It is nice if a tool understands those and for an `emit` suggests only my signals for auto-completion, but if it lists them along all other members it is fine.

So I can see the argument from a puristic view, but in practice, once my built is setup to run it (cmake automoc or something), it simply works.

My issue with Qt is that the API design comes from days very long ago, where they did fantastic work on keeping compatibility, while not adopting to "modern C++" which always requires me to rethink the way I write code. Also integrating with non-Qt code isn't always easy due to reliance on QString (which has value Unicode features), QList (which meanwhile has many std-like APIs) etc. which easily travel through all layers. Thus when staying inside At it's really nice and quite consistent, but integrating is hard.

If it is not semantics essential to the correct operation of your program then this is a different matter. I really am not sure what the contention here or misunderstanding is.

When I write C++ I write it for a C pre-processor and C++ compiler to read in order to obtain a output program which does something specific and in line with the C++ standard.

This is not some high minded purism - this is just pragmatic reality.

If I write something in my source which I expect to result in specific output but the specific output cannot be obtained if I use a standard compliant C Preprosessor and C++ Compiler then either my expectation is wrong or I'm writing something which is not C++.

So if you don't rely on the behaviour that results from MOC being part of the toolchain then sure - you are writing C++ - if you do - it is not C++.

If the semantics (in a sense where it actually affects output) is part of a language, which it is, then changing the semantics of the language makes it a different language.

> So I can see the argument from a puristic view, but in practice, once my built is setup to run it (cmake automoc or something), it simply works.

This does not make it C++.

and indeed :

    echo "#include <QObject>\nclass Foo : public QObject { Q_OBJECT signals: void foo(); public slots: void bar() { emit foo(); } }; int main() { Foo f; f.bar(); }" > /tmp/qt.cpp
    g++ -fPIC -I/usr/include/qt -I/usr/include/qt/QtCore -c /tmp/qt.cpp

So put differently I expected Q_OBJECT (and what it expands to) to have semantic meaning which is not provided by a standard compliant C Preprosessor and C++ Compiler.

And if it was valid output - i.e. the output is the entirety of what I expected from the input - then I would not need to ever use MOC.

Moc just adds some more generated code next to you valid C++ code, generally in a new translation unit. That generated code is just boiler plate which you could write by hand if you wanted. In a sense this is what verdigris does: it provides macros that makes makes it easier to "write it by hand" (the W_OBJECT_IMPL macro somehow expends to the code generated by moc)

No, I understand what it does.

> Moc just adds some more generated code next to you valid C++ code

Code without which the complete output of your toolchain from your source is incomplete and therefore invalid - again if this was not the case you would not need to run MOC at all.

That is not really how moc works though ? moc does not change anything to your original sources nor their translation units; it is not an additional preprocessing step, but a parallel one.

This .o is exactly the same that one that would be produced in a complete Qt program. Moc just generates additional source files with the implementation of some functions whose declaration is given by the Q_... macros. You could also write them by hand or generate them with fancy cmake scripts... it's just not a fun experience.