My approach is that there are a few things that the default answer for should be "no" and then you have to justify (maybe just to yourself) why they're appropriate in this case. Macro Fu in C, template meta-programming in C++ and rewriting from scratch are all examples of these.
In this case I am avoiding the 'throw it all away and start from scratch' approach. It would be infeasible for the intervening period. I am putting together an approach that would get us there in a year or so, but we can lop off smaller chunks to rewrite (the existing architecture is a series of daemons, which helps us there).
In this case I am avoiding the 'throw it all away and start from scratch' approach. It would be infeasible for the intervening period. I am putting together an approach that would get us there in a year or so, but we can lop off smaller chunks to rewrite (the existing architecture is a series of daemons, which helps us there).