python-lxml is likely always already present anyway (due to so many
packages being dependent on it), but at least on my F20 system,
python3-lxml wasn't installed, leading to a failure of "python
setup.py rpm" without an informative error message.

As is now done with libvirt. git head will always have the version
number of the expected *next* release.

Add RPM deps on python/python3-nose, make RPM build run the
test suite and invoke nosetests using correct python binary

Signed-off-by: Daniel P. Berrange <berrange@redhat.com>

The number of parameters in new_params is not guaranteed to be the
same as the number of parameters in params.  Use the correct count
when freeing new_params to avoid crashes.

If pkg-config isn't installed, or a too old libvirt, we can't even
do 'python setup.py --help' without throwing an exception.

Have the pkg-config checks and validation only throw an exception if
being called from the 'build' step.

https://bugzilla.redhat.com/show_bug.cgi?id=1074170

This means we call it twice for a build operation, but I don't think
that's a big deal.

Makes it a bit more clear what all that code is used for, rather than
intermixing it with function definitions.

Besides the comment additions, this is a no-op and just reindents the
block, into a function.

Wrap the new virConnectDomainQemuMonitorEventRegister function
added in libvirt 1.2.3.  This patch copies heavily from
network events (commit 6ea5be0) and from event loop callbacks
in libvirt-override.c, since in the libvirt_qemu module, we
must expose top-level functions rather than class members.

* generator.py (qemu_skip_function): Don't generate event code.
(qemuBuildWrappers): Delay manual portion until after imports.
* libvirt-qemu-override.py (qemuMonitorEventRegister)
(qemuMonitorEventDeregister): New file.
* libvirt-qemu-override.c
(libvirt_qemu_virConnectDomainQemuMonitorEventFreeFunc)
(libvirt_qemu_virConnectDomainQemuMonitorEventCallback)
(libvirt_qemu_virConnectDomainQemuMonitorEventRegister)
(libvirt_qemu_virConnectDomainQemuMonitorEventDeregister)
(libvirt_qemu_lookupPythonFunc, getLibvirtQemuDictObject)
(getLibvirtQemuModuleObject): New functions.

Signed-off-by: Eric Blake <eblake@redhat.com>

Noticed this bug while adding qemu monitor events; there's probably
lots of other misuse of libvirt_virDomainPtrWrap, but for now I'm
limiting the fix to all copied-and-pasted event callbacks, since
I'm about to copy it again in the next patch.  While at it, check
for failure to extract the "conn" key from the opaque callback
struct, and hoist that check to occur before we reach the point
where it is harder to undo on failure (the network code was the
only code that had it in the right place, but then it failed to
restore thread state on failure).

The graphics callback is still not clean; but incremental
improvements are better than nothing.

* libvirt-override.c (libvirt_virConnectDomainEventCallback)
(libvirt_virConnectDomainEvetnLifecycleCallback)
(libvirt_virConnectDomainEventGenericCallback)
(libvirt_virConnectDomainEventRTCChangeCallback)
(libvirt_virConnectDomainEventWatchdogCallback)
(libvirt_virConnectDomainEventIOErrorCallback)
(libvirt_virConnectDomainEventIOErrorReasonCallback)
(libvirt_virConnectDomainEventGraphicsCallback)
(libvirt_virConnectDomainEventBlockJobCallback)
(libvirt_virConnectDomainEventDiskChangeCallback)
(libvirt_virConnectDomainEventTrayChangeCallback)
(libvirt_virConnectDomainEventPMWakeupCallback)
(libvirt_virConnectDomainEventPMSuspendCallback)
(libvirt_virConnectDomainEventBalloonChangeCallback)
(libvirt_virConnectDomainEventPMSuspendDiskCallback)
(libvirt_virConnectDomainEventDeviceRemovedCallback): Don't pass
NULL to PyObject_CallMethod.
(libvirt_virConnectNetworkEventLifecycleCallback): Likewise, and
don't corrupt thread state.

Signed-off-by: Eric Blake <eblake@redhat.com>