Troy J Dawson wrote:
> Constance J. Sieh wrote:
>> On Mon, 29 Jun 2009, Troy Dawson wrote:
>>
>>> Hello,
>>> I have been wanting to update the kernel-module plugin for yum on SL5. I have 
>>> it all ready in the testing area, and it does work so much better than the 
>>> previous kernel module plugin.
>>> The problem is that it does not backport very well to yum 3.0 which is on SL 
>>> 5.0 and 5.1.  To keep things short there is a bug in yum 3.0 (and 2.4) that 
>>> prevents the plugin's from fixing up dependancies when yum gets them wrong. 
>>> This bug is fixed in yum 3.2 (which is what is in SL 5.2 and 5.3)
>>> So why would I worry about upgrading the old yum?
>>> Because along with bug fixes, there is a couple of feature changes.  The 
>>> biggest change is that yum 3.2 automatically has the installonlyn feature. 
>>> This only keeps 'n' kernels on your machine.
>>> This works pretty good and I think many people will think it's great. But I 
>>> don't know if everyone is going to like it, and they might be surprised by 
>>> it.
>>> Anyway, I need opinions.  Should I push the newer yum out to all of SL5?
>>>  Or should I just push it out to SL 5.2 and 5.3?
>> And live with the bugs for 5.0 and 5.1 .  Can you give more detail about 
>> these bugs.
>>
> Well, we could go through yum's bugzilla, but I'll tell about the bug 
> that was biting me.
> 
> If there was a dependancy problem during the normal yum requirements 
> check, and you tried to clean it up in the plugin's section that comes 
> just after that, yum would still think it had the dependancy problem 
> after the plugin section.  Basically, it wouldn't go back and recheck 
> the dependancies.
>  From what I could tell, it also didn't go back and check the 
> dependancies of any of the extra rpm's added during the plugin section. 
>   This could also cause a problem.
> 

I forgot to say, both of these bugs are fixed in yum 3.2.
It keeps track if packages were added and/or removed and/or changed 
during the plugins and rechecks for dependancies if they were.

> It sounds like this should be a simple fix, but it isn't.  The 
> dependancy checking functions were compeltely rewritten between 3.0 and 
> 3.2, so there isn't just a few lines of code to patch.
> 
> Troy


-- 
__________________________________________________
Troy Dawson  [log in to unmask]  (630)840-6468
Fermilab  ComputingDivision/LCSI/CSI LMSS Group
__________________________________________________