"Qt is Java for C++", one of my colleagues used to say, and I saw what he meant once I started working with it. We used to develop with Qt4 a lot and I rarely ran into any issues at all, getting started was really straight forward and the documentation and examples are (were?) pretty comprehensive.

I do still cite Qt's API documentation as the best I've ever used.

I've only done Qt toy apps on the Pi with C++, which was a pretty good experience, could imagine a large Qt codebase in C++ could be stressful.

For Python I like Qt quite a bit because the loose signal/slot approach meshes nicely with a dynamic language.

I work on a sizeable desktop application written in PyQt. While I agree that Qt has a very nineties feel about it, you don't have to buy into that. Specifically, you don't have to use class hierarchies to model your domain. It actually helps to have a language boundary -- we can let Qt be its object-oriented self, but on the Python side it's different. We do a limited amount of subclassing, which lets us hook into user interaction events, but other than that we use whatever works best in Python. As time passes, this seems to involve more and more straight-forward procedural code, less and less object-oriented modeling.

My experience as well - a lot of my nuts and bolts OO work in static languages was making things conform to interfaces and existing class hierarchies. In dynamic languages there's less need for that and hierarchies are shallower and less entangled. None of that is surprising, of course.