rawhide report: 20050420 changes

Jim Cornette fct-cornette at insight.rr.com
Thu Apr 21 02:02:59 UTC 2005


Rodd Clarkson wrote:
> On Wed, 2005-04-20 at 21:33 -0400, Jim Cornette wrote:
> 
> 
>>Same  problem noted here. I also observed that these errors referenced 
>>kernel versions that I no longer have on this system. The error seems to 
>>be related to kernel-devel because the errors included an uninstalled 
>>kernel, however the devel packages ae never removed.
>>
>>kernel-devel-2.6.11-1.1236_FC4 - error referenced
>>kernel-devel-2.6.11-1.1238_FC4 - need to manually remove
>>kernel-devel-2.6.11-1.1240_FC4
>>kernel-devel-2.6.11-1.1251_FC4
>>
>>vs.
>>
>>  rpm -q kernel
>>kernel-2.6.11-1.1240_FC4
>>kernel-2.6.11-1.1251_FC4
> 
> 
> Mmm, I've got the matching kernel and kernel-devel packages installed.
> Now I think about it, this mmap error isn't to do with the kernel but
> with kernel-devel.  The errors occur when kernel-devel is installed, not
> kernel.
> 
> This makes me wonder if it hasn't got something to do with the 'magic'
> kernel-devel uses to try and keep the amount of similar files down so as
> to keep the install size smaller.  As I understand it, kernel-devel
> hardlinks same files so that they are only installed once (rather than
> for each package).  This is part of the reason why installing and
> removing kernel-devel takes a little longer (it had to check if the
> hardlink is the last instance, and thus if it needs to be removed.)  I
> could be wrong about all this, but I remember it being discussed a
> little while back and think it's pretty close to what happens.
> 
> 
> R.
> 

I removed the unused kernel-devel packages and then compiled the vmware 
modules using /usr/bin/vmware-config.pl.

I did not get the lockup that I had when I had kernel-devel versions on 
the system, but no associated kernels to match up to the unassociated 
devel packages.

Anyway, both the hardlink problem and mmap did seem to take forever.
Since carrying a slew of kernels on a system is a rarity, doing away 
with the hardlink concept and reverting to independent files seems to be 
a better approach. Compiling modules seems to be safer with kernel to 
kernel-devel purity, since the purpose for the devel package is to add 
newer and hopefully better differences to the system.

MMAP does take forever, especially for the first kernel-devel removal. 
The second removal did not seem to take as long to remove. I removed 
1236 first (long delay) and then 1238 which seemed to complete quickly.

Jim

-- 
We are the people our parents warned us about.




More information about the fedora-test-list mailing list