unread
this is only a problem for option 1.
Fair point.
the short/medium term impact is so big
Given the scope of it, I think realistically any change suggested here should aim to target 9.0 for the deprecation (and 10 for the change) to give userland a full major release cycle (recent history shows this is ~5 years).
As I said earlier, after a couple of decades of exposure to client projects, I'm not convinced that the top 1000 or even 10000 composer packages is necessarily representative of the majority of userland code in the world.
Cheers
Stephen