Re: [cxx-gtk-utils-general] Build failure on armv5tl

[Barry J. <zen...@ze...>]

On 12/05/16 15:32, Chris Vine wrote:
> On Thu, 12 May 2016 13:14:22 +0100
> Chris Vine <ch...@cv...> wrote:
>> On Thu, 12 May 2016 12:09:26 +0100
>> Barry Jackson <zen...@ze...> wrote:
>>> Ah - seems I attached the wrong file - this is correct.
>>>
>>> Barry
>>>
>>
>> It looks like a compiler bug.  The error is first reported in line
>> 522 of tests/test_task_manager.cpp, and says that std::future<int>
>> is an incomplete type, when clearly it is complete.  It looks as if
>> the compiler cannot find the 'future' class in the std namespace.  It
>> should be able to do so, because the <future> header is included.
>>
>> You could try including a 'using namespace std;' statement after the
>> 'using namespace Cgu;' in the file test_task_manager.cpp.  If that
>> fixes it then it is definitely a compiler bug.
>>
>> The build seems to be using gcc-5.3.1, which should be new enough but
>> is not actually a gnu release, I think (gnu has only released
>> gcc-5.3.0 in the 5.3 series).
>
> On looking at it further, it seems that std::future was not supported
> by gcc-4.8 on armv5 because of the lack of an appropriate atomic
> instruction: the debian error report at
> https://gcc.gnu.org/ml/gcc/2014-07/msg00001.html is very similar to
> yours.
>
> So it looks as if that may remain the case with gcc-5.3 (bummer).  If so
> it looks like a big miss and you are going to find an increasing number
> of C++ packages failing to build for this reason, if that is the
> reason: std::future is hardly ground-breaking.  amd64 seems to be OK.
>
> You may have better luck with clang and their libc++, I don't know.
> You could also look and see if it is fixed in gcc-6.1 (probably not).
>
> Alternatively, if you want a distributable package, you could just omit
> the tests by not having your build system call 'make check' or 'make
> tests'.  Happily the code in question in c++-gtk-utils is templated and
> so not actually instantiated unless it is used (as it is by the test
> suite).  Of course if users actually try calling a function returning a
> std::future object they would then get a compile-time error similar to
> yours, but that is a separate issue, and one which they are probably
> aware of if using armv5.
>
> Chris
>

Many thanks for your rapid response :)
Yes we want a distributable package, so skipping the tests would be 
fine, however we currently don't call 'make check' in the spec, so I'm 
guessing the tests can be disabled in the configuration somewhere?
Our current spec is here:
http://svnweb.mageia.org/packages/cauldron/c%2B%2B-gtk-utils/current/SPECS/c%2B%2B-gtk-utils.spec?view=markup

Barry