The normal Gen monad always splits the seed when doing >>=. This is for very good reasons - it lets you write generators that generate infinite data to the left of a >>= and let's your generators be very lazy!

A traditional GenT m a implementation would inherit this splitting behaviour in order to let you keep writing infinite and lazy things to the left of >>= on the GenT m level. Now, the thing to realize about this is that unless your code is very carefully written to avoid it this means you're going to end up with unnecessary >>=s and thus unnecessary splits.

To get around this issue of unnecessary splits we introduce a threading GenT implementation here that sacrifices letting you do infinite (and to some extent lazy) structures to the left of >>= on the GenT m level, but doesn't prohibit you from doing so on the Gen level.

This drastically reduces the number of seed splits while still letting you write lazy and infinite generators in Gen land by being a little bit more careful. It works great for constrained-generators in particular, which has a tendency to be strict and by design avoids inifinte values.