On Wed, Dec 03, 2014 at 15:58:45 +0000, Daniel Berrange wrote:
On Wed, Dec 03, 2014 at 04:52:25PM +0100, Jiri Denemark wrote:
Both qemu-img and its stripped down version qcow-create are searched for dynamically with virFindFileInPath, functions using them report errors when the required binary is not present, and we don't compile in a fixed path detected by configure anymore. So I don't see a reason for keeping the dependencies for our storage driver. Moreover, it's weired when the dependencies (of the storage driver) change depending on wheter a particular hypervisor driver (xen/qemu) is enabled or not.
Signed-off-by: Jiri Denemark <jdenemar@redhat.com> --- libvirt.spec.in | 18 ------------------ 1 file changed, 18 deletions(-)
Even if using the Xen hypervisor, you can use the qemu-img binary for creating volumes. Since the qemu-img binary is in a separate RPM from the main QEMU emulator binaries, it makes sense to depend on qemu-img if we're enabling QEMU support in libvirt. Similarly the LXC driver can make use of qemu-img for creating images that are then used via qemu-nbd.
So I think our current Requires deps are actually correct here - they ensure that the qemu-img binary is always pulled into the host, no matter what daemon-driver-XXX is installed.
But it still feels wrong. A choice of what hypervisor driver is enabled should not influence dependencies of a storage driver. If we want the storage driver to depend on qemu-img, I think the dependency should be enabled by a separate switch, so that one can enable qemu-img support without enabling qemu driver. Or am I just over-engineering it based on my confusion why storage deps change depending on enabled HV drivers? Jirka