However, most of the code I've changed is used in the old buildsystem as well so somebody could possibly reuse what I'm doing in a Qt5 port. I'm not planning to do that though, Qt 6.2 will be perfectly good for my personal and selfish needs. :)
Well, according to Stroustrup himself C++98/03 is considered "modern C++" (see e.g. https://www.stroustrup.com/DnE2005.pdf ); and I can confirm from personal experience that C++98 and STL were a big step compared to what we had before. C++11 and later have some advantages, but nothing I couldn't live without (personally I don't consider policy based design the way to go). Actually Qt itself nicely demonstrated how to do perfectly well without e.g. move semantics.
C++20 is a wholly more pleasant programming environment than any previous C++. When there is a choice between the old way and the new—which there always is, for backward compatibility—the new way is almost always better.
For some cases, the newer way is better sometimes but not always, e.g. east function return type, with "auto", vs. old-style west. For short functions, deduced return type, i.e. neither one, is often best.
All that said, C++20 coroutines will be much more pleasant to use when thd C++23 library support for them lands.
Because I'm not obsessed with chasing the latest thing?
Well, why would someone want to implement an outdated 90s UI? Apparently there are good reasons.
I have many projects even with Qt4.4 and no need to go to a later Qt version. Qt4 already had everything I ever expected. Qt5 brings a couple of advantages (e.g. better modularization and backend API), but Qt5.4 is good enough for that.
While hackish, it does look pretty promising: https://twitter.com/linusgroh/status/1426904775623385095/