[libvirt] [PATCH] hostdev: Fix build with GCC's static analysis in mdev

Martin Kletzander mkletzan at redhat.com
Mon Mar 27 18:24:58 UTC 2017


On Mon, Mar 27, 2017 at 12:30:23PM -0400, Laine Stump wrote:
>On 03/27/2017 11:40 AM, John Ferlan wrote:
>>
>>
>> On 03/27/2017 11:30 AM, Martin Kletzander wrote:
>>> Similarly to eec3b255d26e7b38bdb0830990569fd91aee661f, fix build with
>>> lv_cv_static_analysis=yes.
>>>
>>> Caused by a4a39d90ab4930750bcbcfccffdf6bb6d310b5d5
>>> ---
>>>  src/util/virhostdev.c | 4 ++--
>>>  1 file changed, 2 insertions(+), 2 deletions(-)
>>>
>>
>> Sure that's another way to fix it... Could have also gone with the
>> removal of NONNULL in the prototype like I did.  IDC whichever way is
>> deemed "more appropriate"...
>>
>> Seeing as drv_name is only ever passed as QEMU_DRIVER_NAME, but dom_name
>> is passed from 'name' which doesn't have the NONNULL on it, that's why I
>> chose removing NONNULL from the prototype.
>
>I think I prefer John's way too. (At least partly because I dislike
>ATTRIBUTE_NONNULL() and would like to see as many of them as possible go
>away).
>

Actually, I do too.  I would love to remove all of them.  But for that
we'd have to reach a decision.  However there are not only cons, some
pros can be that the compiler is able to make better guesses or not have
to be guessing at all.  As a counter-argument, there are way more
attribute we could use and we don't, plus libvirt is not the CPU hogger
where we would have to go optimize so low.  So while the answer is not
clear, I wouldn't, as mentioned before, be against removing all such
occurrences, maybe except public APIs, if someone happens to have a good
reason for that.
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 833 bytes
Desc: Digital signature
URL: <http://listman.redhat.com/archives/libvir-list/attachments/20170327/2fa3e93f/attachment-0001.sig>


More information about the libvir-list mailing list