Re: [PULL 00/16] qtest and gitlab-CI improvements

[Thomas Huth <thuth@redhat.com>]

On 20/12/2021 10.24, Philippe Mathieu-Daudé wrote:
> +Alex & Paolo
> 
> On 12/20/21 07:52, Thomas Huth wrote:
>> On 18/12/2021 17.33, Philippe Mathieu-Daudé wrote:
>>> Hi,
>>>
>>> On 12/15/21 08:33, Thomas Huth wrote:
>>>
>>>> ----------------------------------------------------------------
>>>> * Add virtio-net failover test
>>>> * Make qtests a little bit more flexible with regards to reduced configs
>>>> * Move libssh setup from configure to meson.build
>>>> * Run device-crash-test in CI
>>>> * Add jobs for NetBSD and OpenBSD to the CI
>>>> * Test compilation with MSYS2 in the gitlab-ci, too
>>>> * Add new virtio-iommu test
>>>>
>>>> ----------------------------------------------------------------
>>>
>>>> Laurent Vivier (4):
>>>>         qtest/libqos: add a function to initialize secondary PCI buses
>>>>         tests/qtest: add some tests for virtio-net failover
>>>>         tests/libqtest: add some virtio-net failover migration
>>>> cancelling tests
>>>>         tests/libqtest: add a migration test with two couples of
>>>> failover devices
>>>
>>> On my ASan build directory I'm sometime getting:
>>>
>>> Running test qtest-i386/virtio-net-failover
>>> (process:1558675): GLib-CRITICAL **: 16:19:12.556: g_rand_int: assertion
>>> 'rand != NULL' failed
>>
>> Weird, since the test is not using that function?
> 
> Well it calls g_test_rand_int(), which calls it:
> https://github.com/GNOME/glib/blob/main/glib/gtestutils.c#L1800

Ah, right. I think the problem is that g_test_rand_int() must not be called 
before g_test_init().

But actually, I think in this case we even don't want to use 
g_test_rand_int() since it's about generating a true random temporary file 
name, not a random number for a test case that could be influenced with the 
"--seed" CLI option.
So I think using g_random_int() might be the better choice instead?

  Thomas