Re: [libvirt] [PATCH v2 2/5] tests: Initialize basic capabilities properly
On Wed, 2017-03-29 at 19:50 -0400, John Ferlan wrote:
> The capabilities used in test cases should match those used
> during normal operation for the tests to make any sense.
Could you add just a bit of text about why all -no-acpi is removed
(mostly) and why -no-hpet is added to a couple?
It's essentially the comment from the subsequent patch:
"-no-acpi is not supported on non-x86 even if qemu reports it in -help",
so all non-x64 .args file need to remove
Still not sure I understand why the no-hpet got added, but I suppose
they're necessary. And only a couple of outputs have it... Guess it
depends on what else is defined...
Both are consequences of the test suite not using hardcoded
capabilities anymore: while before this patch NO_ACPI was
hardcoded as present (even though it shouldn't have been,
except for x86 guests) and NO_HPET was hardcoded as *not*
present (even though it should have for x86 guests), after
this patch arch-dependent capabilities are set properly and
so all aarch64 guests lose the -no-acpi in their .args and
the one x86 guest that has
<timer name='hpet' present='no'/>
but didn't have NO_HPET explicitly set (a bug in the test
case) gains the corresponding -no-hpet.
I'll add a short note about that to the commit message, but
we should probably go one step further and drop all uses of
NO_ACPI and NO_HPET from the test cases altogether, since
we're now supposed to be setting them automatically for the
right test cases and setting them manually might just be
hiding existing bugs. I just posted patch 2.5/5 that does
just that :)
--
Andrea Bolognani / Red Hat / Virtualization