moc-generated code error at Qt6 when not at Qt5

JonB

I have code which errors with moc at Qt6 but went through fine at Qt5. I am not looking for a code change to make it work (I have that, see at end) but for the reason for the change.

Here are 2 pairs of files for an absolutely minimal repro:

bar.h:

#ifndef BAR_H
#define BAR_H

class Bar
{
public:
    Bar();
};

#endif // BAR_H

bar.cpp

#include "bar.h"

Bar::Bar() {}

foo.h:

#ifndef FOO_H
#define FOO_H

#include <QList>
#include <QObject>

// I want it to work with the "incomplete" type of next line, rather than #include "bar.h" here ...
class Bar;

class Foo : public QObject
{
    Q_OBJECT
public:
    Foo();
public slots:
    void this_works(Bar *bar);
    void this_errors(QList<Bar *>bars);
};

#endif // FOO_H

foo.cpp:

#include "foo.h"

#include "bar.h"

Foo::Foo() {}

void Foo::this_works(Bar *bar)
{
    Q_UNUSED(bar);
}

void Foo::this_errors(QList<Bar *> bars)
{
    Q_UNUSED(bars);
}

I am "parsimonious" in my use of including .h files in other .h files where I don't want/feel I need to, partly for compilation speed, partly for code separation.

Running build, moc complains with this long message:

g++ -c -pipe -g -std=gnu++1z -Wall -Wextra -fPIC -D_REENTRANT -DQT_QML_DEBUG -DQT_CORE_LIB -I../../../mocerror -I. -I/usr/include/x86_64-linux-gnu/qt6 -I/usr/include/x86_64-linux-gnu/qt6/QtCore -I. -I/usr/lib/x86_64-linux-gnu/qt6/mkspecs/linux-g++ -o moc_foo.o moc_foo.cpp

In file included from /usr/include/x86_64-linux-gnu/qt6/QtCore/qobject.h:18,
                 from /usr/include/x86_64-linux-gnu/qt6/QtCore/QObject:1,
                 from ../../foo.h:5,
                 from moc_foo.cpp:10:
/usr/include/x86_64-linux-gnu/qt6/QtCore/qmetatype.h: In instantiation of ‘struct QtPrivate::IsPointerToTypeDerivedFromQObject<Bar*>’:
/usr/include/x86_64-linux-gnu/qt6/QtCore/qmetatype.h:1102:60:   required from ‘struct QMetaTypeId<Bar*>’
/usr/include/x86_64-linux-gnu/qt6/QtCore/qmetatype.h:1122:38:   required from ‘struct QMetaTypeId2<Bar*>’
/usr/include/x86_64-linux-gnu/qt6/QtCore/qmetatype.h:1585:1:   required from ‘struct QMetaTypeId<QList<Bar*> >’
/usr/include/x86_64-linux-gnu/qt6/QtCore/qmetatype.h:2364:42:   recursively required by substitution of ‘template<class T> struct QtPrivate::BuiltinMetaType<T, typename std::enable_if<QMetaTypeId2<T>::IsBuiltIn, void>::type> [with T = QList<Bar*>]’
/usr/include/x86_64-linux-gnu/qt6/QtCore/qmetatype.h:2364:42:   required from ‘constexpr const QtPrivate::QMetaTypeInterface QtPrivate::QMetaTypeInterfaceWrapper<QList<Bar*> >::metaType’
/usr/include/x86_64-linux-gnu/qt6/QtCore/qmetatype.h:2486:16:   required from ‘constexpr const QtPrivate::QMetaTypeInterface* QtPrivate::qTryMetaTypeInterfaceForType() [with Unique = {anonymous}::qt_meta_stringdata_Foo_t; TypeCompletePair = TypeAndForceComplete<QList<Bar*>, std::integral_constant<bool, false> >]’
/usr/include/x86_64-linux-gnu/qt6/QtCore/qmetatype.h:2537:55:   required from ‘constexpr const QtPrivate::QMetaTypeInterface* const qt_incomplete_metaTypeArray [5]<{anonymous}::qt_meta_stringdata_Foo_t, QtPrivate::TypeAndForceComplete<Foo, std::integral_constant<bool, true> >, QtPrivate::TypeAndForceComplete<void, std::integral_constant<bool, false> >, QtPrivate::TypeAndForceComplete<Bar*, std::integral_constant<bool, false> >, QtPrivate::TypeAndForceComplete<void, std::integral_constant<bool, false> >, QtPrivate::TypeAndForceComplete<QList<Bar*>, std::integral_constant<bool, false> > >’
moc_foo.cpp:95:5:   required from here
/usr/include/x86_64-linux-gnu/qt6/QtCore/qmetatype.h:842:23: error: invalid application of ‘sizeof’ to incomplete type ‘Bar’
  842 |         static_assert(sizeof(T), "Type argument of Q_PROPERTY or Q_DECLARE_METATYPE(T*) must be fully defined");
      |                       ^~~~~~~~~

