Pony’s does this with the added benefit that all the checks happen at compile time. I’m only somewhat familiar with STM and think some of the safety checks are at runtime but could be wrong.
One of the issues with STM is you can't use IO inside the atomic blocks, which is easy enough to get around but if your users are unaware it'll present bugs. Haskell enforces a separation of IO at the type-level, hence this risk is gone so you get every benefit of STM with none of the risk. STM itself eliminates a set of concurrency problems based on the way it works, but you do still have to do some run-time checks to eliminate deadlocks (i.e. use `modify` functions or have the atomic blocks behave like them, or you could end up with a deadlock).
The safety checks are at compile-time, since it only allows pure functions to be run. You can circumvent this of course with unsafePerformIO, but then you are completely on your own.
