Hello Bulat and Anatoly, I have written a first version of an interface to inplace updates in the ST monad for the hmatrix vectors and matrices. Many things must be improved (range checking, documentation, etc.) but I hope that the general idea makes sense. A few usage examples: http://perception.inf.um.es/~aruiz/darcs/hmatrix/examples/inplace.hs Code: http://perception.inf.um.es/~aruiz/darcs/hmatrix/lib/Data/Packed/ST.hs http://perception.inf.um.es/~aruiz/darcs/hmatrix/doc/html/Data-Packed-ST.htm... Any suggestion will be welcome. I'm impressed by the power of the ST monad, it is extremely useful and elegant. Thank you again for your help! In the future I will also try to include efficient conversions to/from standard Haskell arrays and those of other related libraries like Jed Brown's CArray. Thanks, Alberto Bulat Ziganshin wrote:
Hello Alberto,
Tuesday, June 3, 2008, 12:56:50 PM, you wrote:
Good! So you can easily "hide" the IO operations in the ST monad. I will definitely look into it.
from implementation POV ST monad is nothing but renamed IO monad which exports only subset of its operations which are guaranteed to safe. or, saying in other words, it's just type hackery around IO monad that provides safe operations
it's possible to define ST monad and its operations as following:
newtype ST s a = forall s. ST_Constructor (IO a)
unsafeIOtoSt action = ST_Constructor action
runST (ST_Constructor action) = unsafePerformIO action
newtype STRef s a = forall s. STRef (IORef a)
readSTRef (STRef ref) = unsafeIOtoSt (readIORef ref)
and so on. GHC uses technically (but not ideologically!) different implementation where both monads are specializations of one generic type. while Hugs afair uses exactly this approach. you may also look at ArrayRef lib which reimplements arrays/refs for both compilers in more unified way
anyway, because ST is just IO monad modulo type tricks, you can execute any IO action inside ST by lifting it with unsafeIOtoSt