Re: [Qemu-devel] Machine description as data prototype, take 3

[Markus Armbruster]

Paul Brook <address@hidden> writes:

> On Thursday 19 February 2009, Markus Armbruster wrote:
>> Third iteration of the prototype.
>>
>> What about an early merge?  If your answer to that is "yes, but", what
>> exactly do you want changed?
>
> I dislike that you've got everything lumped together. In its current form 
> it's 
> unclear that it's actually an improvement from what we currently have. There 
> still seems to be an awful lot of code that's extremely PC specific, and I 
> can't tell whether/which interfaces achieve separation from the legacy 
> hardcoded PC nastyness and generic machine descriptions.

I think I got the PC nastiness encapsulated in driver methods.  I can
put them into a separate file if you think that would help.

> I'm also worried that we have both user config (image files, vlans, etc) and 
> machine description (devices instantiation, etc) described in the same place. 
> I still believe these should be separate tasks, with clear boundaries between 
> the two.

Actually, we don't.  The machine configuration tree does not contain any
host configuration (if it does, it's an oversight I'd be happy to fix).
That is all in struct dt_host, a completely separate data structure.

The two data structures join in two places:

1. dt_customize_config(), which bridges the gap between our current
configuration system and the new configuration data structure.  Since
the former has machine and host configuration mixed up, dealing with the
two together there is unavoidable.  But keeping them separate from there
on paves the way for keeping them separate from the start some day.

2. The driver config methods, which necessarily use both the machine and
the host configuration.