I agree in the sense that Wayland adoption would have definitely gone quicker if that was the case, however in the long run this approach does make sense (otherwise you will eventually just run into the same sorts of issues X11 had).

Btw what you're describing is not that far off from the normal way of using Wayland protocols in development - you use wayland-scanner to generate C source files from the protocols, and you include those to actually "use" the protocols in your programs. Admittedly all my Wayland development experience has been "client-side", so I really don't know how complex it is to build a compositor, but dwl (minimalist Wayland compositor) is only around 3k lines of code (only slightly more than dwm (minimalist X wm)).