Re: [PATCH] tests: honor $TMPDIR for test_virtio_version

[Richard Henderson <richard.henderson@linaro.org>]

On 9/1/25 16:41, Michael Tokarev wrote:
> On 31.08.2025 15:52, Maxim Cournoyer wrote:
>> Until 10.1.0, the test suite could be run without having a writable
>> /var/tmp in the build environment.  To avoid now requiring /var/tmp in
>> the build environment (which can be a very minimal container like in
>> the case of GNU Guix), consult TMPDIR first, using /var/tmp as a
>> fallback.
> 
>> diff --git a/tests/functional/x86_64/test_virtio_version.py b/tests/functional/x86_64/ 
>> test_virtio_version.py
>> index a5ea73237f..501545f655 100755
>> --- a/tests/functional/x86_64/test_virtio_version.py
>> +++ b/tests/functional/x86_64/test_virtio_version.py
> 
>> @@ -68,7 +70,9 @@ def run_device(self, devtype, opts=None, machine='pc'):
>>           """
>>           Run QEMU with `-device DEVTYPE`, return device info from `query-pci`
>>           """
>> -        with QEMUMachine(self.qemu_bin) as vm:
>> +        with QEMUMachine(
>> +                self.qemu_bin,
>> +                base_temp_dir=os.environ.get('TMPDIR', '/var/tmp')) as vm:
> 
> I don't think /var/tmp is a good choice here, - I'm a bit surprised it
> come from you when your OS doesn't have /var/tmp by default - shouldn't
> it be /tmp here?
> 
> Overall, /var/tmp is a strange choice here and in a few other tests too,
> -- maybe only block.c default is the only right place to have it as the
> fallback/default.

It's possible that it's the testsuite that should be changed to prefer /var/tmp when 
present and TMPDIR is tmpfs.  I don't know enough about what prompted the original change 
in the first place.

But my point is that with *only* this change, it is very likely that this simply exchanges 
one set of failures for another.


r~