IMO the problem with threads is the 'all-or-nothing' approach to shared data. If you want to make use of multiple CPUs, you either use threads, in which case al

IMO the problem with threads is the 'all-or-nothing' approach to shared data. If you want to make use of multiple CPUs, you either use threads, in which case all of your process data is shared, or you use multiple processes, where none of the data is shared. It is the over-sharing that trips people up with threads. Ideally, we need an API that allows selective data sharing, so you greatly reduce the chance of one thread trampling over bits of another. Right now, the common way to do that is with multiple processes and shared memory regions, but it is cumbersome to set up and requires the programmer to treat the shared data very differently (e.g. one process can't just malloc data for a string, say, and then let the other process access it. The other route is through message-passing, but again you have to write your code in a totally different way. Someone please invent multi-threading with selective sharing!