Note that it complains on void this_errors(QList<Bar *>bars); but not on void Foo::this_works(Bar *bar) (try commenting out each method declaration, or moving from public slots: to public:). Why? Why does a function parameter of QList<IncompleteType *> need any more information about IncompleteType than parameter IncompleteType * without QList does? And why any different at Qt6 than at Qt5?

At the time the compiler meets the Bar in slot parameters in foo.h it has only encountered class Bar; declaration. In foo.cpp I actually #include "bar.h". I would like to keep it like that. But to keep Qt6 moc-generated code happy I have to move #include "bar.h" into foo.h, so that it has the "full" definition of class Bar when it sees the slot methods. But why, and why only for QList<Bar *> and why only at Qt6?

Axel Spoerl

I am using gcc 13.3.0, moc 6.9.0.
The only difference in our CMake files is that I require 3.16, you require 3.14.

Axel Spoerl

Hi Jon,

could it be that the build directory is "polluted" by a previous build?
I have copied your code into a new project and it builds like this (Qt 6.7 and 6.8):

19:04:18: Running steps for project JonB...
19:04:18: Starting: "/opt/Qt/Tools/CMake/bin/cmake" --build /home/axel/QtDev/JonB/build/Qt_dev-Debug --target all
[1/6 0.6/sec] Automatic MOC and UIC for target JonB
[2/6 0.2/sec] Building CXX object CMakeFiles/JonB.dir/bar.cpp.o
[3/6 0.3/sec] Building CXX object CMakeFiles/JonB.dir/foo.cpp.o
[4/6 0.4/sec] Building CXX object CMakeFiles/JonB.dir/main.cpp.o
[5/6 0.5/sec] Building CXX object CMakeFiles/JonB.dir/JonB_autogen/mocs_compilation.cpp.o
[6/6 0.6/sec] Linking CXX executable JonB
19:04:29: The process "/opt/Qt/Tools/CMake/bin/cmake" exited normally.
19:04:29: Elapsed time: 00:11.

The only thing I actually added on top, is a main.cpp, creating one instance of each, Foo and Bar on the stack.

JonB

@Axel-Spoerl
Hi Axel, thank you for testing.

All is clean, and reproducible from scratch by me.

This is a compilation issue, so main and creating instances doesn't matter. I don't know why you don't see the same as I do. As I move across I am still using qmake; you use cmake, I cannot see why that would cause a difference but will try tomorrow. I will also get exact Qt versions.

Axel Spoerl

Hi Jon,
Let us see your .pro file.
We’ll fix this together!

JonB

@Axel-Spoerl
I am looking at moving this test over to cmake to see whether it makes any difference/works like you showed. Will take me a while as I discover what that involves!

Meanwhile the .pro file you asked for, which I am using when the compiler generates the error:

QT = core

CONFIG += c++17 cmdline

# You can make your code fail to compile if it uses deprecated APIs.
# In order to do so, uncomment the following line.
#DEFINES += QT_DISABLE_DEPRECATED_BEFORE=0x060000    # disables all the APIs deprecated before Qt 6.0.0

SOURCES += \
        bar.cpp \
        foo.cpp \
        main.cpp

# Default rules for deployment.
qnx: target.path = /tmp/$${TARGET}/bin
else: unix:!android: target.path = /opt/$${TARGET}/bin
!isEmpty(target.path): INSTALLS += target

