2013/12/31 Jean-Baptiste Rouault <jean-baptiste.rouault(a)diateam.net>:
On Monday 30 December 2013 11:26:08 Ryota Ozaki wrote:
> On Mon, Dec 30, 2013 at 5:55 PM, Jean-Baptiste Rouault
>
> <jean-baptiste.rouault(a)diateam.net> wrote:
> > On Sunday 29 December 2013 14:44:10 Ryota Ozaki wrote:
> >> On Wed, Dec 25, 2013 at 12:47 AM, Jean-Baptiste Rouault
> >>
> >> <jean-baptiste.rouault(a)diateam.net> wrote:
> >> > While working on adding virDomain*Stats support to the vbox driver, we
> >> > found bugs in the VirtualBox API C bindings. These bugs have been
> >> > fixed in versions 4.2.20 and 4.3.4.
> >> > However, the changes in the C bindings are incompatible with the
> >> > vbox_CAPI_v4_2.h and vbox_CAPI_v4_3.h files which are bundled in
> >> > libvirt source code. This is why the following patch adds
> >> > vbox_CAPI_v4_2_20.h and vbox_CAPI_v4_3_4.h.
> >> >
> >> > We tried to keep compatibility with older VirtualBox 4.2.x and 4.3.x
> >> > releases so we added a "SPECIAL_VERSION" identifier to
conditionnaly
> >> > include the right header. I'm not really pleased with this
> >> > "SPECIAL_VERSION" identifier, maybe we could instead increase
the
> >> > precision of "VBOX_API_VERSION", for example 4002 would
become
> >> > 4002000. This would permit us to select the right header based on the
> >> > VBOX_API_VERSION only, what do you think ?
> >>
> >> Can we use VBOX_XPCOMC_VERSION instead of adding a new flag?
> >> The version has been bumped up when the incompatibility is introduced.
> >>
> >> ozaki-r
> >
> > The problem is that VBOX_XPCOMC_VERSION is defined in the vbox_CAPI_v*.h
> > headers and we need a flag to choose which header we have to include.
>
> Oops. You're right.
>
> Well, one other idea is to include each vbox_CAPI_X_Y.h in
> the corresponding vbox_VX_Y.c. That's rather straightforward
> for me than including vbox_CAPI_*.h in vbox_tmpl.c according to
> VBOX_API_VERSION.
>
> ozaki-r
This would indeed solve the problem for header inclusion. But what about
future code using the new API ? Will it have to check both VBOX_API_VERSION
and VBOX_XPCOMC_VERSION ?
Wouldn't it be simpler if VBOX_API_VERSION was more precise ? e.g 4003004
Okay, so the actual underlying problem here is that libvirt assumes
that VirtualBox API can only change between minor versions (4.2 ->
4.3), but we have a case here where it changed (or got fixed) between
release version (4.2.19 -> 4.2.20). Using this new SPECIAL_VERSION
define is the least invasive fix for the problem.
The more correct solution would be to make VBOX_API_VERSION represent
the full API version number, instead of just the major and minor part.
This would fix the incorrect assumption in libvirt.
--
Matthias Bolte
http://photron.blogspot.com