Re: [Qemu-devel] [PATCH 0/5]: Improve machine type functions

[Anthony Liguori]

On 02/24/2012 09:20 AM, Luiz Capitulino wrote:
> On Fri, 24 Feb 2012 08:58:55 -0600
> Anthony Liguori<address@hidden>  wrote:
>
> > On 02/24/2012 08:13 AM, Luiz Capitulino wrote:
> > > I was reading some related code yesterday and couldn't resist improving this.
> > > Not sure if this is aligned with any possible QOM work in this area, but I'm
> > > posting this anyway...
> >
> > I'm not sure how easy this is, but the way to do this with QOM would be:
>
> This seems like a cool thing to work on. The two major problems (for me at 
> least)
> is testing and the possible huge amount of manual work,

Heh, I wouldn't quite call it huge :-)

I think you can pretty easily sed the machine init function definitions to 
accept a Machine parameter as the first argument.

I think to introduce the types, you would need to write a quick python script 
that parsed the QEMUMachine declarations and then converted that into a TypeInfo 
+ class_init function.

I wouldn't bother trying to make the code conversion script perfect.  I've found 
it's easiest to convert the easy 90% automatically and then do the remaining 10% 
by hand.

 or do you think that
> most of the work can be automated?
>
> The problem with testing is that, we have a whole lot of machines I have no idea
> what's the best way of testing the conversion work.

Since the conversion shouldn't change any logic, there shouldn't be a huge test 
burden.  What I've done in the past is just write a little script that parses 
the output of -M ? and then walks through and executes qemu with no arguments 
other than -M $name.  It's a bit tedious since you have to manually close the 
window but it's a relatively quick way to check each machine is behaving the 
same as it was before.

Regards,

Anthony Liguori

> [...]
>
> > That's not to say this series isn't a good cleanup.  I haven't actually reviewed
> > it yet.  I just wanted to illustrate the next QOM steps.
>
> Ok. I honestly think that this series is a small first step. Even the QTAILQ
> conversion is worth it IMHO, because it clearly shows what the function is doing
> in a single line.