I was curious if there has been any research on "concurrency analysis" which would determine which thunks might be accessed concurrently by different threads. I imagine that the vast majority are only accessed in a single threaded manner so can use a more efficient representation when doing SMP parallelism with haskell. It seems that some sort of abstract interpretation could provide a conservative answer to this similar to the way update analysis is done. John -- John Meacham - ⑆repetae.net⑆john⑈