The latter. atomicModifyIORef is harder though still, since it is a primop with the same properties as modifyIORef :/
So would it make sense to create a strict modifyIORef' function?
Very much so. In fact, I'd argue the vast majority of uses are for the WHNF-strict version.
I just fixed a leak with atomicModifyIORef that was exactly this problem. If it had at least been documented I wouldn't have had to do that. So I'm going to submit a library proposal to either 1) strictify atomicModifyIORef, 2) add atomicModifyIORef', or at the least 3) add documentation that says "this function leaks". Same story for modifyIORef of course. The only workaround I could find is to immediately read the value back out and 'seq' on it, but it's ugly. So two questions: writeIORef doesn't have this problem. If I am just writing a simple value, is writeIORef atomic? In other words, can I replace 'atomicModifyIORef r (const (x, ())' with 'writeIORef r x'? Any reason to not do solution 1 above?