Re: [Qemu-devel] [PATCH] configure: Force the C standard to gnu11

[Thomas Huth]

On 2019-01-09 14:27, Daniel P. Berrangé wrote:
> On Wed, Jan 09, 2019 at 02:20:02PM +0100, Thomas Huth wrote:
>> On 2019-01-09 13:58, Daniel P. Berrangé wrote:
>>> On Wed, Jan 09, 2019 at 01:52:02PM +0100, Thomas Huth wrote:
>>>> On 2019-01-09 12:44, Daniel P. Berrangé wrote:
>>>>> On Wed, Jan 09, 2019 at 12:25:43PM +0100, Thomas Huth wrote:
>>>>>> On 2019-01-09 11:58, Daniel P. Berrangé wrote:
>>>>>>> On Mon, Jan 07, 2019 at 11:45:26AM +0100, Thomas Huth wrote:
>>>>>>>> Different versions of GCC and Clang use different versions of the C 
>>>>>>>> standard.
>>>>>>>> This repeatedly caused problems already, e.g. with duplicated typedefs:
>>>>>>>>
>>>>>>>>  https://lists.gnu.org/archive/html/qemu-devel/2018-11/msg05829.html
>>>>>>>>
>>>>>>>> or with for-loop variable initializers:
>>>>>>>>
>>>>>>>>  https://lists.gnu.org/archive/html/qemu-devel/2019-01/msg00237.html
>>>>>>>>
>>>>>>>> To avoid these problems, we should enforce the C language version to 
>>>>>>>> the
>>>>>>>> same level for all compilers. Since our minimum compiler versions are
>>>>>>>> GCC v4.8 and Clang v3.4 now, and both basically support "gnu11" 
>>>>>>>> already,
>>>>>>>> this seems to be a good choice.
>>>>>>>
>>>>>>> In 4.x   gnu11 is marked as experimental. I'm not really comfortable
>>>>>>> using experimental features - even if its warning free there's a risk
>>>>>>> it would silently mis-compile something.
>>>>>>>
>>>>>>> gnu99 is ok with 4.x - it is merely "incomplete".
>>>>>>
>>>>>> gnu11 has the big advantage that it also fixes the problem with
>>>>>> duplicated typedefs that are reported by older versions of Clang.
>>>>>>
>>>>>> Are you sure about the experimental character in 4.x? I just looked at
>>>>>> https://gcc.gnu.org/onlinedocs/gcc-4.8.5/gcc/Standards.html and it says:
>>>>>>
>>>>>> "A fourth version of the C standard, known as C11, was published in 2011
>>>>>> as ISO/IEC 9899:2011. GCC has limited incomplete support for parts of
>>>>>> this standard, enabled with -std=c11 or -std=iso9899:2011."
>>>>>>
>>>>>> It does not say anything about "experimental" there. The word
>>>>>> "experimental" is only used for the C++ support, but we hardly have C++
>>>>>> code in QEMU -- if you worry about that, I could simply drop the
>>>>>> "-std=gnu++11" part from my patch?
>>>>>
>>>>> I was looking at the "info gcc" docs on RHEL7, 
>>>>> gcc-4.8.5-16.el7_4.1.x86_64:
>>>>>
>>>>>   "3.4 Options Controlling C Dialect
>>>>>
>>>>>        ....snip...
>>>>>
>>>>>      'gnu11'
>>>>>      'gnu1x'
>>>>>           GNU dialect of ISO C11.  Support is incomplete and
>>>>>           experimental.  The name 'gnu1x' is deprecated."
>>>>
>>>> Ok. Looks like the "Support is incomplete and experimental" sentence has
>>>> been removed with GCC 4.9.0 here. So GCC 4.8 is likely pretty close
>>>> already. IMHO we could give it a try and enable gnu11 for QEMU with GCC
>>>> v4.8, too. If we later find problems, we could still switch back to
>>>> gnu99 instead. Other opinions?
>>>
>>> Our code is already cleanly compiling with gnu99 standard - the problem
>>> is merely that we sometimes introduce regressions due to not enforcing
>>> that standard level. I don't think the features in gnu11 are compelling
>>> enough to justify using something that's declared experimental.
>> What about the duplicated typedef problem? See:
>>
>>  https://lists.gnu.org/archive/html/qemu-devel/2018-11/msg05829.html
>>
>> That one occured with Clang, but I think we've had plenty of these in
>> the past with GCC, too...
> 
> IIUC, That's only a problem because we don't pass any  -std  flag, and
> so get the compilers default, which may be gnu11.  If we explicitly
> set -std=gnu99, that problem will be reported by patchew, travis and
> maintainers own build tests, and thus won't get anywhere near git master.

Ah, right, I just tried it with a newer version of Clang, and indeed it
then prints out a warning if some code with a duplicated typedef is
compiled with "-std=gnu99" (while it remains silent without that
option). So in that case I'm fine with gnu99, too. I'll send a v2 of the
patch...

 Thomas