Re: [PATCH] tests/functional: Set 'pc' machine type instead of relying on the default

[Thomas Huth <thuth@redhat.com>]

On 21/05/2025 17.26, Daniel P. Berrangé wrote:
> On Wed, May 21, 2025 at 05:13:23PM +0200, Thomas Huth wrote:
>> From: Thomas Huth <thuth@redhat.com>
>>
>> In case the default machine has not been compiled into the QEMU
>> binary, the vnc and the cpu_hotplug_props test are currently
>> failing. Add a set_machine('pc') here to make sure that the tests
>> are correctly skipped in case the machine is not available.
> 
> Hmmm, this is making this test be tied exclusively to x86.
> If we can't ever omit the machine, then effectively all
> our tests have to be tied to a specific target arch ?

Since you never can rely on the availability of a default machine (there are 
also some targets like arm that don't have a default machine), you either 
have to set a certain machine explicitly with set_machine(), or use the 
"none" machine (if possible). Speaking of the latter - that's maybe a better 
option for this test, so I'll give that a try.

> What scenario did you find that 'pc' wasn't compiled
> into QEMU  x86 targets ?

Run "configure" with "--without-default-devices".

> Shouldn't a new default
> machine have been picked, or should our test harness
> be picking a default for each targets if no 'set_machine'
> call is present, to preserve target portability ?

We're using something like that in select_machine() in 
tests/functional/test_migration.py already ... we could maybe give it a try 
to turn that into a generic function of the qemu_test module, but I guess 
it's hard to find a default for each and every target that satisfies all the 
needs...

  Thomas