[Development] QNX broken on 5.7

[Hausmann Simon]

Hi,

When building Qt 5.7 for QNX a special patch is applied in the CI that fixes the Dinkumware headers. You were CCed when this was discussed (Subject was "‎QNX and Dinkumware support for constexpr and nullptr").

Simon

  Original Message
From: Rafael Roquetto
Sent: Tuesday, March 8, 2016 18:39
To: development at qt-project.org
Subject: [Development] QNX broken on 5.7


Hello everyone,

I would like to point out that QNX is unable to build the 5.7 branch. The
commit that probably triggered the breakage is:

    d6bb01e1779f1840dfbab57c6ecd615587bbde62
    Force inclusion of <atomic> on QNX systems.


This raises my first question: should this not have been reported by the CI?
This is not the first time that changes go in and break QNX, in spite of
having the CI in place. Would anyone have a clue about what is going on?


Now, concerning the issue itself, it seems to be a bigger problem:

    /usr/include/cpp/xxatomic: In instantiation of 'std::atomic<_Ty*>::atomic(_Ty*) [with _Ty = void(QtMsgType, const char*)]':
    /usr/include/cpp/xxatomic:336:3: error: invalid conversion from 'void (*)(QtMsgType, const char*)' to 'void*' [-fpermissive]
    /usr/include/cpp/xxatomic:873:15: error:   initializing argument 1 of 'void* std::_Atomic_address::operator=(void*)' [-fpermissive]


Caused by the following line inside qlogging.cpp:

static QBasicAtomicPointer<void (QtMsgType, const char*)> msgHandler = Q_BASIC_ATOMIC_INITIALIZER(qDefaultMsgHandler);

The problem can be traced to the Dinkumware's std::atomic implementation.
Ultimately, qDefaultMsgHandler will be passed as an argument to std::atomic's
default ctor, which looks like this:

        _CONST_FUN atomic(_Ty *_Right) _NOEXCEPT
        {   // construct from _Right
            _Atomic_address::operator=(_Right);
        }


_Atomic_address is a base class. As you can see, this ctor is implemented
using operator=, which looks like this:


    inline _ITYPE _ATOMIC_ITYPE::operator=(_ITYPE _Value) _NOEXCEPT
    {   // assign _Value to *this
        atomic_store(this, _Value);
        return _Value;
    }
                `

where _TYPE expands to (void *), thus effectively causing the conversion from
void (*)(QtMsgType, const char*) to (void *), which is the error.

Interestingly, Dinkumware's std::atomic does provide operator= in terms of
_Ty:


    _Ty *operator=(_Ty *_Right) _NOEXCEPT
    {   // assign from _Right
        return static_cast<_Ty *>(_Atomic_address::operator=((void *)_Right));
    }


which makes me think that the ctor is wrongly implemented, since it bypasses
this and calls the base class operator= explicitly.


I am not sure how to work around this. QBasicAtomicPointer won't let me
initialize it with some ugly (void *) pointer.

Does it make sense to revert this patch? Is there a way to fall back to the
old non-C++11 implementation?

Thanks,
Rafael

--
Rafael Roquetto | rafael.roquetto at kdab.com | Software Engineer
Klarälvdalens Datakonsult AB, a KDAB Group company
Tel. Sweden (HQ) +46-563-540090, USA +1-866-777-KDAB(5322)
KDAB - Qt Experts