HEADERS += \
    bar.h \
    foo.h

JonB

@Axel-Spoerl
Before I try to move to cmake, in case that makes any difference. With qmake if you refer to the error in my first post you see it happens on moc_foo.cpp. The relevant moc/compilation lines executed are:

g++ -pipe -g -std=gnu++1z -Wall -Wextra -fPIC -dM -E -o moc_predefs.h /usr/lib/x86_64-linux-gnu/qt6/mkspecs/features/data/dummy.cpp
/usr/lib/qt6/libexec/moc -DQT_QML_DEBUG -DQT_CORE_LIB --include /home/jon/QtTests/mocerror/build/Desktop-Debug/moc_predefs.h -I/usr/lib/x86_64-linux-gnu/qt6/mkspecs/linux-g++ -I/home/jon/QtTests/mocerror -I/usr/include/x86_64-linux-gnu/qt6 -I/usr/include/x86_64-linux-gnu/qt6/QtCore -I. -I/usr/include/c++/13 -I/usr/include/x86_64-linux-gnu/c++/13 -I/usr/include/c++/13/backward -I/usr/lib/gcc/x86_64-linux-gnu/13/include -I/usr/local/include -I/usr/include/x86_64-linux-gnu -I/usr/include ../../foo.h -o moc_foo.cpp
g++ -c -pipe -g -std=gnu++1z -Wall -Wextra -fPIC -D_REENTRANT -DQT_QML_DEBUG -DQT_CORE_LIB -I../../../mocerror -I. -I/usr/include/x86_64-linux-gnu/qt6 -I/usr/include/x86_64-linux-gnu/qt6/QtCore -I. -I/usr/lib/x86_64-linux-gnu/qt6/mkspecs/linux-g++ -o moc_foo.o moc_foo.cpp

My generated moc_foo.cpp file from my pasted code is 151 lines long. The error message includes

required from ‘struct QMetaTypeId<QList<Bar*> >’
/usr/include/x86_64-linux-gnu/qt6/QtCore/qmetatype.h:2364:42: recursively required by substitution of ‘template<class T> struct QtPrivate::BuiltinMetaType<T, typename std::enable_if<QMetaTypeId2<T>::IsBuiltIn, void>::type> [with T = QList<Bar*>]’
QtPrivate::TypeAndForceComplete<QList<Bar*>, std::integral_constant<bool, false>
moc_foo.cpp:95:5: required from here

There I see function:

Q_CONSTINIT const QMetaObject Foo::staticMetaObject = { {
    QMetaObject::SuperData::link<QObject::staticMetaObject>(),
    qt_meta_stringdata_Foo.offsetsAndSizes,
    qt_meta_data_Foo,
    qt_static_metacall,
    nullptr,
    qt_incomplete_metaTypeArray<qt_meta_stringdata_Foo_t,
        // Q_OBJECT / Q_GADGET
        QtPrivate::TypeAndForceComplete<Foo, std::true_type>,
        // method 'this_works'
        QtPrivate::TypeAndForceComplete<void, std::false_type>,
        QtPrivate::TypeAndForceComplete<Bar *, std::false_type>,
        // method 'this_errors'
        QtPrivate::TypeAndForceComplete<void, std::false_type>,
        QtPrivate::TypeAndForceComplete<QList<Bar*>, std::false_type>
    >,
    nullptr
} };

The line generating the error is the

QtPrivate::TypeAndForceComplete<QList<Bar*>

Something in evaluating that is what is complaining that it does not have the "complete" definition of Bar class/type (because of how I wish to arrange my #include "bar.h"). But see how there is a also a line for QtPrivate::TypeAndForceComplete<Bar *, std::false_type>. That does not error. Hence the question: apart from why it has changed at Qt6, why does a slot parameter of QList<Bar *> need to know the full definition of Bar when a slot parameter of Bar * does not?

Ultimately the error comes from the moc-generated code causing the compiler to try to evaluate sizeof(Bar)/whether Bar has a Q_OBJECT (I think)/whatever it is doing with Q_PROPERTY or Q_DECLARE_METATYPE(T*). Which of course fails on an incomplete definition of Bar. I know about copying of signal/slot parameters (e.g. it certainly would complain if I tried to pass Bar or QList<Bar>) but I am passing pointers and I do not see why QList<Bar *> (fails) needs to know any more about Bar than Bar * (succeeds) does.

Do you not have similar in your generated moc_foo.cpp for your build using cmake?

