well, I finally got it working, but I had to hack the heck out of 
> Anaconda to do so.
> Currently this is what had to be changed: <attached>

Unfortunately, changing like this means that some valid cases where
problems occur won't be caught as nicely thus leading to bizarre crashes
for people :-/
> Could we fix some of this in RAWHIDE perhaps, so this hackery won't be 
> neccessary?

The real fix is going through and actually pretty fully auditing the use
of devices throughout anaconda so that we always use the device nodes
in /dev -- at this point, we should be creating them all and so some of
the hoops that have been historically jumped through aren't needed
anymore.

But doing that prior to FC6 is a bit too risky for my taste.
> It seems alot of exceptions occur when a device already exists!

Doing better checks to see if things are really present or not before
trying to create them (or catching the exception in the case where they
already exist) would be better and should be small enough to consider
for FC6

Jeremy