Updating to rawhide is harder than I thought.

So with my work done with the Fedora Board, I have decided to spend more time on things I can fix (technical things) versus those I can not (human interactions). One of those things, was to use more of the Rawhide release because if I can earn LWN.net weekly quotes of the week, I should endure the pain that Jonathan Corbet goes through every day. Hopefully it will also get F16 to be better.

Ok steps to do this:
  1. Install a laptop with Fedora 15.
    This was pretty easy as I am just using the laptop I built for SELF that just has a bunch of development things and a lot of games for the kid (ok me) to play when bored.
  2. Update to rawhide.
    Not as easy as I thought. I figured, hey lets just do a

    yum --enablerepo=rawhide update

    Which gave me after 2 minutes a long list of broken dependencies and such. Hey no problem.. I ran into this with the Fedora 15 betas.. use --skip-broken.

    yum --enablerepo=rawhide --skip-broken update

    This went along a lot further, but died due to packages with the same name installed multiple times. Ok let us get around that.

    package-cleanup --cleandupes

    Tada.. we can now get to our next error: "Protected multilib versions". For some reason various packages are protected from being updated even when the update is showing x86_64 to x86_64. Ok.. maybe this is an error fixed in yum.

    yum --enablerepo=rawhide --skip-broken update y*

    Oh lookie.. I have updated yum.. that was easy. Too easy.. I forgot to update python also. Running yum now gave me a huge traceback:

    --> Processing Dependency: libstartup-notification-1.so.0()(64 bit) for gnome-desktop3-...
    Traceback (most recent call last):
    pkgtup = (nevra_dict['name'], nevra_dict['arch'],
    TypeError: tuple indices must be integers, not str

    Ok maybe an updated python will fix this. Turns out that yum update python worked.. but it didn't fix the issue.. ok time to see if I can work around it in parts. Caveman Oook says to do a

    yum --enablerepo=rawhide --skip-broken update [xyz]*

    and back away down the alphabet over time. Why? Because it is a magic chicken foot that used to work in the days before yum when rpm -Uvh *rpm was used.
  3. Watch laptop catch on fire.
    Well not exactly but the /proc/acpi/ibm/thermal showed the temps over 100C which is quite hot. The system no longer was responsive and trying to log in as root would not work with messages about audit backlog limit exceeded. Turning off the system was about the only hope and turning it back on resulted in a blinking cursor in the left hand corner.
An interesting experiment.. a bit worse than my "Doing and Everything Install" but interesting. Will see if the box will come back after an hour cooling period.


Anonymous said...

It's going surprisingly well for me.

I just did a **minimal** install of Fedora 15 with the DV in a VM, then:
# yum update
# reboot
# yum --enablerepo=rawhide update
# reboot.

The main difference between you and me no doubt comes from the fact I had almost nothing on this box, so no borken deps, no insane amount of updates,...

Still, that was my first ever Rawhide install (I usually install just after the branching), so I'm happy it went so well (especially after reading your blog post, you got me scared :P)

Stephen Smoogen said...

@bochecha Yeah the problems seem to be a) hardware related and b) having a lot of packages installed. Doing it your way is a much saner way of doing it (which I may end up on try 3).