Of all the things I don't like about Qt, the fact that it doesn't play well with templates bugs me the most. You can't do this:

template < typename T >
struct templated_widget : QWidget
{
  Q_OBJECT;

public signals:
  void something_happened(T);
};

It also doesn't play well with the preprocessor. You can't do this:

#define CREATE_WIDGET(name,type) \
struct name ## _widget : QWidget \
{ \
  Q_OBJECT; \
\
public signals: \
  void something_happened(type); \
}

That, mixed with the fact that everything that responds to a signal has to be a Q_OBJECT, makes Qt hard to work in for a C++ programmer. People used to Java or Python style programming probably fair better actually.

I actually spent a lot of time and effort researching and devising a way to gain type safety back and connect a Qt signal to any functor object: http://crazyeddiecpp.blogspot.com/2011/01/quest-for-sane-signals-in-qt-step-1.html

The kind of thing I want to do there is basic, everyday C++ development made next to impossible by the Qt moc...which itself is entirely unnecessary now days, if it ever actually was.

Frankly though, I'm stuck with it because if you want to do automated UI testing, Qt is pretty much the only game in town short of MFC...which is so 1980 (it sucks working in that shit really hard). Some might say WX but it's got even more serious problems. GTKmm would have been my first choice but since it's all owner drawn and doesn't do accessibility...can't be driven by industry standard testing software. Qt is hard enough in that regard (barely works when you modify the accessibility plugin).