On Fri, Apr 27, 2012 at 6:24 AM, Simon Marlow
That concern was why the discussion last time had turned toward reverting
to the previous IORef version, which is safe under a wider array of conditions.
So the options are:
[...] 2. Use atomicModifyIORef;
no way to generate uniques in STM
The difference is if you unsafeIOtoSTM . atomicModifyIORef the runtime system doesn't panic and shut everything down, but if we unsafePerformIO . atomically we do. This makes me lean very heavily towards option 2, as the only contract that we have on Uniques are that they are unique, so if the transaction played back in an edited context, you'd just get a different Unique. This would let you get Uniques in IO and unsafePerformIO, and unsafeIOtoSTM should let it work in STM. -Edward