[Fedora-livecd-list] [PATCH] -fs.py-kickstart.py-chages-for-mulit-partitions

[Bryan Kearney]

Jeremy Katz wrote:
> Mark McLoughlin wrote:
>> On Wed, 2008-05-28 at 14:36 -0400, Jeremy Katz wrote:
>>>> The ability for a tool to create these partitioned disk images via the
>>>> command like, livecd-creator, is very helpful in reproducing known 
>>>> state
>>>> appliances, epically with the rising use of visualization.  Going
>>>> through a actual install in a vm limits you to VM container, xen, kvm,
>>>> vmware and adds extra overhead of building on a machine with extra virt
>>>> resources.  
>>> Reproducing known state should work just as well with a kickstart 
>>> config and doing an install via, eg, virt-install.  Yes, it adds 
>>> overhead, but there are also advantages in functionality and it also 
>>> means *not* maintaining this sort of code in multiple places.  And 
>>> that's a pretty big win as partitioning code is not in the least bit 
>>> trivial. Especially as you start trying to handle more and more of 
>>> the cases.
>>
>> That pretty lame - you're trying to curtail what use cases people make
>> imgcreate support purely because we haven't figured out how to do a
>> better job of sharing code with anaconda.
> 
> If there were multiple people really clamoring for support to do 
> partitioned images, it would be one thing.  But right now, it's one 
> tool.  And to then carry the onus outside of that tool for keeping the 
> non-trivial bits working?
> 
>> The very least we should be doing is allowing imgcreate to be extended
>> externally in this manner ... but that doesn't fix the "multiple copies
>> of this code" issue, it merely pushes it elsewhere and makes it less
>> likely it'll ever be fixed.
> 
> But we are doing the work to allow an external appliance-creator to 
> carry the functionality if desired.  That's why the refactoring patch is 
> reasonable (modulo fixing up the API break).
> 
>> Clearly, one advantage of imgcreate tools over a full blown anaconda
>> install in a VM is that it's far less complex to debug when it all goes
>> wrong. That's a huge advantage to people repeatedly rebuilding images as
>> they develop a LiveCD or an appliance.
> 
> The reason that it's less complex to debug is the fact that the code is 
> more limited in what it supports than anaconda.  Adding more 
> (complicated) pieces is directly in contradiction to keeping things 
> simple to debug
> 

So.. to summarize.. as long as the partitioning code is in the appliance 
creator only.. we are fine. Is that correct?

-- bk