- There's a very interesting discussion on the Ocaml mailing list about Software Transactional Memory for concurrent programming. In particular, see the referenced papers here and here. The basic idea is to wrap up all accesses to shared memory in a STM monad, which is eventually sequenced atomically. The neat thing about doing this is that the concurrent mechanisms compose; see the papers for how it all works out. Now, there's no reason another language couldn't do this, but Haskell is ideal because the pure semantics and the type system conspire to prevent any non-pure accesses from within the atomic section that are not to "registered" STM slots (i.e. the atomic sequencer won't thread parts of the IO monad through the computation---only STM monads are allowed). Pretty slick stuff.
- The nbody integrator paper I'm working on is coming along nicely (I'm a physicist, though you wouldn't know it to look at the blog); just some quick (i.e. a day or two long) simulations to run and then a lot of editing before it's ready for submission. Some time soon I may post a quick synopsis of the idea; for now you can read about what it's not (though similar) in Jack Wisdom's Symplectic Maps for the N-Body Problem (with Holman). If that tickles your fancy, have a look at Jack's Symplectic Correctors (also with Holman). If that doesn't make your jaw drop then you must not be much of a dynamicist!
08 March 2006
My god! It's been about a month since I last posted (about PLT Scheme's new JIT). Just to prime the pump (hopefully more to follow more quickly):