Changing something closer to the root of a large dependency tree triggers a lot of recompilation. The lack of real modules and reliance of `#include` in C++ makes dependency graphs larger than they could be.

Not necessarily. Say I'm editing "bob.cc", and in the mean time edited "common.h" #included by many many files. While I'm editing "bob.cc", there is no need for me to rebuild the entire project until I want to see how it links to everything else (which can be postponed quite a bit if you have some tests); so I either run "make bob.cc" or "make test_bob.cc && ./test_bob". I don't know much C++ and all this is theoretical, but I think this is the ideal workflow with C/C++ development, and this is why incremental builds are useful, especially if the code is nicely separated into testable units.

Not full recompilations, but when we're working on a feature release, switching git branches routinely requires a lot of files to be rebuilt due to header changes.

ccache is your friend.

When you use templates, recompiling a single file can take ten minutes.

Is there an open source project where I can test this? I don't know much about C++ (I know some C), and would like to see this for myself.