On 07/07/2011 10:28 AM, Eric Blake wrote:
> +++ b/tests/statstest.c
> @@ -17,7 +17,7 @@ static void testQuietError(void *userData ATTRIBUTE_UNUSED,
>
> static int testDevice(const char *path, int expect)
> {
> - int actual = xenLinuxDomainDeviceID(NULL, 1, path);
> + int actual = xenLinuxDomainDeviceID(1, path);
ACK.
This was a regression introduced in commit 60d769a1, back in April. So
why did we not detect it until now? What is different about your test
environment that triggered compilation of this code while others (myself
included) have been happily unaware of its existence?
Found it - statstest is guarded by the automake conditional
ENABLE_XEN_TESTS, set at configure time according to whether xen sockets
were usable.
But to avoid this type of failure in the future, I'd rather have the
test unconditionally compiled if xen headers are present, and merely
have the test have an early opt-opt return 77 (for skip) if the xen
socket cannot be connected to at runtime.
--
Eric Blake eblake(a)redhat.com +1-801-349-2682
Libvirt virtualization library
http://libvirt.org