On one of my coding projects, I only did realize recently the runtime component model I did choose is not "powerful"/"flexible"/"robust"/etc enough.

I brutally started with a component model dealing only with raw component pointers: "nope, I need reference counting". I am good for some severe re-factoring (something alien to corpo software).

Guys when you start to code project which you know will be fairly sized, you go directly for a "more than bare pointers" component model ?

If yes, how far do you go? gobject level? (from on assembly coding perspective, ofc)