Re: [Qemu-devel] [PATCH 1/3] tests: use qtest_pc_boot()/qtest_pc_shutdown() in virtio tests

[Greg Kurz <groug@kaod.org>]

On Fri, 30 Sep 2016 11:13:33 +0200
Laurent Vivier <lvivier@redhat.com> wrote:

> On 30/09/2016 10:33, Greg Kurz wrote:
> > On Thu, 29 Sep 2016 19:15:05 +0200
> > Laurent Vivier <lvivier@redhat.com> wrote:
> >   
> >> [...]
> >>  static void pci_nop(void)
> >>  {
> >> -    qvirtio_9p_start();
> >> -    qvirtio_9p_stop();
> >> +    QOSState *qs;
> >> +
> >> +    qs = qvirtio_9p_start();
> >> +    g_assert(qs);  
> > 
> > The appropriate macro to use here is: g_assert_nonnull().  
> 
> OK
> 
> > 
> > BTW, how can qs be NULL ?  
> 
> we should not know what happens in  qtest_pc_boot() (or
> qtest_spapr_boot(), or qtest_XXX_boot())
> 

What is the point in letting qtest_XXX_boot() return NULL then if
it is always followed by g_assert() in the test program code ?

I'd rather move the assertion there and document that it cannot
return NULL, since it is always unrecoverable in the test code.

> So I think it i better to check it before to use it.
> [...]
> >> +static QOSState *pci_test_start(void)
> >>  {
> >> -    char *cmdline;
> >> +    QOSState *qs = NULL;  
> > 
> > Why setting qs to NULL ? It is necessarily set...  
> 
> Yes, I've mixed my patches: later we add a "if (arch == pc) { qs = }
> else if (arch == spapr) { qs = }" and this case qs can be uninitialized.
> 

Ok, I've realized that when reading the other patch. :)

> [...]
> >> +    qtest_shutdown(qs);  
> > 
> > The title is "tests: use qtest_pc_boot()/qtest_pc_shutdown() in virtio tests"
> > 
> > Even if qtest_shutdown() happens to be equivalent to qtest_pc_shutdown() when
> > qtest_pc_boot() was called, I would rather stick to the title, and convert
> > all qtest_pc_shutdown() to qtest_shutdown() in patch 3/3... 
> > 
> > No big deal though, you may s/qtest_pc_shutdown/qtest_shutdown in the title
> > as well :)  
> 
> I'm to change all qtest_pc_shutdown() to qtest_shutdown() here
> 

Ok.

Cheers.

--
Greg