JonB

@Axel-Spoerl
I have now changed over to cmake (project mocerror2). FWIW my CMakeLists.txt:

cmake_minimum_required(VERSION 3.14)

project(mocerror2 LANGUAGES CXX)

set(CMAKE_AUTOUIC ON)
set(CMAKE_AUTOMOC ON)
set(CMAKE_AUTORCC ON)

set(CMAKE_CXX_STANDARD 17)
set(CMAKE_CXX_STANDARD_REQUIRED ON)

find_package(QT NAMES Qt6 Qt5 REQUIRED COMPONENTS Core)
find_package(Qt${QT_VERSION_MAJOR} REQUIRED COMPONENTS Core)

add_executable(mocerror2
  main.cpp
  bar.cpp bar.h foo.cpp foo.h
)
target_link_libraries(mocerror2 Qt${QT_VERSION_MAJOR}::Core)

include(GNUInstallDirs)
install(TARGETS mocerror2
    LIBRARY DESTINATION ${CMAKE_INSTALL_LIBDIR}
    RUNTIME DESTINATION ${CMAKE_INSTALL_BINDIR}
)

As I expected it makes no difference and I get same error:

[ 14%] Automatic MOC and UIC for target mocerror2
[ 14%] Built target mocerror2_autogen
[ 28%] Building CXX object CMakeFiles/mocerror2.dir/mocerror2_autogen/mocs_compilation.cpp.o
In file included from /usr/include/x86_64-linux-gnu/qt6/QtCore/qobject.h:18,
                 from /usr/include/x86_64-linux-gnu/qt6/QtCore/QObject:1,
                 from /home/jon/QtTests/mocerror2/build/Desktop-Debug/mocerror2_autogen/EWIEGA46WW/../../../../foo.h:5,
                 from /home/jon/QtTests/mocerror2/build/Desktop-Debug/mocerror2_autogen/EWIEGA46WW/moc_foo.cpp:10,
                 from /home/jon/QtTests/mocerror2/build/Desktop-Debug/mocerror2_autogen/mocs_compilation.cpp:2:
/usr/include/x86_64-linux-gnu/qt6/QtCore/qmetatype.h: In instantiation of ‘struct QtPrivate::IsPointerToTypeDerivedFromQObject<Bar*>’:
/usr/include/x86_64-linux-gnu/qt6/QtCore/qmetatype.h:1102:60:   required from ‘struct QMetaTypeId<Bar*>’
/usr/include/x86_64-linux-gnu/qt6/QtCore/qmetatype.h:1122:38:   required from ‘struct QMetaTypeId2<Bar*>’
/usr/include/x86_64-linux-gnu/qt6/QtCore/qmetatype.h:1585:1:   required from ‘struct QMetaTypeId<QList<Bar*> >’
/usr/include/x86_64-linux-gnu/qt6/QtCore/qmetatype.h:2364:42:   recursively required by substitution of ‘template<class T> struct QtPrivate::BuiltinMetaType<T, typename std::enable_if<QMetaTypeId2<T>::IsBuiltIn, void>::type> [with T = QList<Bar*>]’
/usr/include/x86_64-linux-gnu/qt6/QtCore/qmetatype.h:2364:42:   required from ‘constexpr const QtPrivate::QMetaTypeInterface QtPrivate::QMetaTypeInterfaceWrapper<QList<Bar*> >::metaType’
/usr/include/x86_64-linux-gnu/qt6/QtCore/qmetatype.h:2486:16:   required from ‘constexpr const QtPrivate::QMetaTypeInterface* QtPrivate::qTryMetaTypeInterfaceForType() [with Unique = {anonymous}::qt_meta_stringdata_Foo_t; TypeCompletePair = TypeAndForceComplete<QList<Bar*>, std::integral_constant<bool, false> >]’
/usr/include/x86_64-linux-gnu/qt6/QtCore/qmetatype.h:2537:55:   required from ‘constexpr const QtPrivate::QMetaTypeInterface* const qt_incomplete_metaTypeArray [5]<{anonymous}::qt_meta_stringdata_Foo_t, QtPrivate::TypeAndForceComplete<Foo, std::integral_constant<bool, true> >, QtPrivate::TypeAndForceComplete<void, std::integral_constant<bool, false> >, QtPrivate::TypeAndForceComplete<Bar*, std::integral_constant<bool, false> >, QtPrivate::TypeAndForceComplete<void, std::integral_constant<bool, false> >, QtPrivate::TypeAndForceComplete<QList<Bar*>, std::integral_constant<bool, false> > >’
/home/jon/QtTests/mocerror2/build/Desktop-Debug/mocerror2_autogen/EWIEGA46WW/moc_foo.cpp:95:5:   required from here
/usr/include/x86_64-linux-gnu/qt6/QtCore/qmetatype.h:842:23: error: invalid application of ‘sizeof’ to incomplete type ‘Bar’
  842 |         static_assert(sizeof(T), "Type argument of Q_PROPERTY or Q_DECLARE_METATYPE(T*) must be fully defined");
      |                       ^~~~~~~~~
gmake[2]: *** [CMakeFiles/mocerror2.dir/build.make:82: CMakeFiles/mocerror2.dir/mocerror2_autogen/mocs_compilation.cpp.o] Error 1
gmake[1]: *** [CMakeFiles/Makefile2:86: CMakeFiles/mocerror2.dir/all] Error 2
gmake: *** [Makefile:136: all] Error 2
10:38:41: The process "/usr/bin/cmake" exited with code 2.

