Re: [Qemu-devel] [PATCH] move 'unsafe' to end of caching modes in help

[Avi Kivity]

 On 07/26/2010 10:03 PM, Anthony Liguori wrote:
> On 07/26/2010 11:29 AM, Avi Kivity wrote:
> >  On 07/26/2010 07:26 PM, Avi Kivity wrote:
> > > > The help output is *not* a supported interface. 
> > >
> > > There is no supported, usable interface for this.
> >
> > Well actually, libvirt could probe this by starting qemu with cache=x 
> > for various x and seeing if it breaks.  But the milk has already been 
> > spilled.
>
> So we're stuck with supporting the help output as an interface 
> forever?   Even with a capabilities system, what about old versions of 
> libvirt?

No, the rate of crap generation seems to increase all the time, we have 
to get rid of it eventually.  If we provide a replacement, give a 
warning, wait some months for libvirt^Wusers to catch up, and throw away 
the deprecated feature, we can remove any feature we like.  This is 
still a fast moving field and users to upgrade.  Just not every week.

> At some point in time, old versions of libvirt are going to stop 
> working with new versions of QEMU because the help output changes.  If 
> libvirt switches to something more reliable (like versioning), then 
> the gap is closed until there is a capabilities system.
>
> There will be a breakage though and we shouldn't pretend that taking 
> this patch does anything other than delay that breakage.

There's a difference between
- planned breakage with a warning ahead of time
- unplanned breakage
- unplanned breakage with unwillingness to fix

--
I have a truly marvellous patch that fixes the bug which this
signature is too narrow to contain.