On Tue, Aug 22, 2017 at 04:07:09PM -0300, Eduardo Habkost wrote: > (CCing Cleber and Stefan) > > On Tue, Aug 22, 2017 at 07:19:45AM -0300, Philippe Mathieu-Daudé wrote: > [...] > > Can we predict how the python scripts will evolve? Only fast-testing? > > > > I guess it depends on how you define "fast". Does "fast-testing" > include a full device-crash-test run (that could take ~30 > minutes) or the full set of iotests? > > > > Is there some users hacking on qemu.py unaware they can/should use > > libvirt-python? > > I believe the existing users of qemu.py wouldn't want to have > dependencies on libvirt or other external modules (e.g. code that > test specific QMP commands and/or is run by "make check").
Yes, most QEMU tests interact with QEMU at a lower level than the libvirt API. They may exercise new features that are not available via libvirt so we'd end up bypassing it anyway. It is easier to debug a test case that interacts directly with QEMU. Investigating a failed test where libvirt is involved is more time-consuming. libvirt and tools built on it, like virt-builder, are good for tests that need a fully-fledged operating system inside the guest. Avocado Virt is aimed at higher-level tests involving guests too: https://avocado-virt.readthedocs.io/en/latest/WritingTests.html Stefan
