Re: [Qemu-devel] [RFC PATCH] pc: change default machine model and versions

[Kevin Wolf]

Am 10.04.2012 14:10, schrieb Andreas Färber:
> Am 10.04.2012 12:32, schrieb Peter Maydell:
>> On 10 April 2012 11:30, Andreas Färber <address@hidden> wrote:
>>> Independent of what frequency of machine versions we offer, I think
>>> defaulting to pc-1.0 is a bad idea. The people fiddling with QEMU
>>> command lines are mostly developers, others can be expected to use a
>>> more convenient frontend, such as libvirt or oVirt. I would hope that
>>> updating any of these frontends to supply a -M pc-1.0 argument would be
>>> possible without much effort. Expecting developers to type -M pc-next
>>> again and again will result in pc-next not being tested as much and thus
>>> broken quickly.
>>
>> To throw in a tangential non-x86 perspective, I would prefer that
>> there be no -M default at all. For ARM at any rate there is no
>> machine which is inherently preferable or better than any other;
>> the current default is integratorcp simply because it happened to
>> be the first one implemented, and this tends to lead to user errors
>> where people don't realise that not passing in a -M option results
>> in their getting an ancient and barely maintained model...
> 
> I'd be fine with such an explicit choice, too. As a side-effect that
> would simplify QOM'ification by not needing to create and iterate O(n)
> classes (wasting memory during regular execution) just to find the
> default one and then possibly override it with another.
> 
> I could generalize and clarify my goal as avoiding unintentionally using
> the "wrong" machine.

Not having a default may be a good choice for ARM, but why can't this be
a per-target decision? For x86 you don't really have this problem, and
one of qemu's strengths is that it doesn't need much configuration, but
even just a 'qemu-system-x86_64 disk.img' can be enough to get something
booted.

Kevin