The tiny package in this repository is an exploration of an idea I had for how to use unsafePerformIO in a way that is "obviously" safe:
the only IO actions embedded in pure thunks are innocuous, merely adding a message to a queue.
This queue is then observed elsewhere, in the monad IO.
That's it; I just wanted a little library that codified that idea without imposing too many opinions about how it should be used.
EG I've still forced the user to choose between seq and pseq to embed their instrumentation into their pure code;
that aspect is going to be tricky no matter how you go about it, so I wanted the signatures to make it clear that I'm in no way trying to solve that question for you.
Also, this is my first deployment of weak pointers. I think it adds some interesting expressiveness (eg you can be notified when a certain subset of thunks are all either forced or GC'd), but we'll need to come up with some good stress tests before I'm confident about it.