Re: [PATCH][1/10] Device model cleanup.

[Arun Sharma <arun.sharma@intel.com>]

Ian Pratt wrote:

> This is useful progress, but I have a couple of comments/questions.
> 
> * should the empty config options be appearing in the sxp? I don't think
> so. Further, perhaps they should be wrapped in a (ioemu ...) section?
> 
> * we should make it such that disks can be specified using the same
> syntax as for normal paravirt drives e.g. file:, phy: etc.
>

Yes, will implement both of these.

> * I presume that the vncviewer is being forked from xm, but that the
> vncconect is being issued from xend, which is why it needs the $DISPLAY
> ?    

Rolf applied a patch sometime back that has two separate ways of connect 
a vncviewer to device models:

a) vncviewer -listen (port communicated to xend and then to qemu via 
-vncconnect). This is forked from xm.
b) libvncserver within qemu-dm (-vncport). This is forked from xend.

And then

c) the user may choose vnc=0 sdl=1.


It's really case (c) that requires $DISPLAY. Also, I think the end user 
needs only one of (a) and (b), not both. If $DISPLAY is not set in xm's 
environment, (a) will fail. In that sense (b) is more reliable.

> 
> * we still need some tests when building to ensure libvncserver and
> libsvg are installed. As the current behaviour when they're not is
> confusing -- I think its better to refuse to build ioemu if they're not
> installed.

Not all distributions are shipping libvncserver (for eg: FC4 didn't). 
Once we resolve that I think it's a reasonable thing to do.

> 
> * we should move qemu-dm from /usr/bin to /usr/lib/xen/bin
> 
> * files like mem-map.sxp should probably live in /usr/lib/xen/boot
>
> * presumably we no longer need bochsrc ?

Thanks for the comments. You should see some patches today.

	-Arun