I do not understand why you do not get same error as I do.

My gcc is 13.2.0. My moc is 6.4.2 as is my Qt6.

Axel Spoerl

I am using gcc 13.3.0, moc 6.9.0.
The only difference in our CMake files is that I require 3.16, you require 3.14.

JonB

@Axel-Spoerl
Thank you. So what would you like me to do about our difference in behaviour? :)
[I fetch all my Qt stuff from apt as supplied with Ubuntu 24.04 distro.]

J.Hilk

I actually remember something similar, early when QVector was nuked for QList to prosper.

I think if you change it from QList<Bar *>bars to QList<Bar *> *bars it actually works. Axel is using a way later Qt6 version ( 6.8 vs 6.4.2) QList was probably touched again in between those versions, to fix this :D

JonB

@J-Hilk said in moc-generated code error at Qt6 when not at Qt5:

I think if you change it from QList<Bar *>bars to QList<Bar *> *bars it actually works.

But those are quite different types! :) Are you just showing me that to say what it could cope with?

Axel is using a way later Qt6 version ( 6.8 vs 6.4.2) QList was probably touched again in between those versions, to fix this :D

Ah! Well in that case it's obviously not to be reported. Thanks for the heads-up.

With Qt5 I always had Qt 5.15 (or maybe 5.12) via apt under various Ubuntu releases over the years. And I have to say that was just great, I didn't encounter problems and I didn't need later versions. I suspected Qt6 would be problematic, looks like I get a quite older version and am going to have issues which really need a later version from TQtC :( Ho-hum....

J.Hilk

@JonB what was the reason again, why you don't fetch the latest version from the installer or the repo (to compile yourself) ?

JonB

@J-Hilk

I have never even attempted to download from TQtC, having seen the endless problems with downloads/mirrors. Fills me with dread....
...As does when I see people who for whatever reason (e.g. maybe to do with configuration, whatever it happens) get an error during compilation of the whole of Qt....
I don't have the disk space in VM for Qt sources + compilation outputs.
I don't have the memory in VM to compile it. Not to mention the requirements for QWebEngine.
I share (development) with other people using Qt from Ubuntu distro. They are not going to want to fetch/compile/install or take from me.

Am I being difficult? ;-) It was all so simple with Qt5, everything worked. Now beginning to think of backing out of the whole of Qt6 and reverting to Qt5. But that would be a shame for my answering questions on this forum.

J.Hilk

@JonB I Understand, I myself are still on 5.15.17 and I stay with 5 as long as possible.

But Android support is apparently already dropped for Qt5. So I have at most a year until I have to switch over :(

JonB

@J-Hilk said in moc-generated code error at Qt6 when not at Qt5:

@JonB I Understand, I myself are still on 5.15.17 and I stay with 5 as long as possible.

OMG! What platform/OS version are you on?

J.Hilk

@JonB windows, macOS, iOS, iPadOS & Android. Still so far all the latest available (not betas)

JonB

@J-Hilk Right, but not my Ubuntu 24.04 which I am moving to, and doing Qt as part of it.

J.Hilk

@JonB non of our customers use Linux, only reason why we support macOS is because its my development environment and its only a few steps more to deploy for it as well ;)