So maybe 1) we shouldn't wrote systems that are tens of millions of anarchically intercoupled lines of code or 2) we should write 100 100,000 line subsystems and link them together in a rigorous way.
They key is then to simply avoid emergent interactions between modules/subsystems. That's pretty doable.
What doesn't happen is that people never reason about the cost of defects. Even if they do some reasoning about this, they don't apply that to other components of systems.
Exactly what I said above. Verifying individual components and their straight-forward compisition is how it was done in the past. People just ignore that.
They key is then to simply avoid emergent interactions between modules/subsystems. That's pretty doable.
What doesn't happen is that people never reason about the cost of defects. Even if they do some reasoning about this, they don't apply that to other components of systems.