So, this series fixes a few issues that are very tightly coupled, so buckle up if you want to read further, if not, here you go: TL;DR: Issue: there's a test passing even though the code doesn't really work Result: fix both the code and the test LONG STORY: The main issue is that we allow starting a domain with multiple OpenGL-enabled graphics devices: <graphics type='spice'> <listen type='none'/> <gl enable='yes' rendernode='foo'/> </graphics> <graphics type='egl-headless'/> But hey, we have both a validation code handling this and a test case and everything seems to work. Well, not really, since the validation code for graphics was never run because it runs as a callback from our device iterator which only iterates over devices which provide boot info data which graphics do not provide. Any attempt to pass some arbitrary boot info data for graphics will cause troubles in other callbacks that are invoked from within the iterator. The way to enable it is to introduce some flags to alter the behaviour of the iterator. So why didn't we this scenario? We did, but the test case was lacking some capabilities and since it's a negative test case, it happily accepted any kind of error from the library, therefore, negative versions of DO_TEST_CAPS_LATEST have been introduced to mitigate failures like these in the future. Erik -- libvir-list mailing list libvir-list@xxxxxxxxxx https://www.redhat.com/mailman/listinfo/libvir-list
