Let 'c2h' convert CStrings to Haskell Strings, and 'h2c' convert Haskell Strings to CStrings. (If I understand correctly, c2h . h2c === id, but h2c . c2h is not the identity on all inputs;
That is correct. CStrings are 8-bits, and Haskell Strings are 32-bits. Converting from Haskell to C loses information, unless you use a multi-byte encoding on the C side (for instance, UTF8).
So actually I am incorrect, and h2c . c2h is the identity but c2h . h2c is not?
I suggest you look at the utf8-string package, for instance Codec.Binary.UTF8.String.{encode,decode}, which convert Haskell strings to/from a list of Word8, which can then be transferred via the FFI to wherever you like.
This package was very helpful; I looked at the source to see how the utf8 encoding was done. It looks as if the functionality I want is technically feasible but not implemented yet; it shouldn't be too much trouble to implement it myself, by imitating the existing 'decode' function but changing its behavior when it runs out of input in the middle of a utf8-character. Also key is the property s1 ++ s2 == decode (encode s1)) ++ decode (encode s2)) holds. Thanks, Eric