On 3/18/21 12:10 PM, Peter Maydell wrote: > On Thu, 18 Mar 2021 at 10:45, Paolo Bonzini <pbonz...@redhat.com> wrote: >> >> On 18/03/21 11:40, Peter Maydell wrote: >>> On Thu, 18 Mar 2021 at 10:37, Paolo Bonzini <pbonz...@redhat.com> wrote: >>>> >>>> On 18/03/21 11:06, Laurent Vivier wrote: >>>>> This also removes the virtio-devices test, I think we should keep the >>>>> files, but in the files to disable the PCI part when it is not >>>>> available. >>>> >>>> I think we should just shuffle the targets in the gitlab YAML to bypass >>>> the issue. >>> >>> Then we'll hit it again later. I'm pretty sure this isn't the >>> first time we've run into "some test makes dubious assumptions"... >> >> We can both fix qemu-iotests and CI configuration, but m68k is certainly >> not the culprit here. And we are going to make more assumptions over >> time, not fewer, in order to keep the CI time at bay. > > I don't see why CI time is relevant to whether the test says > "I require X,Y,Z, so don't run me on configs without those" > or whether it just randomly assumes X,Y,Z are always present > or that if it says "I require W" than W must imply X,Y,Z...
Recently we changed a bit our view and are trying to have smarter tests. In particular building target/device agnostic tests, and have the test queries the QEMU binary what features/devices are available before running. This will take some time before we get there, unlikely for the 6.0 release. For short term, Paolo's "shuffle gitlab YAML" suggestion is simpler.
