Some general points about recoders: 1. We don't really need to work out a single recoder interface that's best for every purpose. All the types people have proposed can happily coexist, and in many cases it's easy to write functions which convert from one to another. 2. There are regional encodings in use which are not stateless (JIS is the only one I know of personally), so I think that a currentLocaleEncoding function will have to return a stateful interface. 3. I expect that most conversion functions will be implemented in C code, for three reasons: (a) performance; (b) the C code already exists, is well-tested, knows about correct locale handling on different systems, etc., and (c) every Haskell executable will become hundreds of kilobytes larger if it has to include translation tables for every locale in which it might be run. We have little choice but to use the C library's tables, and the only interface to them is the C conversion functions. 4. Existing C conversion libraries (whose interfaces we can't change) store their state in opaque data structures and provide no way to copy that state. So all use has to be single-threaded, and if we want to enforce that statically we can't use an explicitly threaded interface. ST and IO are the only options. -- Ben