Re: [Qemu-devel] [PATCH V2 2/6] libqblock type and structure defines

[Blue Swirl]

On Thu, Sep 13, 2012 at 3:49 AM, Eric Blake <address@hidden> wrote:
> On 09/12/2012 09:33 PM, Eric Blake wrote:
>>>  OK ,then I think
>>> #if __GNUC__ >= 4
>>> ....
>>> #else
>>>   [warn name space pollution may happen]
>>> #endif
>>> would be better.
>>
>> It may be shorter, but it is definitely not better, at least not in the
>> current context of qemu.  Using the short form will fail a -Werror
>> build, unless you also write a patch to qemu's configure to quit
>> supplying -Wundef during builds.  But as touching configure has a bigger
>> impact to the overall qemu project, you're going to need a lot more
>> buy-in from other developers that -Wundef is not helping qemu gain any
>> portability, and that it is safe to ditch it (or get enough
>> counter-arguments from other developers why qemu insists on the
>> anachronistic style enforced by -Wundef, at which point you must comply
>> and use the longer form).
>
> On second thought, this _particular_ usage will never fail a -Wundef
> -Werror build, precisely because -Wundef is a gcc warning, which impies
> the warning is only ever useful in the same scenarios that the __GNUC__
> macro is always defined (that is, __GNUC__ is undefined only on a
> non-gcc compiler, but what non-gcc compiler supports -Wundef -Werror?).

The library could be used by a project that does not use GCC or pick
CFLAGS from QEMU configuration. Supporting for example MSVC or C++
users for the library could be interesting one day, even if we didn't
support MSVC or C++ at all for building the rest of QEMU.

>
> But why should this line be the one exemption to the rules?  Either qemu
> insists on the -Wundef style of coding (and you should use the long form
> to conform to that style, on the off-chance that someone ever wants to
> port to a non-gcc compiler, even in this one place where gcc can't warn
> you about the violation of that style), or we should change the qemu
> style (at which point, the short form is nicer here, but it also implies
> the potential for cleaning up lots of other places to also use short
> forms and rely on preprocessor 0 computation).
>
> --
> Eric Blake   address@hidden    +1-919-301-3266
> Libvirt virtualization library http://libvirt.org
>